Merge branch 'master' into ocricola-cleanup

.gitignore

@@ -52,6 +52,8 @@ training/wordlist2dawg
 *.patch
 
 # ignore compilation files
+build/*
+/bin
 */.deps/*
 */.libs/*
 *.lo
@@ -63,8 +65,6 @@ training/wordlist2dawg
 *.jar
 
 # tessdata
-*.cube.*
-*.tesseract_cube.*
 *.traineddata
 
 # OpenCL
@@ -73,5 +73,10 @@ kernel*.bin
 
 # build dirs
 /build*
+/.cppan
 /cppan
+/*.dll
+/*.lib
+/*.exe
+/*.lnk
 /win*

.travis.yml

@@ -7,11 +7,11 @@ sudo: required
 
 os:
   - linux
-  #- osx
+  - osx
 
-branches:
+#branches:
-  only:
+  #only:
-    - master
+    #- master
 
 addons:
   apt:
@@ -25,16 +25,16 @@ before_install:
   - if [[ $TRAVIS_OS_NAME == linux   ]]; then LINUX=true; fi
   - if [[ $TRAVIS_OS_NAME == osx     ]]; then OSX=true; fi
 
-  - if [[ $OSX ]]; then brew update; fi
+  #- if [[ $OSX ]]; then brew update; fi
 
-  - export LEPT_VER=1.73
+  - export LEPT_VER=1.74.1
 
 install:
-  - if [[ $OSX ]]; then brew install icu4c pango; brew link --force gettext; fi
+  #- if [[ $OSX ]]; then brew install icu4c pango; brew link --force gettext; fi
-  - if [[ $OSX ]]; then export ICU_ROOT=/usr/local/opt/icu4c ; fi
+  #- if [[ $OSX ]]; then export ICU_ROOT=/usr/local/opt/icu4c ; fi
-  - wget https://www.cmake.org/files/v3.6/cmake-3.6.1-Linux-x86_64.sh
+  - if [[ $LINUX ]]; then wget https://www.cmake.org/files/v3.7/cmake-3.7.2-Linux-x86_64.sh; fi
-  - sudo sh cmake-3.6.1-Linux-x86_64.sh --skip-license --prefix=/usr
+  - if [[ $LINUX ]]; then sudo sh cmake-3.7.2-Linux-x86_64.sh --skip-license --prefix=/usr; fi
-  - wget -O leptonica.zip https://github.com/DanBloomberg/leptonica/archive/v$LEPT_VER.zip
+  - wget -O leptonica.zip https://github.com/DanBloomberg/leptonica/archive/$LEPT_VER.zip
   - unzip leptonica.zip -d .
   - cmake -Hleptonica-$LEPT_VER -Bleptonica-$LEPT_VER/build
   - make -C leptonica-$LEPT_VER/build

AUTHORS

@@ -2,12 +2,14 @@ Ray Smith (lead developer) <theraysmith@gmail.com>
 Ahmad Abdulkader
 Rika Antonova
 Nicholas Beato
+Jeff Breidenbach
 Samuel Charron
 Phil Cheatle
 Simon Crouch
 David Eger
 Sheelagh Huddleston
 Dan Johnson
+Rajesh Katikam
 Thomas Kielbus
 Dar-Shyang Lee
 Zongyi (Joe) Liu
@@ -26,3 +28,15 @@ Joern Wanke
 Ping Ping Xiu
 Andrew Ziem
 Oscar Zuniga
+
+Community Contributors:
+Zdenko Podobný (Maintainer)
+Jim Regan (Maintainer)
+James R Barlow
+Amit Dovev
+Martin Ettl
+Tom Morris
+Tobias Müller
+Egor Pugin
+Sundar M. Vaidya
+Stefan Weil

CMakeLists.txt

@@ -10,6 +10,12 @@
 
 cmake_minimum_required(VERSION 2.8.11)
 
+if (NOT APPVEYOR)
+if (WIN32 AND (CMAKE_VERSION VERSION_EQUAL 3.6 OR (CMAKE_VERSION VERSION_GREATER 3.6 AND CMAKE_VERSION VERSION_LESS 3.7)))
+    message(FATAL_ERROR "You have bugged CMake version 3.6 which is known to not work with tesseract. Please, upgrade CMake.")
+endif()
+endif()
+
 # In-source builds are disabled.
 if (${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR})
     message(FATAL_ERROR
@@ -40,25 +46,31 @@ set_property(GLOBAL PROPERTY PREDEFINED_TARGETS_FOLDER "CMake Targets")
 
 project(tesseract C CXX)
 
-set(VERSION_MAJOR 3)
+set(VERSION_MAJOR 4)
-set(VERSION_MINOR 05)
+set(VERSION_MINOR 00)
 set(VERSION_PLAIN ${VERSION_MAJOR}.${VERSION_MINOR})
 
-set(MINIMUM_LEPTONICA_VERSION 1.71)
+set(MINIMUM_LEPTONICA_VERSION 1.74)
 
-if(NOT EXISTS ${PROJECT_SOURCE_DIR}/cppan)
+if(NOT EXISTS ${PROJECT_SOURCE_DIR}/.cppan)
     if (NOT Leptonica_DIR AND NOT MSVC)
         find_package(PkgConfig REQUIRED)
-        pkg_check_modules(Leptonica REQUIRED lept)
+        pkg_check_modules(Leptonica REQUIRED lept>=${MINIMUM_LEPTONICA_VERSION})
     else()
         find_package(Leptonica ${MINIMUM_LEPTONICA_VERSION} REQUIRED CONFIG)
     endif()
 else()
-    add_subdirectory(cppan)
+    if (STATIC)
+        set(CPPAN_BUILD_SHARED_LIBS 0)
+    else()
+        set(CPPAN_BUILD_SHARED_LIBS 1)
+    endif()
+    add_subdirectory(.cppan)
 endif()
 
 find_package(OpenCL QUIET)
-find_package(PkgConfig)
+
+option(BUILD_TRAINING_TOOLS "Build training tools" ON)
 
 ###############################################################################
 #
@@ -76,6 +88,9 @@ if (WIN32)
         add_definitions(-D_CRT_SECURE_NO_WARNINGS)
 
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
+        if (APPVEYOR)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /W0")
+        endif()
     endif()
 
     set(LIB_Ws2_32 Ws2_32)
@@ -104,12 +119,7 @@ include(Configure)
 
 configure_file(${AUTOCONFIG_SRC} ${AUTOCONFIG} @ONLY)
 
-set(INCLUDE_DIR
+set(INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/include" "${CMAKE_INSTALL_PREFIX}/include/tesseract")
-    ${CMAKE_SOURCE_DIR}/api
-    ${CMAKE_SOURCE_DIR}/ccmain
-    ${CMAKE_SOURCE_DIR}/ccstruct
-    ${CMAKE_SOURCE_DIR}/ccutil
-)
 
 configure_file(
     ${CMAKE_SOURCE_DIR}/cmake/templates/TesseractConfig-version.cmake.in
@@ -137,14 +147,14 @@ include_directories(${Leptonica_INCLUDE_DIRS})
 include_directories(${CMAKE_BINARY_DIR})
 
 include_directories(api)
+include_directories(arch)
 include_directories(ccmain)
 include_directories(ccstruct)
 include_directories(ccutil)
 include_directories(classify)
-include_directories(cube)
 include_directories(cutil)
 include_directories(dict)
-include_directories(neural_networks/runtime)
+include_directories(lstm)
 include_directories(opencl)
 include_directories(textord)
 include_directories(vs2010/port)
@@ -159,14 +169,14 @@ string(SUBSTRING ${VERSION_MINOR} 0 1 VERSION_MINOR_0)
 string(SUBSTRING ${VERSION_MINOR} 1 1 VERSION_MINOR_1)
 
 file(GLOB tesseract_src
+    arch/*.cpp
     ccmain/*.cpp
     ccstruct/*.cpp
     ccutil/*.cpp
     classify/*.cpp
-    cube/*.cpp
     cutil/*.cpp
     dict/*.cpp
-    neural_networks/runtime/*.cpp
+    lstm/*.cpp
     opencl/*.cpp
     textord/*.cpp
     viewer/*.cpp
@@ -174,14 +184,14 @@ file(GLOB tesseract_src
 )
 file(GLOB tesseract_hdr
     api/*.h
+    arch/*.h
     ccmain/*.h
     ccstruct/*.h
     ccutil/*.h
     classify/*.h
-    cube/*.h
     cutil/*.h
     dict/*.h
-    neural_networks/runtime/*.h
+    lstm/*.h
     opencl/*.h
     textord/*.h
     viewer/*.h
@@ -201,25 +211,40 @@ set(tesseract_src ${tesseract_src}
     api/pdfrenderer.cpp
 )
 
-add_library                     (tesseract ${LIBRARY_TYPE} ${tesseract_src} ${tesseract_hdr})
-if (NOT STATIC)
-target_compile_definitions      (tesseract PUBLIC -DTESS_EXPORTS)
-endif()
-target_link_libraries           (tesseract ${LIB_Ws2_32} ${LIB_pthread})
-set_target_properties           (tesseract PROPERTIES VERSION ${VERSION_MAJOR}.${VERSION_MINOR_0}.${VERSION_MINOR_1})
-set_target_properties           (tesseract PROPERTIES SOVERSION ${VERSION_MAJOR}.${VERSION_MINOR_0}.${VERSION_MINOR_1})
 if (WIN32)
-set_target_properties           (tesseract PROPERTIES OUTPUT_NAME tesseract${VERSION_MAJOR}${VERSION_MINOR})
+    set_source_files_properties(
-set_target_properties           (tesseract PROPERTIES DEBUG_OUTPUT_NAME tesseract${VERSION_MAJOR}${VERSION_MINOR}d)
+        ${CMAKE_CURRENT_SOURCE_DIR}/arch/dotproductsse.cpp
+        PROPERTIES COMPILE_DEFINITIONS __SSE4_1__)
+    if (MSVC)
+        set_source_files_properties(
+            ${CMAKE_CURRENT_SOURCE_DIR}/arch/dotproductavx.cpp
+            PROPERTIES COMPILE_FLAGS "/arch:AVX")
+    endif()
+endif()
+
+add_library                     (libtesseract ${LIBRARY_TYPE} ${tesseract_src} ${tesseract_hdr})
+if (NOT STATIC)
+target_compile_definitions      (libtesseract
+    PRIVATE -DTESS_EXPORTS
+    INTERFACE -DTESS_IMPORTS
+)
+set_target_properties           (libtesseract PROPERTIES WINDOWS_EXPORT_ALL_SYMBOLS True)
+endif()
+target_link_libraries           (libtesseract ${LIB_Ws2_32} ${LIB_pthread})
+set_target_properties           (libtesseract PROPERTIES VERSION ${VERSION_MAJOR}.${VERSION_MINOR_0}.${VERSION_MINOR_1})
+set_target_properties           (libtesseract PROPERTIES SOVERSION ${VERSION_MAJOR}.${VERSION_MINOR_0}.${VERSION_MINOR_1})
+if (WIN32)
+set_target_properties           (libtesseract PROPERTIES OUTPUT_NAME tesseract${VERSION_MAJOR}${VERSION_MINOR})
+set_target_properties           (libtesseract PROPERTIES DEBUG_OUTPUT_NAME tesseract${VERSION_MAJOR}${VERSION_MINOR}d)
 endif()
 
 if (NOT CPPAN_BUILD)
-    target_link_libraries       (tesseract ${Leptonica_LIBRARIES})
+    target_link_libraries       (libtesseract ${Leptonica_LIBRARIES})
-    export(TARGETS tesseract FILE ${CMAKE_BINARY_DIR}/TesseractTargets.cmake)
+    export(TARGETS libtesseract FILE ${CMAKE_BINARY_DIR}/TesseractTargets.cmake)
 else()
-    target_link_libraries       (tesseract cppan)
+    target_link_libraries       (libtesseract pvt.cppan.demo.danbloomberg.leptonica)
     file(WRITE ${CMAKE_BINARY_DIR}/TesseractTargets.cmake "include(${CMAKE_BINARY_DIR}/cppan.cmake)\n")
-    export(TARGETS tesseract APPEND FILE ${CMAKE_BINARY_DIR}/TesseractTargets.cmake)
+    export(TARGETS libtesseract APPEND FILE ${CMAKE_BINARY_DIR}/TesseractTargets.cmake)
 endif()
 
 ########################################
@@ -231,12 +256,97 @@ set(tesseractmain_src
     vs2010/tesseract/resource.h
     vs2010/tesseract/tesseract.rc
 )
-add_executable                  (tesseractmain ${tesseractmain_src})
+add_executable                  (tesseract ${tesseractmain_src})
-target_link_libraries           (tesseractmain tesseract)
+target_link_libraries           (tesseract libtesseract)
-set_target_properties           (tesseractmain PROPERTIES OUTPUT_NAME tesseract)
 
 ########################################
 
+if (BUILD_TRAINING_TOOLS)
 add_subdirectory(training)
+endif()
+
+get_target_property(tesseract_NAME libtesseract NAME)
+get_target_property(tesseract_VERSION libtesseract VERSION)
+get_target_property(tesseract_OUTPUT_NAME libtesseract OUTPUT_NAME)
+configure_file(tesseract.pc.cmake ${CMAKE_CURRENT_BINARY_DIR}/tesseract.pc @ONLY)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/tesseract.pc DESTINATION lib/pkgconfig)
+install(TARGETS tesseract RUNTIME DESTINATION bin LIBRARY DESTINATION lib ARCHIVE DESTINATION lib)
+install(TARGETS libtesseract EXPORT TesseractTargets RUNTIME DESTINATION bin LIBRARY DESTINATION lib ARCHIVE DESTINATION lib)
+install(EXPORT TesseractTargets DESTINATION cmake)
+install(FILES
+    ${CMAKE_BINARY_DIR}/TesseractConfig.cmake
+    ${CMAKE_BINARY_DIR}/TesseractConfig-version.cmake
+    DESTINATION cmake)
+
+install(FILES
+    # from api/makefile.am
+    api/apitypes.h
+    api/baseapi.h 
+    api/capi.h 
+    api/renderer.h
+
+    #from arch/makefile.am
+    arch/dotproductavx.h
+    arch/dotproductsse.h 
+    arch/simddetect.h
+
+    #from ccmain/makefile.am
+    ccmain/thresholder.h
+    ccmain/ltrresultiterator.h 
+    ccmain/pageiterator.h 
+    ccmain/resultiterator.h 
+    ccmain/osdetect.h
+
+    #from ccstruct/makefile.am
+    ccstruct/publictypes.h 
+
+    #from ccutil/makefile.am
+    ccutil/basedir.h 
+    ccutil/errcode.h 
+    ccutil/fileerr.h 
+    ccutil/genericvector.h 
+    ccutil/helpers.h 
+    ccutil/host.h 
+    ccutil/memry.h 
+    ccutil/ndminx.h
+    ccutil/params.h
+    ccutil/ocrclass.h 
+    ccutil/platform.h 
+    ccutil/serialis.h 
+    ccutil/strngs.h 
+    ccutil/tesscallback.h
+    ccutil/unichar.h 
+    ccutil/unicharcompress.h 
+    ccutil/unicharmap.h 
+    ccutil/unicharset.h
+
+    #from lstm/makefile.am
+    lstm/convolve.h
+    lstm/ctc.h 
+    lstm/fullyconnected.h 
+    lstm/functions.h 
+    lstm/input.h
+    lstm/lstm.h 
+    lstm/lstmrecognizer.h 
+    lstm/lstmtrainer.h 
+    lstm/maxpool.h
+    lstm/networkbuilder.h 
+    lstm/network.h 
+    lstm/networkio.h 
+    lstm/networkscratch.h
+    lstm/parallel.h 
+    lstm/plumbing.h 
+    lstm/recodebeam.h 
+    lstm/reconfig.h 
+    lstm/reversed.h
+    lstm/series.h 
+    lstm/static_shape.h 
+    lstm/stridemap.h 
+    lstm/tfnetwork.h 
+    lstm/weightmatrix.h
+
+    #${CMAKE_BINARY_DIR}/src/endianness.h
+    DESTINATION include/tesseract)
+
 
 ###############################################################################

CONTRIBUTING.md

@@ -24,7 +24,7 @@ When creating an issue, please report your operating system, including its speci
 
 Search through open and closed issues to see if similar issue has been reported already (and sometimes also has been solved). 
 
-Similary, before you post your question in the forum, search through past threads to see if similar question has been asked already.
+Similarly, before you post your question in the forum, search through past threads to see if similar question has been asked already.
 
 Read the [wiki](https://github.com/tesseract-ocr/tesseract/wiki) before you report your issue or ask a question in the forum.
 
@@ -44,7 +44,7 @@ When attaching a file to the issue report / forum ...
 
 Do not attach programs or libraries to your issues/posts.
 
-For large files or for programs, add a link to a iocation where they can be downloaded (your site, Git repo, Google Drive, Dropbox etc.)
+For large files or for programs, add a link to a location where they can be downloaded (your site, Git repo, Google Drive, Dropbox etc.)
 
 Attaching a multi-page TIFF image is useful only if you have problem with multi-page functionality, otherwise attach only one or a few single page images. 
 

COPYING

@@ -17,5 +17,5 @@ in this distribution is now licensed under the Apache License:
 Other Dependencies and Licenses:
 ================================
 
-Tesseract uses Leptonica library (http://leptonica.com/) with a very weakly
+Tesseract uses Leptonica library (http://leptonica.com/) which essentially
-restricted copyright license (http://leptonica.com/about-the-license.html)
+uses a BSD 2-clause license. (http://leptonica.com/about-the-license.html)

ChangeLog

@@ -1,3 +1,46 @@
+2017-03-24 - V4.00.00-alpha
+  * Added new neural network system based on LSTMs, with major accuracy gains.
+  * Improvements to PDF rendering.
+  * Fixes to trainingdata rendering.  
+  * Added LSTM models+lang models to 101 languages. (tessdata repository)
+  * Improved multi-page TIFF handling.
+  * Fixed damage to binary images when processing PDFs.
+  * Fixes to training process to allow incremental training from a recognition model.
+  * Made LSTM the default engine, pushed cube out.
+  * Deleted cube code.
+  * Changed OEModes --oem 0 for legacy tesseract engine, --oem 1 for LSTM, --oem 2 for both, --oem 3 for default.
+  * Avoid use of Leptonica debug parameters or functions.
+  * Fixed multi-language mode.
+  * Removed support for VS2010.
+  * Added Support for VS2015 and VS2017 with CPPAN.
+  * Implemented invisible text only for PDF.
+  * Added AVX / SSE support for Windows
+  * Enabled OpenMP support.
+  * Miscellaneous Fixes.
+
+2017-02-16 - V3.05.00
+  * Made some fine tuning to the hOCR output.
+  * Added TSV as another optional output format.
+  * Fixed ABI break introduced in 3.04.00 with the AnalyseLayout() method.
+  * text2image tool - Enable all OpenType ligatures available in a font. This feature requires Pango 1.38 or newer.
+  * Training tools - Replaced asserts with tprintf() and exit(1).
+  * Fixed Cygwin compatibility.
+  * Improved multipage tiff processing.
+  * Improved the embedded pdf font (pdf.ttf).
+  * Enable selection of OCR engine mode from command line.
+  * Changed tesseract command line parameter '-psm' to '--psm'.
+  * Write output of tesseract --help, --version and --list-langs to stdout instead of stderr.
+  * Added new C API for orientation and script detection, removed the old one.
+  * Increased minimum autoconf version to 2.59.
+  * Removed dead code.
+  * Require Leptonica 1.74 or higher.
+  * Fixed many compiler warning.
+  * Fixed memory and resource leaks.
+  * Fixed some issues with the 'Cube' OCR engine.
+  * Fixed some openCL issues.
+  * Added option to build Tesseract with CMake build system.
+  * Implemented CPPAN support for easy Windows building.
+  
 2016-02-17 - V3.04.01
   * Added OSD renderer for psm 0. Works for single page and multi-page images.
   * Improve tesstrain.sh script.

INSTALL

@@ -45,7 +45,7 @@ The simplest way to compile this package is:
      `sh ./configure' instead to prevent `csh' from trying to execute
      `configure' itself.
 
-     Running `configure' takes awhile.  While running, it prints some
+     Running `configure' takes a while.  While running, it prints some
      messages telling which features it is checking for.
 
   2. Type `make' to compile the package.

INSTALL.GIT.md

@@ -3,11 +3,12 @@
 If you have cloned Tesseract from GitHub, you must generate
 the configure script.
 
-If you have tesseract 3.0x installation in your system, please remove it
+If you have tesseract 4.0x installation in your system, please remove it
 before new build.
 
 Known dependencies for training tools (excluding leptonica):
- * compiler with c++ support
+ * compiler with c++11 support
+ * autoconf-archive
  * pango-devel
  * cairo-devel
  * icu-devel
@@ -24,7 +25,7 @@ So, the steps for making Tesseract are:
 You need to install at least English language and OSD data files to TESSDATA_PREFIX
 directory. You can retrieve single file with tools like [wget](https://www.gnu.org/software/wget/), [curl](https://curl.haxx.se/), [GithubDownloader](https://github.com/intezer/GithubDownloader) or browser.
 
-All language data files can be retrieved from git repository (usefull only for packagers!):
+All language data files can be retrieved from git repository (useful only for packagers!):
 
     $ git clone https://github.com/tesseract-ocr/tessdata.git tesseract-ocr.tessdata
 

LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

Makefile.am

@@ -4,9 +4,12 @@ ACLOCAL_AMFLAGS = -I m4
 if ENABLE_TRAINING
 TRAINING_SUBDIR = training
 training:
+	$(MAKE)
 	@cd "$(top_builddir)/training" && $(MAKE)
 training-install:
 	@cd "$(top_builddir)/training" && $(MAKE) install
+training-uninstall:
+	@cd "$(top_builddir)/training" && $(MAKE) uninstall
 clean-local:
 	@cd "$(top_builddir)/training" && $(MAKE) clean
 else
@@ -16,10 +19,7 @@ endif
 
 .PHONY: install-langs ScrollView.jar install-jars training
 
-SUBDIRS = ccutil viewer cutil opencl ccstruct dict classify wordrec textord
+SUBDIRS = arch ccutil viewer cutil opencl ccstruct dict classify wordrec textord lstm
-if !NO_CUBE_BUILD
-    SUBDIRS += neural_networks/runtime cube
-endif
 SUBDIRS += ccmain api . tessdata doc
 
 EXTRA_DIST = README.md\
@@ -35,14 +35,14 @@ dist-hook:
 # Need to remove .svn directories from directories
 # added using EXTRA_DIST. $(distdir)/tessdata would in
 # theory suffice.
-	rm -rf `find $(distdir) -name .svn`
+	rm -rf $(find $(distdir) -name .svn)
-	rm -rf `find $(distdir) -name .git`
+	rm -rf $(find $(distdir) -name .git)
-	rm -rf `find $(distdir) -name .deps`
+	rm -rf $(find $(distdir) -name .deps)
-	rm -rf `find $(distdir) -name .libs`
+	rm -rf $(find $(distdir) -name .libs)
-	rm -rf `find $(distdir) -name *.o`
+	rm -rf $(find $(distdir) -name *.o)
-	rm -rf `find $(distdir) -name *.lo`
+	rm -rf $(find $(distdir) -name *.lo)
-	rm -rf `find $(distdir) -name *.la`
+	rm -rf $(find $(distdir) -name *.la)
-	rm -rf `find $(distdir)/training -executable -type f`
+	rm -rf $(find $(distdir)/training -executable -type f)
 	rm -rf $(distdir)/doc/html/*
 
 ScrollView.jar:

README.md

@@ -1,29 +1,30 @@
 [![Build Status](https://travis-ci.org/tesseract-ocr/tesseract.svg?branch=master)](https://travis-ci.org/tesseract-ocr/tesseract)
-[![Build status](https://ci.appveyor.com/api/projects/status/miah0ikfsf0j3819?svg=true)](https://ci.appveyor.com/project/zdenop/tesseract/)
+[![Build status](https://ci.appveyor.com/api/projects/status/miah0ikfsf0j3819/branch/master?svg=true)](https://ci.appveyor.com/project/zdenop/tesseract/)
 
 For the latest online version of the README.md see:
     
   https://github.com/tesseract-ocr/tesseract/blob/master/README.md
 
-#About
+# About
 
 This package contains an OCR engine - `libtesseract` and a command line program - `tesseract`.
 
 The lead developer is Ray Smith. The maintainer is Zdenko Podobny. 
-For a list of contributors see [AUTHORS](https://github.com/tesseract-ocr/tesseract/blob/master/AUTHORS) and github's log of [contributors](https://github.com/tesseract-ocr/tesseract/graphs/contributors).
+For a list of contributors see [AUTHORS](https://github.com/tesseract-ocr/tesseract/blob/master/AUTHORS)
+and GitHub's log of [contributors](https://github.com/tesseract-ocr/tesseract/graphs/contributors).
 
 Tesseract has unicode (UTF-8) support, and can recognize more than 100
 languages "out of the box". It can be trained to recognize other languages. See [Tesseract Training](https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract) for more information. 
 
 Tesseract supports various output formats: plain-text, hocr(html), pdf.
 
-This project does not include a GUI application. If you need one, please see the [3rdParty](https://github.com/tesseract-ocr/tesseract/wiki/3rdParty) wiki page.
+This project does not include a GUI application. If you need one, please see the [3rdParty](https://github.com/tesseract-ocr/tesseract/wiki/User-Projects-%E2%80%93-3rdParty) wiki page.
 
 You should note that in many cases, in order to get better OCR results, you'll need to [improve the quality](https://github.com/tesseract-ocr/tesseract/wiki/ImproveQuality) of the image you are giving Tesseract.
 
-The latest stable version is 3.04.01, released in February 2016.
+The latest stable version is 3.05.00, released in February 2017.
 
-#Brief history
+# Brief history
 
 Tesseract was originally developed at Hewlett-Packard Laboratories Bristol and
 at Hewlett-Packard Co, Greeley Colorado between 1985 and 1994, with some
@@ -33,13 +34,13 @@ In 2005 Tesseract was open sourced by HP. Since 2006 it is developed by Google.
 
 [Release Notes](https://github.com/tesseract-ocr/tesseract/wiki/ReleaseNotes)
 
-#For developers
+# For developers
 
 Developers can use `libtesseract` [C](https://github.com/tesseract-ocr/tesseract/blob/master/api/capi.h) or [C++](https://github.com/tesseract-ocr/tesseract/blob/master/api/baseapi.h) API to build their own application. If you need bindings to `libtesseract` for other programming languages, please see the [wrapper](https://github.com/tesseract-ocr/tesseract/wiki/AddOns#tesseract-wrappers) section on AddOns wiki page.
 
 Documentation of Tesseract generated from source code by doxygen can be found on [tesseract-ocr.github.io](http://tesseract-ocr.github.io/).
 
-#License
+# License
 
     The code in this repository is licensed under the Apache License, Version 2.0 (the "License");
     you may not use this file except in compliance with the License.
@@ -55,19 +56,27 @@ Documentation of Tesseract generated from source code by doxygen can be found on
 
 **NOTE**: This software depends on other packages that may be licensed under different open source licenses.
 
-#Installing Tesseract
+# Installing Tesseract
 
 You can either [Install Tesseract via pre-built binary package](https://github.com/tesseract-ocr/tesseract/wiki) or [build it from source](https://github.com/tesseract-ocr/tesseract/wiki/Compiling).
 
-#Running Tesseract
+## Supported Compilers
+
+* GCC 4.8 and above
+* Clang 3.4 and above
+* MSVC 2015, 2017
+
+Other compilers might work, but are not officially supported.
+
+# Running Tesseract
 
 Basic command line usage:
 
-    tesseract imagename outputbase [-l lang] [-psm pagesegmode] [configfiles...]
+    tesseract imagename outputbase [-l lang] [--psm pagesegmode] [configfiles...]
 
 For more information about the various command line options use `tesseract --help` or `man tesseract`. 
 
-#Support
+# Support
 
 Mailing-lists:
 * [tesseract-ocr](https://groups.google.com/d/forum/tesseract-ocr) - For tesseract users. 

android/jni/Android.mk

@@ -4,7 +4,7 @@ include $(CLEAR_VARS)
 LOCAL_MODULE := tesseract-$(APP_ABI)
 
 LOCAL_STATIC_LIBRARIES := \
-    mobile_base \
+    base \
     leptonica-$(APP_ABI)
 
 LOCAL_C_INCLUDES := $(APP_C_INCLUDES)
@@ -30,13 +30,6 @@ $(info local path=$(LOCAL_PATH))
 LOCAL_SRC_FILES := $(wildcard $(LOCAL_PATH)/../../api/*.cpp $(LOCAL_PATH)/../../ccmain/*.cpp $(LOCAL_PATH)/../../ccstruct/*.cpp $(LOCAL_PATH)/../../ccutil/*.cpp $(LOCAL_PATH)/../../classify/*.cpp $(LOCAL_PATH)/../../cutil/*.cpp $(LOCAL_PATH)/../../dict/*.cpp $(LOCAL_PATH)/../../image/*.cpp $(LOCAL_PATH)/../../textord/*.cpp $(LOCAL_PATH)/../../viewer/*.cpp $(LOCAL_PATH)/../../wordrec/*.cpp)
 
 EXPLICIT_SRC_EXCLUDES := \
-  $(LOCAL_PATH)/../../ccmain/cubeclassifier.cpp \
-  $(LOCAL_PATH)/../../ccmain/cubeclassifier.h  \
-  $(LOCAL_PATH)/../../ccmain/cube_control.cpp \
-  $(LOCAL_PATH)/../../ccmain/cube_reco_context.cpp \
-  $(LOCAL_PATH)/../../ccmain/cube_reco_context.h \
-  $(LOCAL_PATH)/../../ccmain/tesseract_cube_combiner.cpp \
-  $(LOCAL_PATH)/../../ccmain/tesseract_cube_combiner.h \
   $(LOCAL_PATH)/../../api/pdfrenderer.cpp \
   $(LOCAL_PATH)/../../api/tesseractmain.cpp \
 
@@ -47,11 +40,10 @@ LOCAL_SRC_FILES := $(LOCAL_SRC_FILES:$(LOCAL_PATH)/%=%)
 $(info local src files  = $(LOCAL_SRC_FILES))
 
 LOCAL_LDLIBS := -ldl -llog -ljnigraphics
-LOCAL_CFLAGS := -DANDROID_BUILD -DNO_CUBE_BUILD -DGRAPHICS_DISABLED
+LOCAL_CFLAGS := -DANDROID_BUILD -DGRAPHICS_DISABLED
 
 include $(BUILD_SHARED_LIBRARY)
 
-$(call import-module,mobile/base)
+$(call import-module,base/port)
-$(call import-module,mobile/base)
 $(call import-module,mobile/util/hash)
 $(call import-module,third_party/leptonica/android/jni)

api/Makefile.am

@@ -1,6 +1,7 @@
 AM_CPPFLAGS += -DLOCALEDIR=\"$(localedir)\"\
     -DUSE_STD_NAMESPACE \
-    -I$(top_srcdir)/ccutil -I$(top_srcdir)/ccstruct -I$(top_srcdir)/cube \
+    -I$(top_srcdir)/arch -I$(top_srcdir)/lstm \
+    -I$(top_srcdir)/ccutil -I$(top_srcdir)/ccstruct \
     -I$(top_srcdir)/viewer \
     -I$(top_srcdir)/textord -I$(top_srcdir)/dict \
     -I$(top_srcdir)/classify -I$(top_srcdir)/ccmain \
@@ -27,15 +28,15 @@ libtesseract_api_la_LIBADD = \
     ../wordrec/libtesseract_wordrec.la \
     ../classify/libtesseract_classify.la \
     ../dict/libtesseract_dict.la \
+    ../arch/libtesseract_arch.la \
+    ../arch/libtesseract_avx.la \
+    ../arch/libtesseract_sse.la \
+    ../lstm/libtesseract_lstm.la \
     ../ccstruct/libtesseract_ccstruct.la \
     ../cutil/libtesseract_cutil.la \
     ../viewer/libtesseract_viewer.la \
     ../ccutil/libtesseract_ccutil.la \
     ../opencl/libtesseract_opencl.la 
-    if !NO_CUBE_BUILD
-        libtesseract_api_la_LIBADD += ../cube/libtesseract_cube.la \
-            ../neural_networks/runtime/libtesseract_neural.la \
-    endif
 endif
 
 libtesseract_api_la_CPPFLAGS = $(AM_CPPFLAGS)
@@ -45,7 +46,7 @@ endif
 libtesseract_api_la_SOURCES = baseapi.cpp capi.cpp renderer.cpp pdfrenderer.cpp
 
 lib_LTLIBRARIES += libtesseract.la
-libtesseract_la_LDFLAGS = 
+libtesseract_la_LDFLAGS = $(LEPTONICA_LIBS) $(OPENCL_LDFLAGS)
 libtesseract_la_SOURCES =
 # Dummy C++ source to cause C++ linking.
 # see http://www.gnu.org/s/hello/manual/automake/Libtool-Convenience-Libraries.html#Libtool-Convenience-Libraries
@@ -57,15 +58,15 @@ libtesseract_la_LIBADD = \
     ../wordrec/libtesseract_wordrec.la \
     ../classify/libtesseract_classify.la \
     ../dict/libtesseract_dict.la \
+    ../arch/libtesseract_arch.la \
+    ../arch/libtesseract_avx.la \
+    ../arch/libtesseract_sse.la \
+    ../lstm/libtesseract_lstm.la \
     ../ccstruct/libtesseract_ccstruct.la \
     ../cutil/libtesseract_cutil.la \
     ../viewer/libtesseract_viewer.la \
     ../ccutil/libtesseract_ccutil.la \
     ../opencl/libtesseract_opencl.la 
-if !NO_CUBE_BUILD
-libtesseract_la_LIBADD += ../cube/libtesseract_cube.la \
-        ../neural_networks/runtime/libtesseract_neural.la
-endif
 
 libtesseract_la_LDFLAGS += -version-info $(GENERIC_LIBRARY_VERSION) -no-undefined
 
@@ -81,9 +82,10 @@ tesseract_LDADD = libtesseract.la
 
 tesseract_LDFLAGS = $(OPENCL_LDFLAGS)
 
-if OPENMP
+tesseract_LDADD += $(LEPTONICA_LIBS)
-tesseract_LDADD += $(OPENMP_CFLAGS)
+tesseract_LDADD += $(OPENMP_CXXFLAGS)
-endif
+
+tesseract_LDADD += -ltiff
 
 if T_WIN
 tesseract_LDADD += -lws2_32
@@ -92,4 +94,3 @@ endif
 if ADD_RT
 tesseract_LDADD += -lrt
 endif
-

api/apitypes.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_API_APITYPES_H__
+#ifndef TESSERACT_API_APITYPES_H_
-#define TESSERACT_API_APITYPES_H__
+#define TESSERACT_API_APITYPES_H_
 
 #include "publictypes.h"
 
@@ -30,4 +30,4 @@
 // than the lower-level one, and lower-level code should be sure to include
 // only the lower-level file.
 
-#endif  // TESSERACT_API_APITYPES_H__
+#endif  // TESSERACT_API_APITYPES_H_

api/baseapi.cpp

@@ -34,8 +34,6 @@
 // workaround for stdlib.h with -std=c++11 for _splitpath and _MAX_FNAME
 #undef __STRICT_ANSI__
 #endif  // _MSC_VER
-#include <stdlib.h>
-#include <windows.h>
 #include <fcntl.h>
 #include <io.h>
 #else
@@ -110,27 +108,30 @@ const int kMinCredibleResolution = 70;
 const int kMaxCredibleResolution = 2400;
 
 TessBaseAPI::TessBaseAPI()
-  : tesseract_(NULL),
+    : tesseract_(nullptr),
-    osd_tesseract_(NULL),
+      osd_tesseract_(nullptr),
-    equ_detect_(NULL),
+      equ_detect_(nullptr),
+      reader_(nullptr),
       // Thresholder is initialized to NULL here, but will be set before use by:
       // A constructor of a derived API,  SetThresholder(), or
       // created implicitly when used in InternalSetImage.
-    thresholder_(NULL),
+      thresholder_(nullptr),
-    paragraph_models_(NULL),
+      paragraph_models_(nullptr),
-    block_list_(NULL),
+      block_list_(nullptr),
-    page_res_(NULL),
+      page_res_(nullptr),
-    input_file_(NULL),
+      input_file_(nullptr),
-    input_image_(NULL),
+      output_file_(nullptr),
-    output_file_(NULL),
+      datapath_(nullptr),
-    datapath_(NULL),
+      language_(nullptr),
-    language_(NULL),
       last_oem_requested_(OEM_DEFAULT),
       recognition_done_(false),
       truth_cb_(NULL),
-    rect_left_(0), rect_top_(0), rect_width_(0), rect_height_(0),
+      rect_left_(0),
-    image_width_(0), image_height_(0) {
+      rect_top_(0),
-}
+      rect_width_(0),
+      rect_height_(0),
+      image_width_(0),
+      image_height_(0) {}
 
 TessBaseAPI::~TessBaseAPI() {
   End();
@@ -278,20 +279,33 @@ int TessBaseAPI::Init(const char* datapath, const char* language,
                       const GenericVector<STRING> *vars_vec,
                       const GenericVector<STRING> *vars_values,
                       bool set_only_non_debug_params) {
+  return Init(datapath, 0, language, oem, configs, configs_size, vars_vec,
+              vars_values, set_only_non_debug_params, nullptr);
+}
+
+// In-memory version reads the traineddata file directly from the given
+// data[data_size] array. Also implements the version with a datapath in data,
+// flagged by data_size = 0.
+int TessBaseAPI::Init(const char* data, int data_size, const char* language,
+                      OcrEngineMode oem, char** configs, int configs_size,
+                      const GenericVector<STRING>* vars_vec,
+                      const GenericVector<STRING>* vars_values,
+                      bool set_only_non_debug_params, FileReader reader) {
   PERF_COUNT_START("TessBaseAPI::Init")
   // Default language is "eng".
-  if (language == NULL) language = "eng";
+  if (language == nullptr) language = "eng";
+  STRING datapath = data_size == 0 ? data : language;
   // If the datapath, OcrEngineMode or the language have changed - start again.
   // Note that the language_ field stores the last requested language that was
   // initialized successfully, while tesseract_->lang stores the language
   // actually used. They differ only if the requested language was NULL, in
   // which case tesseract_->lang is set to the Tesseract default ("eng").
-  if (tesseract_ != NULL &&
+  if (tesseract_ != nullptr &&
-      (datapath_ == NULL || language_ == NULL ||
+      (datapath_ == nullptr || language_ == nullptr || *datapath_ != datapath ||
-       *datapath_ != datapath || last_oem_requested_ != oem ||
+       last_oem_requested_ != oem ||
        (*language_ != language && tesseract_->lang != language))) {
     delete tesseract_;
-    tesseract_ = NULL;
+    tesseract_ = nullptr;
   }
   // PERF_COUNT_SUB("delete tesseract_")
 #ifdef USE_OPENCL
@@ -300,19 +314,25 @@ int TessBaseAPI::Init(const char* datapath, const char* language,
 #endif
   PERF_COUNT_SUB("OD::InitEnv()")
   bool reset_classifier = true;
-  if (tesseract_ == NULL) {
+  if (tesseract_ == nullptr) {
     reset_classifier = false;
     tesseract_ = new Tesseract;
+    if (reader != nullptr) reader_ = reader;
+    TessdataManager mgr(reader_);
+    if (data_size != 0) {
+      mgr.LoadMemBuffer(language, data, data_size);
+    }
     if (tesseract_->init_tesseract(
-        datapath, output_file_ != NULL ? output_file_->string() : NULL,
+            datapath.string(),
+            output_file_ != nullptr ? output_file_->string() : nullptr,
             language, oem, configs, configs_size, vars_vec, vars_values,
-        set_only_non_debug_params) != 0) {
+            set_only_non_debug_params, &mgr) != 0) {
       return -1;
     }
   }
   PERF_COUNT_SUB("update tesseract_")
   // Update datapath and language requested for the last valid initialization.
-  if (datapath_ == NULL)
+  if (datapath_ == nullptr)
     datapath_ = new STRING(datapath);
   else
     *datapath_ = datapath;
@@ -320,7 +340,7 @@ int TessBaseAPI::Init(const char* datapath, const char* language,
       (strcmp(tesseract_->datadir.string(), "") != 0))
      *datapath_ = tesseract_->datadir;
 
-  if (language_ == NULL)
+  if (language_ == nullptr)
     language_ = new STRING(language);
   else
     *language_ = language;
@@ -424,7 +444,8 @@ int TessBaseAPI::InitLangMod(const char* datapath, const char* language) {
     tesseract_ = new Tesseract;
   else
     ParamUtils::ResetToDefaults(tesseract_->params());
-  return tesseract_->init_tesseract_lm(datapath, NULL, language);
+  TessdataManager mgr;
+  return tesseract_->init_tesseract_lm(datapath, NULL, language, &mgr);
 }
 
 /**
@@ -434,7 +455,7 @@ int TessBaseAPI::InitLangMod(const char* datapath, const char* language) {
 void TessBaseAPI::InitForAnalysePage() {
   if (tesseract_ == NULL) {
     tesseract_ = new Tesseract;
-    tesseract_->InitAdaptiveClassifier(false);
+    tesseract_->InitAdaptiveClassifier(nullptr);
   }
 }
 
@@ -515,9 +536,7 @@ void TessBaseAPI::ClearAdaptiveClassifier() {
 
 /**
  * Provide an image for Tesseract to recognize. Format is as
- * TesseractRect above. Does not copy the image buffer, or take
+ * TesseractRect above. Copies the image buffer and converts to Pix.
- * ownership. The source image may be destroyed after Recognize is called,
- * either explicitly or implicitly via one of the Get*Text functions.
  * SetImage clears all recognition results, and sets the rectangle to the
  * full image, so it may be followed immediately by a GetUTF8Text, and it
  * will automatically perform recognition.
@@ -525,9 +544,11 @@ void TessBaseAPI::ClearAdaptiveClassifier() {
 void TessBaseAPI::SetImage(const unsigned char* imagedata,
                            int width, int height,
                            int bytes_per_pixel, int bytes_per_line) {
-  if (InternalSetImage())
+  if (InternalSetImage()) {
     thresholder_->SetImage(imagedata, width, height,
                            bytes_per_pixel, bytes_per_line);
+    SetInputImage(thresholder_->GetPixRect());
+  }
 }
 
 void TessBaseAPI::SetSourceResolution(int ppi) {
@@ -539,18 +560,17 @@ void TessBaseAPI::SetSourceResolution(int ppi) {
 
 /**
  * Provide an image for Tesseract to recognize. As with SetImage above,
- * Tesseract doesn't take a copy or ownership or pixDestroy the image, so
+ * Tesseract takes its own copy of the image, so it need not persist until
- * it must persist until after Recognize.
+ * after Recognize.
  * Pix vs raw, which to use?
- * Use Pix where possible. A future version of Tesseract may choose to use Pix
+ * Use Pix where possible. Tesseract uses Pix as its internal representation
- * as its internal representation and discard IMAGE altogether.
+ * and it is therefore more efficient to provide a Pix directly.
- * Because of that, an implementation that sources and targets Pix may end up
- * with less copies than an implementation that does not.
  */
 void TessBaseAPI::SetImage(Pix* pix) {
-  if (InternalSetImage())
+  if (InternalSetImage()) {
     thresholder_->SetImage(pix);
-  SetInputImage(pix);
+    SetInputImage(thresholder_->GetPixRect());
+  }
 }
 
 /**
@@ -693,8 +713,8 @@ Boxa* TessBaseAPI::GetComponentImages(PageIteratorLevel level,
       if (pixa != NULL) {
         Pix* pix = NULL;
         if (raw_image) {
-          pix = page_it->GetImage(level, raw_padding, input_image_,
+          pix = page_it->GetImage(level, raw_padding, GetInputImage(), &left,
-                                  &left, &top);
+                                  &top);
         } else {
           pix = page_it->GetBinaryImage(level);
         }
@@ -747,53 +767,6 @@ void TessBaseAPI::DumpPGM(const char* filename) {
   fclose(fp);
 }
 
-#ifndef NO_CUBE_BUILD
-/**
- * Placeholder for call to Cube and test that the input data is correct.
- * reskew is the direction of baselines in the skewed image in
- * normalized (cos theta, sin theta) form, so (0.866, 0.5) would represent
- * a 30 degree anticlockwise skew.
- */
-int CubeAPITest(Boxa* boxa_blocks, Pixa* pixa_blocks,
-                Boxa* boxa_words, Pixa* pixa_words,
-                const FCOORD& reskew, Pix* page_pix,
-                PAGE_RES* page_res) {
-  int block_count = boxaGetCount(boxa_blocks);
-  ASSERT_HOST(block_count == pixaGetCount(pixa_blocks));
-  // Write each block to the current directory as junk_write_display.nnn.png.
-  for (int i = 0; i < block_count; ++i) {
-    Pix* pix = pixaGetPix(pixa_blocks, i, L_CLONE);
-    pixDisplayWrite(pix, 1);
-  }
-  int word_count = boxaGetCount(boxa_words);
-  ASSERT_HOST(word_count == pixaGetCount(pixa_words));
-  int pr_word = 0;
-  PAGE_RES_IT page_res_it(page_res);
-  for (page_res_it.restart_page(); page_res_it.word () != NULL;
-       page_res_it.forward(), ++pr_word) {
-    WERD_RES *word = page_res_it.word();
-    WERD_CHOICE* choice = word->best_choice;
-    // Write the first 100 words to files names wordims/<wordstring>.tif.
-    if (pr_word < 100) {
-      STRING filename("wordims/");
-      if (choice != NULL) {
-        filename += choice->unichar_string();
-      } else {
-        char numbuf[32];
-        filename += "unclassified";
-        snprintf(numbuf, 32, "%03d", pr_word);
-        filename += numbuf;
-      }
-      filename += ".tif";
-      Pix* pix = pixaGetPix(pixa_words, pr_word, L_CLONE);
-      pixWrite(filename.string(), pix, IFF_TIFF_G4);
-    }
-  }
-  ASSERT_HOST(pr_word == word_count);
-  return 0;
-}
-#endif  // NO_CUBE_BUILD
-
 /**
  * Runs page layout analysis in the mode set by SetPageSegMode.
  * May optionally be called prior to Recognize to get access to just
@@ -809,9 +782,7 @@ int CubeAPITest(Boxa* boxa_blocks, Pixa* pixa_blocks,
  * has not been subjected to a call of Init, SetImage, Recognize, Clear, End
  * DetectOS, or anything else that changes the internal PAGE_RES.
  */
-PageIterator* TessBaseAPI::AnalyseLayout() {
+PageIterator* TessBaseAPI::AnalyseLayout() { return AnalyseLayout(false); }
-  return AnalyseLayout(false);
-}
 
 PageIterator* TessBaseAPI::AnalyseLayout(bool merge_similar_words) {
   if (FindLines() == 0) {
@@ -836,7 +807,6 @@ int TessBaseAPI::Recognize(ETEXT_DESC* monitor) {
     return -1;
   if (FindLines() != 0)
     return -1;
-  if (page_res_ != NULL)
   delete page_res_;
   if (block_list_->empty()) {
     page_res_ = new PAGE_RES(false, block_list_,
@@ -851,13 +821,17 @@ int TessBaseAPI::Recognize(ETEXT_DESC* monitor) {
   } else if (tesseract_->tessedit_resegment_from_boxes) {
     page_res_ = tesseract_->ApplyBoxes(*input_file_, false, block_list_);
   } else {
-    // TODO(rays) LSTM here.
+    page_res_ = new PAGE_RES(tesseract_->AnyLSTMLang(),
-    page_res_ = new PAGE_RES(false,
                              block_list_, &tesseract_->prev_word_best_choice_);
   }
   if (page_res_ == NULL) {
     return -1;
   }
+  if (tesseract_->tessedit_train_line_recognizer) {
+    tesseract_->TrainLineRecognizer(*input_file_, *output_file_, block_list_);
+    tesseract_->CorrectClassifyWords(page_res_);
+    return 0;
+  }
   if (tesseract_->tessedit_make_boxes_from_boxes) {
     tesseract_->CorrectClassifyWords(page_res_);
     return 0;
@@ -940,17 +914,10 @@ int TessBaseAPI::RecognizeForChopTest(ETEXT_DESC* monitor) {
   return 0;
 }
 
-void TessBaseAPI::SetInputImage(Pix *pix) {
+// Takes ownership of the input pix.
-  if (input_image_)
+void TessBaseAPI::SetInputImage(Pix* pix) { tesseract_->set_pix_original(pix); }
-    pixDestroy(&input_image_);
-  input_image_ = NULL;
-  if (pix)
-    input_image_ = pixCopy(NULL, pix);
-}
 
-Pix* TessBaseAPI::GetInputImage() {
+Pix* TessBaseAPI::GetInputImage() { return tesseract_->pix_original(); }
-  return input_image_;
-}
 
 const char * TessBaseAPI::GetInputName() {
   if (input_file_)
@@ -994,8 +961,7 @@ bool TessBaseAPI::ProcessPagesFileList(FILE *flist,
   }
 
   // Begin producing output
-  const char* kUnknownTitle = "";
+  if (renderer && !renderer->BeginDocument(unknown_title_)) {
-  if (renderer && !renderer->BeginDocument(kUnknownTitle)) {
     return false;
   }
 
@@ -1038,26 +1004,13 @@ bool TessBaseAPI::ProcessPagesMultipageTiff(const l_uint8 *data,
                                             int tessedit_page_number) {
 #ifndef ANDROID_BUILD
   Pix *pix = NULL;
-#ifdef USE_OPENCL
-  OpenclDevice od;
-#endif  // USE_OPENCL
   int page = (tessedit_page_number >= 0) ? tessedit_page_number : 0;
+  size_t offset = 0;
   for (; ; ++page) {
     if (tessedit_page_number >= 0)
       page = tessedit_page_number;
-#ifdef USE_OPENCL
+    pix = (data) ? pixReadMemFromMultipageTiff(data, size, &offset)
-    if ( od.selectedDeviceIsOpenCL() ) {
+                 : pixReadFromMultipageTiff(filename, &offset);
-      pix = (data) ?
-          od.pixReadMemTiffCl(data, size, page) :
-          od.pixReadTiffCl(filename, page);
-    } else {
-#endif  // USE_OPENCL
-      pix = (data) ?
-          pixReadMemTiff(data, size, page) :
-          pixReadTiff(filename, page);
-#ifdef USE_OPENCL
-    }
-#endif  // USE_OPENCL
     if (pix == NULL) break;
     tprintf("Page %d\n", page + 1);
     char page_str[kMaxIntSize];
@@ -1068,6 +1021,7 @@ bool TessBaseAPI::ProcessPagesMultipageTiff(const l_uint8 *data,
     pixDestroy(&pix);
     if (!r) return false;
     if (tessedit_page_number >= 0) break;
+    if (!offset) break;
   }
   return true;
 #else
@@ -1107,7 +1061,6 @@ bool TessBaseAPI::ProcessPagesInternal(const char* filename,
                                        const char* retry_config,
                                        int timeout_millisec,
                                        TessResultRenderer* renderer) {
-#ifndef ANDROID_BUILD
   PERF_COUNT_START("ProcessPages")
   bool stdInput = !strcmp(filename, "stdin") || !strcmp(filename, "-");
   if (stdInput) {
@@ -1142,7 +1095,15 @@ bool TessBaseAPI::ProcessPagesInternal(const char* filename,
 
   // Maybe we have a filelist
   if (r != 0 || format == IFF_UNKNOWN) {
-    STRING s(buf.c_str());
+    STRING s;
+    if (stdInput) {
+      s = buf.c_str();
+    } else {
+      std::ifstream t(filename);
+      std::string u((std::istreambuf_iterator<char>(t)),
+                    std::istreambuf_iterator<char>());
+      s = u.c_str();
+    }
     return ProcessPagesFileList(NULL, &s, retry_config,
                                 timeout_millisec, renderer,
                                 tesseract_->tessedit_page_number);
@@ -1164,8 +1125,7 @@ bool TessBaseAPI::ProcessPagesInternal(const char* filename,
   }
 
   // Begin the output
-  const char* kUnknownTitle = "";
+  if (renderer && !renderer->BeginDocument(unknown_title_)) {
-  if (renderer && !renderer->BeginDocument(kUnknownTitle)) {
     pixDestroy(&pix);
     return false;
   }
@@ -1187,9 +1147,6 @@ bool TessBaseAPI::ProcessPagesInternal(const char* filename,
   }
   PERF_COUNT_END
   return true;
-#else
-  return false;
-#endif
 }
 
 bool TessBaseAPI::ProcessPage(Pix* pix, int page_index, const char* filename,
@@ -1379,8 +1336,9 @@ static void AddBaselineCoordsTohOCR(const PageIterator *it,
   hocr_str->add_str_double(" ", round(p0 * 1000.0) / 1000.0);
 }
 
-static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1, int num2) {
+static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1,
-  const unsigned long BUFSIZE = 64;
+                        int num2) {
+  const size_t BUFSIZE = 64;
   char id_buffer[BUFSIZE];
   if (num2 >= 0) {
     snprintf(id_buffer, BUFSIZE - 1, "%s_%d_%d", base.c_str(), num1, num2);
@@ -1393,8 +1351,7 @@ static void AddIdTohOCR(STRING* hocr_str, const std::string base, int num1, int
   *hocr_str += "'";
 }
 
-static void AddBoxTohOCR(const ResultIterator *it,
+static void AddBoxTohOCR(const ResultIterator* it, PageIteratorLevel level,
-                         PageIteratorLevel level,
                          STRING* hocr_str) {
   int left, top, right, bottom;
   it->BoundingBox(level, &left, &top, &right, &bottom);
@@ -1410,7 +1367,7 @@ static void AddBoxTohOCR(const ResultIterator *it,
     // add custom height measures
     float row_height, descenders, ascenders;  // row attributes
     it->RowAttributes(&row_height, &descenders, &ascenders);
-    // TODO: Do we want to limit these to a single decimal place?
+    // TODO(rays): Do we want to limit these to a single decimal place?
     hocr_str->add_str_double("; x_size ", row_height);
     hocr_str->add_str_double("; x_descenders ", descenders * -1);
     hocr_str->add_str_double("; x_ascenders ", ascenders);
@@ -1418,8 +1375,7 @@ static void AddBoxTohOCR(const ResultIterator *it,
   *hocr_str += "\">";
 }
 
-static void AddBoxToTSV(const PageIterator *it,
+static void AddBoxToTSV(const PageIterator* it, PageIteratorLevel level,
-                         PageIteratorLevel level,
                         STRING* hocr_str) {
   int left, top, right, bottom;
   it->BoundingBox(level, &left, &top, &right, &bottom);
@@ -1429,8 +1385,6 @@ static void AddBoxToTSV(const PageIterator *it,
   hocr_str->add_str_int("\t", bottom - top);
 }
 
-
-
 /**
  * Make a HTML-formatted string with hOCR markup from the internal
  * data structures.
@@ -1440,7 +1394,7 @@ static void AddBoxToTSV(const PageIterator *it,
  * STL removed from original patch submission and refactored by rays.
  */
 char* TessBaseAPI::GetHOCRText(int page_number) {
-  return GetHOCRText(NULL,page_number);
+  return GetHOCRText(NULL, page_number);
 }
 
 /**
@@ -1452,8 +1406,7 @@ char* TessBaseAPI::GetHOCRText(int page_number) {
  * STL removed from original patch submission and refactored by rays.
  */
 char* TessBaseAPI::GetHOCRText(ETEXT_DESC* monitor, int page_number) {
-  if (tesseract_ == NULL ||
+  if (tesseract_ == NULL || (page_res_ == NULL && Recognize(monitor) < 0))
-      (page_res_ == NULL && Recognize(monitor) < 0))
     return NULL;
 
   int lcnt = 1, bcnt = 1, pcnt = 1, wcnt = 1;
@@ -1470,13 +1423,13 @@ char* TessBaseAPI::GetHOCRText(ETEXT_DESC* monitor, int page_number) {
 
 #ifdef _WIN32
   // convert input name from ANSI encoding to utf-8
-  int str16_len = MultiByteToWideChar(CP_ACP, 0, input_file_->string(), -1,
+  int str16_len =
-                                      NULL, 0);
+      MultiByteToWideChar(CP_ACP, 0, input_file_->string(), -1, NULL, 0);
   wchar_t *uni16_str = new WCHAR[str16_len];
   str16_len = MultiByteToWideChar(CP_ACP, 0, input_file_->string(), -1,
                                   uni16_str, str16_len);
-  int utf8_len = WideCharToMultiByte(CP_UTF8, 0, uni16_str, str16_len, NULL,
+  int utf8_len = WideCharToMultiByte(CP_UTF8, 0, uni16_str, str16_len, NULL, 0,
-                                     0, NULL, NULL);
+                                     NULL, NULL);
   char *utf8_str = new char[utf8_len];
   WideCharToMultiByte(CP_UTF8, 0, uni16_str, str16_len, utf8_str,
                       utf8_len, NULL, NULL);
@@ -1567,8 +1520,12 @@ char* TessBaseAPI::GetHOCRText(ETEXT_DESC* monitor, int page_number) {
     }
     switch (res_it->WordDirection()) {
       // Only emit direction if different from current paragraph direction
-      case DIR_LEFT_TO_RIGHT: if (!para_is_ltr) hocr_str += " dir='ltr'"; break;
+      case DIR_LEFT_TO_RIGHT:
-      case DIR_RIGHT_TO_LEFT: if (para_is_ltr) hocr_str += " dir='rtl'"; break;
+        if (!para_is_ltr) hocr_str += " dir='ltr'";
+        break;
+      case DIR_RIGHT_TO_LEFT:
+        if (para_is_ltr) hocr_str += " dir='rtl'";
+        break;
       case DIR_MIX:
       case DIR_NEUTRAL:
       default:  // Do nothing.
@@ -1620,8 +1577,7 @@ char* TessBaseAPI::GetHOCRText(ETEXT_DESC* monitor, int page_number) {
  * page_number is 0-based but will appear in the output as 1-based.
  */
 char* TessBaseAPI::GetTSVText(int page_number) {
-  if (tesseract_ == NULL ||
+  if (tesseract_ == NULL || (page_res_ == NULL && Recognize(NULL) < 0))
-      (page_res_ == NULL && Recognize(NULL) < 0))
     return NULL;
 
   int lcnt = 1, bcnt = 1, pcnt = 1, wcnt = 1;
@@ -1629,7 +1585,8 @@ char* TessBaseAPI::GetTSVText(int page_number) {
 
   STRING tsv_str("");
 
-  int page_num = page_id, block_num = 0, par_num = 0, line_num = 0, word_num = 0;
+  int page_num = page_id, block_num = 0, par_num = 0, line_num = 0,
+      word_num = 0;
 
   tsv_str.add_str_int("1\t", page_num);  // level 1 - page
   tsv_str.add_str_int("\t", block_num);
@@ -1642,7 +1599,7 @@ char* TessBaseAPI::GetTSVText(int page_number) {
   tsv_str.add_str_int("\t", rect_height_);
   tsv_str += "\t-1\t\n";
 
-  ResultIterator *res_it = GetIterator();
+  ResultIterator* res_it = GetIterator();
   while (!res_it->Empty(RIL_BLOCK)) {
     if (res_it->Empty(RIL_WORD)) {
       res_it->Next(RIL_WORD);
@@ -1683,13 +1640,7 @@ char* TessBaseAPI::GetTSVText(int page_number) {
 
     // Now, process the word...
     int left, top, right, bottom;
-    bool bold, italic, underlined, monospace, serif, smallcaps;
-    int pointsize, font_id;
-    const char *font_name;
     res_it->BoundingBox(RIL_WORD, &left, &top, &right, &bottom);
-    font_name = res_it->WordFontAttributes(&bold, &italic, &underlined,
-                                           &monospace, &serif, &smallcaps,
-                                           &pointsize, &font_id);
     word_num++;
     tsv_str.add_str_int("5\t", page_num);  // level 5 - word
     tsv_str.add_str_int("\t", block_num);
@@ -1716,7 +1667,7 @@ char* TessBaseAPI::GetTSVText(int page_number) {
     wcnt++;
   }
 
-  char *ret = new char[tsv_str.length() + 1];
+  char* ret = new char[tsv_str.length() + 1];
   strcpy(ret, tsv_str.string());
   delete res_it;
   return ret;
@@ -1760,7 +1711,7 @@ char* TessBaseAPI::GetBoxText(int page_number) {
   int total_length = blob_count * kBytesPerBoxFileLine + utf8_length +
       kMaxBytesPerLine;
   char* result = new char[total_length];
-  strcpy(result, "\0");
+  result[0] = '\0';
   int output_length = 0;
   LTRResultIterator* it = GetLTRIterator();
   do {
@@ -1907,43 +1858,70 @@ char* TessBaseAPI::GetUNLVText() {
   return result;
 }
 
-  /**
+/**
+ * Detect the orientation of the input image and apparent script (alphabet).
+ * orient_deg is the detected clockwise rotation of the input image in degrees
+ * (0, 90, 180, 270)
+ * orient_conf is the confidence (15.0 is reasonably confident)
+ * script_name is an ASCII string, the name of the script, e.g. "Latin"
+ * script_conf is confidence level in the script
+ * Returns true on success and writes values to each parameter as an output
+ */
+bool TessBaseAPI::DetectOrientationScript(int* orient_deg, float* orient_conf,
+                                          const char** script_name,
+                                          float* script_conf) {
+  OSResults osr;
+
+  bool osd = DetectOS(&osr);
+  if (!osd) {
+    return false;
+  }
+
+  int orient_id = osr.best_result.orientation_id;
+  int script_id = osr.get_best_script(orient_id);
+  if (orient_conf) *orient_conf = osr.best_result.oconfidence;
+  if (orient_deg) *orient_deg = orient_id * 90;  // convert quadrant to degrees
+
+  if (script_name) {
+    const char* script = osr.unicharset->get_script_from_script_id(script_id);
+
+    *script_name = script;
+  }
+
+  if (script_conf) *script_conf = osr.best_result.sconfidence;
+
+  return true;
+}
+
+/**
  * The recognized text is returned as a char* which is coded
  * as UTF8 and must be freed with the delete [] operator.
  * page_number is a 0-based page index that will appear in the osd file.
  */
 char* TessBaseAPI::GetOsdText(int page_number) {
-  OSResults osr;
+  int orient_deg;
+  float orient_conf;
+  const char* script_name;
+  float script_conf;
 
-  bool osd = DetectOS(&osr);
+  if (!DetectOrientationScript(&orient_deg, &orient_conf, &script_name,
-  if (!osd) {
+                               &script_conf))
     return NULL;
-  }
-
-  int orient_id = osr.best_result.orientation_id;
-  int script_id = osr.get_best_script(orient_id);
-  float orient_conf = osr.best_result.oconfidence;
-  float script_conf = osr.best_result.sconfidence;
-  const char* script_name = 
-      osr.unicharset->get_script_from_script_id(script_id);
-
-  // clockwise orientation of the input image, in degrees
-  int orient_deg = orient_id * 90; 
 
   // clockwise rotation needed to make the page upright
-  int rotate =  OrientationIdToValue(orient_id);
+  int rotate = OrientationIdToValue(orient_deg / 90);
 
-  char* osd_buf = new char[255];
+  const int kOsdBufsize = 255;
-  snprintf(osd_buf, 255,
+  char* osd_buf = new char[kOsdBufsize];
+  snprintf(osd_buf, kOsdBufsize,
            "Page number: %d\n"
            "Orientation in degrees: %d\n"
            "Rotate: %d\n"
            "Orientation confidence: %.2f\n"
            "Script: %s\n"
            "Script confidence: %.2f\n",
-          page_number,
+           page_number, orient_deg, rotate, orient_conf, script_name,
-          orient_deg, rotate, orient_conf,
+           script_conf);
-          script_name, script_conf);
 
   return osd_buf;
 }
@@ -2020,8 +1998,7 @@ bool TessBaseAPI::AdaptToWordStr(PageSegMode mode, const char* wordstr) {
     for (t = 0; text[t] != '\0'; ++t) {
       if (text[t] == '\n' || text[t] == ' ')
         continue;
-      while (wordstr[w] != '\0' && wordstr[w] == ' ')
+      while (wordstr[w] == ' ') ++w;
-        ++w;
       if (text[t] != wordstr[w])
         break;
       ++w;
@@ -2063,7 +2040,7 @@ void TessBaseAPI::Clear() {
   if (thresholder_ != NULL)
     thresholder_->Clear();
   ClearResults();
-  SetInputImage(NULL);
+  if (tesseract_ != NULL) SetInputImage(NULL);
 }
 
 /**
@@ -2073,6 +2050,7 @@ void TessBaseAPI::Clear() {
  * other than Init and anything declared above it in the class definition.
  */
 void TessBaseAPI::End() {
+  Clear();
   if (thresholder_ != NULL) {
     delete thresholder_;
     thresholder_ = NULL;
@@ -2108,10 +2086,6 @@ void TessBaseAPI::End() {
     delete input_file_;
     input_file_ = NULL;
   }
-  if (input_image_ != NULL) {
-    pixDestroy(&input_image_);
-    input_image_ = NULL;
-  }
   if (output_file_ != NULL) {
     delete output_file_;
     output_file_ = NULL;
@@ -2242,6 +2216,8 @@ void TessBaseAPI::Threshold(Pix** pix) {
   if (y_res < kMinCredibleResolution || y_res > kMaxCredibleResolution) {
     // Use the minimum default resolution, as it is safer to under-estimate
     // than over-estimate resolution.
+    tprintf("Warning. Invalid resolution %d dpi. Using %d instead.\n", y_res,
+            kMinCredibleResolution);
     thresholder_->SetSourceYResolution(kMinCredibleResolution);
   }
   PageSegMode pageseg_mode =
@@ -2286,7 +2262,7 @@ int TessBaseAPI::FindLines() {
   }
   if (tesseract_ == NULL) {
     tesseract_ = new Tesseract;
-    tesseract_->InitAdaptiveClassifier(false);
+    tesseract_->InitAdaptiveClassifier(nullptr);
   }
   if (tesseract_->pix_binary() == NULL)
     Threshold(tesseract_->mutable_pix_binary());
@@ -2308,14 +2284,16 @@ int TessBaseAPI::FindLines() {
 
   Tesseract* osd_tess = osd_tesseract_;
   OSResults osr;
-  if (PSM_OSD_ENABLED(tesseract_->tessedit_pageseg_mode) && osd_tess == NULL) {
+  if (PSM_OSD_ENABLED(tesseract_->tessedit_pageseg_mode) &&
+      osd_tess == nullptr) {
     if (strcmp(language_->string(), "osd") == 0) {
       osd_tess = tesseract_;
     } else {
       osd_tesseract_ = new Tesseract;
-      if (osd_tesseract_->init_tesseract(
+      TessdataManager mgr(reader_);
-          datapath_->string(), NULL, "osd", OEM_TESSERACT_ONLY,
+      if (osd_tesseract_->init_tesseract(datapath_->string(), nullptr, "osd",
-          NULL, 0, NULL, NULL, false) == 0) {
+                                         OEM_TESSERACT_ONLY, nullptr, 0,
+                                         nullptr, nullptr, false, &mgr) == 0) {
         osd_tess = osd_tesseract_;
         osd_tesseract_->set_source_resolution(
             thresholder_->GetSourceYResolution());
@@ -2323,7 +2301,7 @@ int TessBaseAPI::FindLines() {
         tprintf("Warning: Auto orientation and script detection requested,"
                 " but osd language failed to load\n");
         delete osd_tesseract_;
-        osd_tesseract_ = NULL;
+        osd_tesseract_ = nullptr;
       }
     }
   }
@@ -2766,7 +2744,7 @@ void TessBaseAPI::GetFeaturesForBlob(TBLOB* blob,
   INT_FX_RESULT_STRUCT fx_info;
   tesseract_->ExtractFeatures(*blob, false, &bl_features,
                               &cn_features, &fx_info, &outline_counts);
-  if (cn_features.size() == 0 || cn_features.size() > MAX_NUM_INT_FEATURES) {
+  if (cn_features.empty() || cn_features.size() > MAX_NUM_INT_FEATURES) {
     *num_features = 0;
     return;  // Feature extraction failed.
   }
@@ -2847,13 +2825,6 @@ int TessBaseAPI::NumDawgs() const {
   return tesseract_ == NULL ? 0 : tesseract_->getDict().NumDawgs();
 }
 
-#ifndef NO_CUBE_BUILD
-/** Return a pointer to underlying CubeRecoContext object if present. */
-CubeRecoContext *TessBaseAPI::GetCubeRecoContext() const {
-  return (tesseract_ == NULL) ? NULL : tesseract_->GetCubeRecoContext();
-}
-#endif  // NO_CUBE_BUILD
-
 /** Escape a char string - remove <>&"' with HTML codes. */
 STRING HOcrEscape(const char* text) {
   STRING ret;

api/baseapi.h

@@ -17,11 +17,11 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_API_BASEAPI_H__
+#ifndef TESSERACT_API_BASEAPI_H_
-#define TESSERACT_API_BASEAPI_H__
+#define TESSERACT_API_BASEAPI_H_
 
-#define TESSERACT_VERSION_STR "3.05.00dev"
+#define TESSERACT_VERSION_STR "4.00.00alpha"
-#define TESSERACT_VERSION 0x030500
+#define TESSERACT_VERSION 0x040000
 #define MAKE_VERSION(major, minor, patch) (((major) << 16) | ((minor) << 8) | \
                                             (patch))
 
@@ -29,14 +29,15 @@
 // To avoid collision with other typenames include the ABSOLUTE MINIMUM
 // complexity of includes here. Use forward declarations wherever possible
 // and hide includes of complex types in baseapi.cpp.
-#include "platform.h"
 #include "apitypes.h"
+#include "pageiterator.h"
+#include "platform.h"
+#include "publictypes.h"
+#include "resultiterator.h"
+#include "serialis.h"
+#include "tesscallback.h"
 #include "thresholder.h"
 #include "unichar.h"
-#include "tesscallback.h"
-#include "publictypes.h"
-#include "pageiterator.h"
-#include "resultiterator.h"
 
 template <typename T> class GenericVector;
 class PAGE_RES;
@@ -65,9 +66,6 @@ struct TBLOB;
 
 namespace tesseract {
 
-#ifndef NO_CUBE_BUILD
-class CubeRecoContext;
-#endif  // NO_CUBE_BUILD
 class Dawg;
 class Dict;
 class EquationDetect;
@@ -142,6 +140,7 @@ class TESS_API TessBaseAPI {
    * is stored in the PDF so we need that as well.
    */
   const char* GetInputName();
+  // Takes ownership of the input pix.
   void SetInputImage(Pix *pix);
   Pix* GetInputImage();
   int GetSourceYResolution();
@@ -239,6 +238,13 @@ class TESS_API TessBaseAPI {
   int Init(const char* datapath, const char* language) {
     return Init(datapath, language, OEM_DEFAULT, NULL, 0, NULL, NULL, false);
   }
+  // In-memory version reads the traineddata file directly from the given
+  // data[data_size] array, and/or reads data via a FileReader.
+  int Init(const char* data, int data_size, const char* language,
+           OcrEngineMode mode, char** configs, int configs_size,
+           const GenericVector<STRING>* vars_vec,
+           const GenericVector<STRING>* vars_values,
+           bool set_only_non_debug_params, FileReader reader);
 
   /**
    * Returns the languages string used in the last valid initialization.
@@ -333,9 +339,7 @@ class TESS_API TessBaseAPI {
 
   /**
    * Provide an image for Tesseract to recognize. Format is as
-   * TesseractRect above. Does not copy the image buffer, or take
+   * TesseractRect above. Copies the image buffer and converts to Pix.
-   * ownership. The source image may be destroyed after Recognize is called,
-   * either explicitly or implicitly via one of the Get*Text functions.
    * SetImage clears all recognition results, and sets the rectangle to the
    * full image, so it may be followed immediately by a GetUTF8Text, and it
    * will automatically perform recognition.
@@ -345,13 +349,11 @@ class TESS_API TessBaseAPI {
 
   /**
    * Provide an image for Tesseract to recognize. As with SetImage above,
-   * Tesseract doesn't take a copy or ownership or pixDestroy the image, so
+   * Tesseract takes its own copy of the image, so it need not persist until
-   * it must persist until after Recognize.
+   * after Recognize.
    * Pix vs raw, which to use?
-   * Use Pix where possible. A future version of Tesseract may choose to use Pix
+   * Use Pix where possible. Tesseract uses Pix as its internal representation
-   * as its internal representation and discard IMAGE altogether.
+   * and it is therefore more efficient to provide a Pix directly.
-   * Because of that, an implementation that sources and targets Pix may end up
-   * with less copies than an implementation that does not.
    */
   void SetImage(Pix* pix);
 
@@ -376,7 +378,6 @@ class TESS_API TessBaseAPI {
    * delete it when it it is replaced or the API is destructed.
    */
   void SetThresholder(ImageThresholder* thresholder) {
-    if (thresholder_ != NULL)
     delete thresholder_;
     thresholder_ = thresholder;
     ClearResults();
@@ -622,6 +623,18 @@ class TESS_API TessBaseAPI {
    */
   char* GetUNLVText();
 
+  /**
+   * Detect the orientation of the input image and apparent script (alphabet).
+   * orient_deg is the detected clockwise rotation of the input image in degrees
+   * (0, 90, 180, 270)
+   * orient_conf is the confidence (15.0 is reasonably confident)
+   * script_name is an ASCII string, the name of the script, e.g. "Latin"
+   * script_conf is confidence level in the script
+   * Returns true on success and writes values to each parameter as an output
+   */
+  bool DetectOrientationScript(int* orient_deg, float* orient_conf,
+                               const char** script_name, float* script_conf);
+
   /**
    * The recognized text is returned as a char* which is coded
    * as UTF8 and must be freed with the delete [] operator.
@@ -750,21 +763,12 @@ class TESS_API TessBaseAPI {
    */
   static void NormalizeTBLOB(TBLOB *tblob, ROW *row, bool numeric_mode);
 
-  Tesseract* tesseract() const {
+  Tesseract* tesseract() const { return tesseract_; }
-    return tesseract_;
-  }
 
-  OcrEngineMode oem() const {
+  OcrEngineMode oem() const { return last_oem_requested_; }
-    return last_oem_requested_;
-  }
 
   void InitTruthCallback(TruthCallback *cb) { truth_cb_ = cb; }
 
-#ifndef NO_CUBE_BUILD
-  /** Return a pointer to underlying CubeRecoContext object if present. */
-  CubeRecoContext *GetCubeRecoContext() const;
-#endif  // NO_CUBE_BUILD
-
   void set_min_orientation_margin(double margin);
 
   /**
@@ -855,9 +859,7 @@ class TESS_API TessBaseAPI {
                                     int** y1,
                                     PAGE_RES* page_res);
 
-  TESS_LOCAL const PAGE_RES* GetPageRes() const {
+  TESS_LOCAL const PAGE_RES* GetPageRes() const { return page_res_; }
-    return page_res_;
-  };
   /* @} */
 
 
@@ -865,12 +867,12 @@ class TESS_API TessBaseAPI {
   Tesseract*        tesseract_;       ///< The underlying data object.
   Tesseract*        osd_tesseract_;   ///< For orientation & script detection.
   EquationDetect*   equ_detect_;      ///<The equation detector.
+  FileReader reader_;                 ///< Reads files from any filesystem.
   ImageThresholder* thresholder_;     ///< Image thresholding module.
   GenericVector<ParagraphModel *>* paragraph_models_;
   BLOCK_LIST*       block_list_;      ///< The page layout.
   PAGE_RES*         page_res_;        ///< The page-level data.
   STRING*           input_file_;      ///< Name used by training code.
-  Pix*              input_image_;     ///< Image used for searchable PDF
   STRING*           output_file_;     ///< Name used by debug code.
   STRING*           datapath_;        ///< Current location of tessdata.
   STRING*           language_;        ///< Last initialized language.
@@ -898,7 +900,7 @@ class TESS_API TessBaseAPI {
                             const char* retry_config, int timeout_millisec,
                             TessResultRenderer* renderer,
                             int tessedit_page_number);
-  // TIFF supports multipage so gets special consideration
+  // TIFF supports multipage so gets special consideration.
   bool ProcessPagesMultipageTiff(const unsigned char *data,
                                  size_t size,
                                  const char* filename,
@@ -906,10 +908,16 @@ class TESS_API TessBaseAPI {
                                  int timeout_millisec,
                                  TessResultRenderer* renderer,
                                  int tessedit_page_number);
+  // There's currently no way to pass a document title from the
+  // Tesseract command line, and we have multiple places that choose
+  // to set the title to an empty string. Using a single named
+  // variable will hopefully reduce confusion if the situation changes
+  // in the future.
+  const char *unknown_title_ = "";
 };  // class TessBaseAPI.
 
 /** Escape a char string - remove &<>"' with HTML codes. */
 STRING HOcrEscape(const char* text);
 }  // namespace tesseract.
 
-#endif  // TESSERACT_API_BASEAPI_H__
+#endif  // TESSERACT_API_BASEAPI_H_

api/capi.cpp

@@ -64,9 +64,10 @@ TESS_API TessResultRenderer* TESS_CALL TessHOcrRendererCreate2(const char* outpu
     return new TessHOcrRenderer(outputbase, font_info);
 }
 
-TESS_API TessResultRenderer* TESS_CALL TessPDFRendererCreate(const char* outputbase, const char* datadir)
+TESS_API TessResultRenderer* TESS_CALL TessPDFRendererCreate(const char* outputbase, const char* datadir,
+                                                             BOOL textonly)
 {
-    return new TessPDFRenderer(outputbase, datadir);
+    return new TessPDFRenderer(outputbase, datadir, textonly);
 }
 
 TESS_API TessResultRenderer* TESS_CALL TessUnlvRendererCreate(const char* outputbase)
@@ -538,9 +539,18 @@ TESS_API void TESS_CALL TessBaseAPISetProbabilityInContextFunc(TessBaseAPI* hand
 
 TESS_API BOOL TESS_CALL TessBaseAPIDetectOS(TessBaseAPI* handle, OSResults* results)
 {
-    return handle->DetectOS(results) ? TRUE : FALSE;
+    return FALSE; // Unsafe ABI, return FALSE always
 }
 
+TESS_API BOOL TESS_CALL TessBaseAPIDetectOrientationScript(TessBaseAPI* handle,
+                                                            int* orient_deg, float* orient_conf, const char** script_name, float* script_conf)
+{
+    bool success;
+    success = handle->DetectOrientationScript(orient_deg, orient_conf, script_name, script_conf);
+    return (BOOL)success;
+}
+
+
 TESS_API void TESS_CALL TessBaseAPIGetFeaturesForBlob(TessBaseAPI* handle, TBLOB* blob, INT_FEATURE_STRUCT* int_features,
                                                             int* num_features, int* FeatureOutlineIndex)
 {
@@ -598,13 +608,6 @@ TESS_API void TESS_CALL TessBaseAPIInitTruthCallback(TessBaseAPI* handle, TessTr
     handle->InitTruthCallback(cb);
 }
 
-#ifndef NO_CUBE_BUILD
-TESS_API TessCubeRecoContext* TESS_CALL TessBaseAPIGetCubeRecoContext(const TessBaseAPI* handle)
-{
-    return handle->GetCubeRecoContext();
-}
-#endif  // NO_CUBE_BUILD
-
 TESS_API void TESS_CALL TessBaseAPISetMinOrientationMargin(TessBaseAPI* handle, double margin)
 {
     handle->set_min_orientation_margin(margin);

api/capi.h

@@ -68,9 +68,6 @@ typedef tesseract::ProbabilityInContextFunc TessProbabilityInContextFunc;
 typedef tesseract::FillLatticeFunc TessFillLatticeFunc;
 typedef tesseract::Dawg TessDawg;
 typedef tesseract::TruthCallback TessTruthCallback;
-#ifndef NO_CUBE_BUILD
-typedef tesseract::CubeRecoContext TessCubeRecoContext;
-#endif  // NO_CUBE_BUILD
 typedef tesseract::Orientation TessOrientation;
 typedef tesseract::ParagraphJustification TessParagraphJustification;
 typedef tesseract::WritingDirection TessWritingDirection;
@@ -88,7 +85,7 @@ typedef struct TessPageIterator TessPageIterator;
 typedef struct TessResultIterator TessResultIterator;
 typedef struct TessMutableIterator TessMutableIterator;
 typedef struct TessChoiceIterator TessChoiceIterator;
-typedef enum TessOcrEngineMode     { OEM_TESSERACT_ONLY, OEM_CUBE_ONLY, OEM_TESSERACT_CUBE_COMBINED, OEM_DEFAULT } TessOcrEngineMode;
+typedef enum TessOcrEngineMode     { OEM_TESSERACT_ONLY, OEM_LSTM_ONLY, OEM_TESSERACT_LSTM_COMBINED, OEM_DEFAULT } TessOcrEngineMode;
 typedef enum TessPageSegMode       { PSM_OSD_ONLY, PSM_AUTO_OSD, PSM_AUTO_ONLY, PSM_AUTO, PSM_SINGLE_COLUMN, PSM_SINGLE_BLOCK_VERT_TEXT,
                                      PSM_SINGLE_BLOCK, PSM_SINGLE_LINE, PSM_SINGLE_WORD, PSM_CIRCLE_WORD, PSM_SINGLE_CHAR, PSM_SPARSE_TEXT,
                                      PSM_SPARSE_TEXT_OSD, PSM_COUNT } TessPageSegMode;
@@ -122,7 +119,8 @@ TESS_API void  TESS_CALL TessDeleteBlockList(BLOCK_LIST* block_list);
 TESS_API TessResultRenderer* TESS_CALL TessTextRendererCreate(const char* outputbase);
 TESS_API TessResultRenderer* TESS_CALL TessHOcrRendererCreate(const char* outputbase);
 TESS_API TessResultRenderer* TESS_CALL TessHOcrRendererCreate2(const char* outputbase, BOOL font_info);
-TESS_API TessResultRenderer* TESS_CALL TessPDFRendererCreate(const char* outputbase, const char* datadir);
+TESS_API TessResultRenderer* TESS_CALL TessPDFRendererCreate(const char* outputbase, const char* datadir,
+                                                             BOOL textonly);
 TESS_API TessResultRenderer* TESS_CALL TessUnlvRendererCreate(const char* outputbase);
 TESS_API TessResultRenderer* TESS_CALL TessBoxTextRendererCreate(const char* outputbase);
 
@@ -285,7 +283,10 @@ TESS_API void  TESS_CALL TessBaseAPIClearPersistentCache(TessBaseAPI* handle);
 TESS_API void  TESS_CALL TessBaseAPISetProbabilityInContextFunc(TessBaseAPI* handle, TessProbabilityInContextFunc f);
 
 TESS_API void  TESS_CALL TessBaseAPISetFillLatticeFunc(TessBaseAPI* handle, TessFillLatticeFunc f);
-TESS_API BOOL  TESS_CALL TessBaseAPIDetectOS(TessBaseAPI* handle, OSResults* results);
+
+// Call TessDeleteText(*best_script_name) to free memory allocated by this function
+TESS_API BOOL  TESS_CALL TessBaseAPIDetectOrientationScript(TessBaseAPI* handle,
+                                                            int* orient_deg, float* orient_conf, const char **script_name, float* script_conf);
 
 TESS_API void  TESS_CALL TessBaseAPIGetFeaturesForBlob(TessBaseAPI* handle, TBLOB* blob, INT_FEATURE_STRUCT* int_features,
                                                        int* num_features, int* FeatureOutlineIndex);
@@ -313,11 +314,6 @@ TESS_API void  TESS_CALL TessNormalizeTBLOB(TBLOB* tblob, ROW* row, BOOL numeric
 TESS_API TessOcrEngineMode
                TESS_CALL TessBaseAPIOem(const TessBaseAPI* handle);
 TESS_API void  TESS_CALL TessBaseAPIInitTruthCallback(TessBaseAPI* handle, TessTruthCallback* cb);
-
-#ifndef NO_CUBE_BUILD
-TESS_API TessCubeRecoContext*
-               TESS_CALL TessBaseAPIGetCubeRecoContext(const TessBaseAPI* handle);
-#endif  // NO_CUBE_BUILD
 #endif
 
 TESS_API void  TESS_CALL TessBaseAPISetMinOrientationMargin(TessBaseAPI* handle, double margin);

api/pdfrenderer.cpp

@@ -20,12 +20,12 @@
 #include "config_auto.h"
 #endif
 
+#include "allheaders.h"
 #include "baseapi.h"
-#include "renderer.h"
 #include "math.h"
+#include "renderer.h"
 #include "strngs.h"
 #include "tprintf.h"
-#include "allheaders.h"
 
 #ifdef _MSC_VER
 #include "mathfix.h"
@@ -159,7 +159,7 @@ CIDToGIDMap.
 
 OK there is a small problem there, if I use GID 0 then Acrobat gets
 upset about it and complains it cannot extract the font. If I set the
-CIDToGIDMap so that all the entries are 1 instead, its happy. Totally
+CIDToGIDMap so that all the entries are 1 instead, it's happy. Totally
 mad......
 
 */
@@ -169,19 +169,26 @@ namespace tesseract {
 // Use for PDF object fragments. Must be large enough
 // to hold a colormap with 256 colors in the verbose
 // PDF representation.
-const int kBasicBufSize = 2048;
+static const int kBasicBufSize = 2048;
 
 // If the font is 10 pts, nominal character width is 5 pts
-const int kCharWidth = 2;
+static const int kCharWidth = 2;
+
+// Used for memory allocation. A codepoint must take no more than this
+// many bytes, when written in the PDF way. e.g. "<0063>" for the
+// letter 'c'
+static const int kMaxBytesPerCodepoint = 20;
 
 /**********************************************************************
  * PDF Renderer interface implementation
  **********************************************************************/
 
-TessPDFRenderer::TessPDFRenderer(const char* outputbase, const char *datadir)
+TessPDFRenderer::TessPDFRenderer(const char *outputbase, const char *datadir,
+                                 bool textonly)
     : TessResultRenderer(outputbase, "pdf") {
   obj_  = 0;
   datadir_ = datadir;
+  textonly_ = textonly;
   offsets_.push_back(0);
 }
 
@@ -282,7 +289,7 @@ void AffineMatrix(int writing_direction,
   }
 }
 
-// There are some really stupid PDF viewers in the wild, such as
+// There are some really awkward PDF viewers in the wild, such as
 // 'Preview' which ships with the Mac. They do a better job with text
 // selection and highlighting when given perfectly flat baseline
 // instead of very slightly tilted. We clip small tilts to appease
@@ -302,6 +309,23 @@ void ClipBaseline(int ppi, int x1, int y1, int x2, int y2,
     *line_y1 = *line_y2 = (y1 + y2) / 2;
 }
 
+bool CodepointToUtf16be(int code, char utf16[kMaxBytesPerCodepoint]) {
+  if ((code > 0xD7FF && code < 0xE000) || code > 0x10FFFF) {
+    tprintf("Dropping invalid codepoint %d\n", code);
+    return false;
+  }
+  if (code < 0x10000) {
+    snprintf(utf16, kMaxBytesPerCodepoint, "%04X", code);
+  } else {
+    int a = code - 0x010000;
+    int high_surrogate = (0x03FF & (a >> 10)) + 0xD800;
+    int low_surrogate = (0x03FF & a) + 0xDC00;
+    snprintf(utf16, kMaxBytesPerCodepoint,
+             "%04X%04X", high_surrogate, low_surrogate);
+  }
+  return true;
+}
+
 char* TessPDFRenderer::GetPDFTextObjects(TessBaseAPI* api,
                                          double width, double height) {
   STRING pdf_str("");
@@ -326,7 +350,11 @@ char* TessPDFRenderer::GetPDFTextObjects(TessBaseAPI* api,
   pdf_str.add_str_double("", prec(width));
   pdf_str += " 0 0 ";
   pdf_str.add_str_double("", prec(height));
-  pdf_str += " 0 0 cm /Im1 Do Q\n";
+  pdf_str += " 0 0 cm";
+  if (!textonly_) {
+    pdf_str += " /Im1 Do";
+  }
+  pdf_str += " Q\n";
 
   int line_x1 = 0;
   int line_y1 = 0;
@@ -436,27 +464,15 @@ char* TessPDFRenderer::GetPDFTextObjects(TessBaseAPI* api,
       if (grapheme && grapheme[0] != '\0') {
         GenericVector<int> unicodes;
         UNICHAR::UTF8ToUnicode(grapheme, &unicodes);
-        char utf16[20];
+        char utf16[kMaxBytesPerCodepoint];
         for (int i = 0; i < unicodes.length(); i++) {
           int code = unicodes[i];
-          // Convert to UTF-16BE https://en.wikipedia.org/wiki/UTF-16
+          if (CodepointToUtf16be(code, utf16)) {
-          if ((code > 0xD7FF && code < 0xE000) || code > 0x10FFFF) {
-                tprintf("Dropping invalid codepoint %d\n", code);
-                continue;
-          }
-          if (code < 0x10000) {
-            snprintf(utf16, sizeof(utf16), "<%04X>", code);
-          } else {
-            int a = code - 0x010000;
-            int high_surrogate = (0x03FF & (a >> 10)) + 0xD800;
-            int low_surrogate = (0x03FF & a) + 0xDC00;
-            snprintf(utf16, sizeof(utf16), "<%04X%04X>",
-                     high_surrogate, low_surrogate);
-          }
             pdf_word += utf16;
             pdf_word_len++;
           }
         }
+      }
       delete []grapheme;
       res_it->Next(RIL_SYMBOL);
     } while (!res_it->Empty(RIL_BLOCK) && !res_it->IsAtBeginningOf(RIL_WORD));
@@ -465,9 +481,9 @@ char* TessPDFRenderer::GetPDFTextObjects(TessBaseAPI* api,
           kCharWidth * prec(100.0 * word_length / (fontsize * pdf_word_len));
       pdf_str.add_str_double("", h_stretch);
       pdf_str += " Tz";          // horizontal stretch
-      pdf_str += " [ ";
+      pdf_str += " [ <";
       pdf_str += pdf_word;       // UTF-16BE representation
-      pdf_str += " ] TJ";        // show the text
+      pdf_str += "> ] TJ";       // show the text
     }
     if (last_word_in_line) {
       pdf_str += " \n";
@@ -567,7 +583,8 @@ bool TessPDFRenderer::BeginDocumentHandler() {
                "<<\n"
                "  /Length %lu /Filter /FlateDecode\n"
                ">>\n"
-               "stream\n", (unsigned long)len);
+               "stream\n",
+               (unsigned long)len);
   if (n >= sizeof(buf)) {
     lept_free(comp);
     return false;
@@ -619,7 +636,6 @@ bool TessPDFRenderer::BeginDocumentHandler() {
   AppendPDFObject(buf);
 
   // FONT DESCRIPTOR
-  const int kCharHeight = 2;  // Effect: highlights are half height
   n = snprintf(buf, sizeof(buf),
                "7 0 obj\n"
                "<<\n"
@@ -635,10 +651,10 @@ bool TessPDFRenderer::BeginDocumentHandler() {
                "  /Type /FontDescriptor\n"
                ">>\n"
                "endobj\n",
-               1000 / kCharHeight,
+               1000,
-               1000 / kCharHeight,
+               1000,
                1000 / kCharWidth,
-               1000 / kCharHeight,
+               1000,
                8L      // Font data
                );
   if (n >= sizeof(buf)) return false;
@@ -703,11 +719,6 @@ bool TessPDFRenderer::imageToPDFObj(Pix *pix,
   L_COMP_DATA *cid = NULL;
   const int kJpegQuality = 85;
 
-  // TODO(jbreiden) Leptonica 1.71 doesn't correctly handle certain
-  // types of PNG files, especially if there are 2 samples per pixel.
-  // We can get rid of this logic after Leptonica 1.72 is released and
-  // has propagated everywhere. Bug discussion as follows.
-  // https://code.google.com/p/tesseract-ocr/issues/detail?id=1300
   int format, sad;
   findFileFormat(filename, &format);
   if (pixGetSpp(pix) == 4 && format == IFF_PNG) {
@@ -819,10 +830,6 @@ bool TessPDFRenderer::imageToPDFObj(Pix *pix,
   *pdf_object_size =
       b1_len + colorspace_len + b2_len + cid->nbytescomp + b3_len;
   *pdf_object = new char[*pdf_object_size];
-  if (!pdf_object) {
-    l_CIDataDestroy(&cid);
-    return false;
-  }
 
   char *p = *pdf_object;
   memcpy(p, b1, b1_len);
@@ -841,6 +848,7 @@ bool TessPDFRenderer::imageToPDFObj(Pix *pix,
 bool TessPDFRenderer::AddImageHandler(TessBaseAPI* api) {
   size_t n;
   char buf[kBasicBufSize];
+  char buf2[kBasicBufSize];
   Pix *pix = api->GetInputImage();
   char *filename = (char *)api->GetInputName();
   int ppi = api->GetSourceYResolution();
@@ -849,6 +857,9 @@ bool TessPDFRenderer::AddImageHandler(TessBaseAPI* api) {
   double width = pixGetWidth(pix) * 72.0 / ppi;
   double height = pixGetHeight(pix) * 72.0 / ppi;
 
+  snprintf(buf2, sizeof(buf2), "/XObject << /Im1 %ld 0 R >>\n", obj_ + 2);
+  const char *xobject = (textonly_) ? "" : buf2;
+
   // PAGE
   n = snprintf(buf, sizeof(buf),
                "%ld 0 obj\n"
@@ -859,7 +870,7 @@ bool TessPDFRenderer::AddImageHandler(TessBaseAPI* api) {
                "  /Contents %ld 0 R\n"
                "  /Resources\n"
                "  <<\n"
-               "    /XObject << /Im1 %ld 0 R >>\n"
+               "    %s"
                "    /ProcSet [ /PDF /Text /ImageB /ImageI /ImageC ]\n"
                "    /Font << /f-0-0 %ld 0 R >>\n"
                "  >>\n"
@@ -867,10 +878,9 @@ bool TessPDFRenderer::AddImageHandler(TessBaseAPI* api) {
                "endobj\n",
                obj_,
                2L,  // Pages object
-               width,
+               width, height,
-               height,
                obj_ + 1,  // Contents object
-               obj_ + 2,      // Image object
+               xobject,   // Image object
                3L);       // Type0 Font
   if (n >= sizeof(buf)) return false;
   pages_.push_back(obj_);
@@ -908,13 +918,15 @@ bool TessPDFRenderer::AddImageHandler(TessBaseAPI* api) {
   objsize += strlen(b2);
   AppendPDFObjectDIY(objsize);
 
-  char *pdf_object;
+  if (!textonly_) {
+    char *pdf_object = nullptr;
     if (!imageToPDFObj(pix, filename, obj_, &pdf_object, &objsize)) {
       return false;
     }
     AppendData(pdf_object, objsize);
     AppendPDFObjectDIY(objsize);
     delete[] pdf_object;
+  }
   return true;
 }
 
@@ -958,15 +970,27 @@ bool TessPDFRenderer::EndDocumentHandler() {
   offsets_.back() += pages_objsize;    // manipulation #2
 
   // INFO
+  STRING utf16_title = "FEFF";  // byte_order_marker
+  GenericVector<int> unicodes;
+  UNICHAR::UTF8ToUnicode(title(), &unicodes);
+  char utf16[kMaxBytesPerCodepoint];
+  for (int i = 0; i < unicodes.length(); i++) {
+    int code = unicodes[i];
+    if (CodepointToUtf16be(code, utf16)) {
+      utf16_title += utf16;
+    }
+  }
+
   char* datestr = l_getFormattedDate();
   n = snprintf(buf, sizeof(buf),
                "%ld 0 obj\n"
                "<<\n"
                "  /Producer (Tesseract %s)\n"
                "  /CreationDate (D:%s)\n"
-               "  /Title (%s)"
+               "  /Title <%s>\n"
                ">>\n"
-               "endobj\n", obj_, TESSERACT_VERSION_STR, datestr, title());
+               "endobj\n",
+               obj_, TESSERACT_VERSION_STR, datestr, utf16_title.c_str());
   lept_free(datestr);
   if (n >= sizeof(buf)) return false;
   AppendPDFObject(buf);

api/renderer.cpp

@@ -198,25 +198,25 @@ bool TessHOcrRenderer::AddImageHandler(TessBaseAPI* api) {
 /**********************************************************************
  * TSV Text Renderer interface implementation
  **********************************************************************/
-TessTsvRenderer::TessTsvRenderer(const char *outputbase)
+TessTsvRenderer::TessTsvRenderer(const char* outputbase)
     : TessResultRenderer(outputbase, "tsv") {
   font_info_ = false;
 }
 
-TessTsvRenderer::TessTsvRenderer(const char *outputbase, bool font_info)
+TessTsvRenderer::TessTsvRenderer(const char* outputbase, bool font_info)
     : TessResultRenderer(outputbase, "tsv") {
   font_info_ = font_info;
 }
 
 bool TessTsvRenderer::BeginDocumentHandler() {
   // Output TSV column headings
-  AppendString("level\tpage_num\tblock_num\tpar_num\tline_num\tword_num\tleft\ttop\twidth\theight\tconf\ttext\n");
+  AppendString(
+      "level\tpage_num\tblock_num\tpar_num\tline_num\tword_"
+      "num\tleft\ttop\twidth\theight\tconf\ttext\n");
   return true;
 }
 
-bool TessTsvRenderer::EndDocumentHandler() {
+bool TessTsvRenderer::EndDocumentHandler() { return true; }
-  return true;
-}
 
 bool TessTsvRenderer::AddImageHandler(TessBaseAPI* api) {
   char* tsv = api->GetTSVText(imagenum());
@@ -266,8 +266,7 @@ bool TessBoxTextRenderer::AddImageHandler(TessBaseAPI* api) {
  * Osd Text Renderer interface implementation
  **********************************************************************/
 TessOsdRenderer::TessOsdRenderer(const char* outputbase)
-    : TessResultRenderer(outputbase, "osd") {
+    : TessResultRenderer(outputbase, "osd") {}
-}
 
 bool TessOsdRenderer::AddImageHandler(TessBaseAPI* api) {
   char* osd = api->GetOsdText(imagenum());

api/renderer.h

@@ -15,8 +15,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_API_RENDERER_H__
+#ifndef TESSERACT_API_RENDERER_H_
-#define TESSERACT_API_RENDERER_H__
+#define TESSERACT_API_RENDERER_H_
 
 // To avoid collision with other typenames include the ABSOLUTE MINIMUM
 // complexity of includes here. Use forward declarations wherever possible
@@ -57,6 +57,7 @@ class TESS_API TessResultRenderer {
     /**
      * Starts a new document with the given title.
      * This clears the contents of the output data.
+     * Title should use UTF-8 encoding.
      */
     bool BeginDocument(const char* title);
 
@@ -77,7 +78,7 @@ class TESS_API TessResultRenderer {
     bool EndDocument();
 
     const char* file_extension() const { return file_extension_; }
-    const char* title() const { return title_; }
+    const char* title() const { return title_.c_str(); }
 
     /**
      * Returns the index of the last image given to AddImage
@@ -126,7 +127,7 @@ class TESS_API TessResultRenderer {
 
   private:
     const char* file_extension_;  // standard extension for generated output
-    const char* title_;           // title of document being renderered
+    STRING title_;                // title of document being renderered
     int imagenum_;                // index of last image added
 
     FILE* fout_;                  // output file pointer
@@ -153,12 +154,12 @@ class TESS_API TessHOcrRenderer : public TessResultRenderer {
   explicit TessHOcrRenderer(const char *outputbase, bool font_info);
   explicit TessHOcrRenderer(const char *outputbase);
 
-protected:
+ protected:
   virtual bool BeginDocumentHandler();
   virtual bool AddImageHandler(TessBaseAPI* api);
   virtual bool EndDocumentHandler();
 
-private:
+ private:
   bool font_info_;  // whether to print font information
 };
 
@@ -167,15 +168,15 @@ private:
  */
 class TESS_API TessTsvRenderer : public TessResultRenderer {
  public:
-  explicit TessTsvRenderer(const char *outputbase, bool font_info);
+  explicit TessTsvRenderer(const char* outputbase, bool font_info);
-  explicit TessTsvRenderer(const char *outputbase);
+  explicit TessTsvRenderer(const char* outputbase);
 
-protected:
+ protected:
   virtual bool BeginDocumentHandler();
   virtual bool AddImageHandler(TessBaseAPI* api);
   virtual bool EndDocumentHandler();
 
-private:
+ private:
   bool font_info_;              // whether to print font information
 };
 
@@ -186,30 +187,30 @@ class TESS_API TessPDFRenderer : public TessResultRenderer {
  public:
   // datadir is the location of the TESSDATA. We need it because
   // we load a custom PDF font from this location.
-  TessPDFRenderer(const char *outputbase, const char *datadir);
+  TessPDFRenderer(const char* outputbase, const char* datadir, bool textonly);
 
-protected:
+ protected:
   virtual bool BeginDocumentHandler();
   virtual bool AddImageHandler(TessBaseAPI* api);
   virtual bool EndDocumentHandler();
 
-private:
+ private:
   // We don't want to have every image in memory at once,
   // so we store some metadata as we go along producing
-  // PDFs one page at a time. At the end that metadata is
+  // PDFs one page at a time. At the end, that metadata is
   // used to make everything that isn't easily handled in a
   // streaming fashion.
   long int obj_;                     // counter for PDF objects
   GenericVector<long int> offsets_;  // offset of every PDF object in bytes
   GenericVector<long int> pages_;    // object number for every /Page object
   const char *datadir_;              // where to find the custom font
+  bool textonly_;                    // skip images if set
   // Bookkeeping only. DIY = Do It Yourself.
   void AppendPDFObjectDIY(size_t objectsize);
   // Bookkeeping + emit data.
   void AppendPDFObject(const char *data);
   // Create the /Contents object for an entire page.
-  static char* GetPDFTextObjects(TessBaseAPI* api,
+  char* GetPDFTextObjects(TessBaseAPI* api, double width, double height);
-                                 double width, double height);
   // Turn an image into a PDF object. Only transcode if we have to.
   static bool imageToPDFObj(Pix *pix, char *filename, long int objnum,
                           char **pdf_object, long int *pdf_object_size);
@@ -251,4 +252,4 @@ class TESS_API TessOsdRenderer : public TessResultRenderer {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_API_RENDERER_H__
+#endif  // TESSERACT_API_RENDERER_H_

api/tesseractmain.cpp

@@ -27,16 +27,16 @@
 #include "allheaders.h"
 #include "baseapi.h"
 #include "basedir.h"
-#include "renderer.h"
-#include "strngs.h"
-#include "tprintf.h"
 #include "openclwrapper.h"
 #include "osdetect.h"
+#include "renderer.h"
+#include "simddetect.h"
+#include "strngs.h"
+#include "tprintf.h"
 
 #if defined(HAVE_TIFFIO_H) && defined(_WIN32)
 
 #include <tiffio.h>
-#include <windows.h>
 
 static void Win32WarningHandler(const char* module, const char* fmt,
                                 va_list ap) {
@@ -51,7 +51,7 @@ static void Win32WarningHandler(const char* module, const char* fmt,
 #endif /* HAVE_TIFFIO_H &&  _WIN32 */
 
 void PrintVersionInfo() {
-    char *versionStrP;
+  char* versionStrP;
 
   printf("tesseract %s\n", tesseract::TessBaseAPI::Version());
 
@@ -64,33 +64,45 @@ void PrintVersionInfo() {
   lept_free(versionStrP);
 
 #ifdef USE_OPENCL
-    cl_platform_id platform;
+  cl_platform_id platform[4];
   cl_uint num_platforms;
-    cl_device_id devices[2];
-    cl_uint num_devices;
-    char info[256];
-    int i;
 
   printf(" OpenCL info:\n");
-    clGetPlatformIDs(1, &platform, &num_platforms);
+  if (clGetPlatformIDs(4, platform, &num_platforms) == CL_SUCCESS) {
-    printf("  Found %d platforms.\n", num_platforms);
+    printf("  Found %u platform(s).\n", num_platforms);
-    clGetPlatformInfo(platform, CL_PLATFORM_NAME, 256, info, 0);
+    for (unsigned n = 0; n < num_platforms; n++) {
-    printf("  Platform name: %s.\n", info);
+      char info[256];
-    clGetPlatformInfo(platform, CL_PLATFORM_VERSION, 256, info, 0);
+      if (clGetPlatformInfo(platform[n], CL_PLATFORM_NAME, 256, info, 0) ==
+          CL_SUCCESS) {
+        printf("  Platform %u name: %s.\n", n + 1, info);
+      }
+      if (clGetPlatformInfo(platform[n], CL_PLATFORM_VERSION, 256, info, 0) ==
+          CL_SUCCESS) {
         printf("  Version: %s.\n", info);
-    clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 2, devices, &num_devices);
+      }
-    printf("  Found %d devices.\n", num_devices);
+      cl_device_id devices[2];
-    for (i = 0; i < num_devices; ++i) {
+      cl_uint num_devices;
-      clGetDeviceInfo(devices[i], CL_DEVICE_NAME, 256, info, 0);
+      if (clGetDeviceIDs(platform[n], CL_DEVICE_TYPE_ALL, 2, devices,
-      printf("    Device %d name: %s.\n", i+1, info);
+                         &num_devices) == CL_SUCCESS) {
+        printf("  Found %u device(s).\n", num_devices);
+        for (unsigned i = 0; i < num_devices; ++i) {
+          if (clGetDeviceInfo(devices[i], CL_DEVICE_NAME, 256, info, 0) ==
+              CL_SUCCESS) {
+            printf("    Device %u name: %s.\n", i + 1, info);
+          }
+        }
+      }
+    }
     }
 #endif
+    if (SIMDDetect::IsAVXAvailable()) printf(" Found AVX\n");
+    if (SIMDDetect::IsSSEAvailable()) printf(" Found SSE\n");
 }
 
 void PrintUsage(const char* program) {
   printf(
       "Usage:\n"
-      "  %s --help | --help-psm | --version\n"
+      "  %s --help | --help-psm | --help-oem | --version\n"
       "  %s --list-langs [--tessdata-dir PATH]\n"
       "  %s --print-parameters [options...] [configfile...]\n"
       "  %s imagename|stdin outputbase|stdout [options...] [configfile...]\n",
@@ -111,16 +123,22 @@ void PrintHelpForPSM() {
       "  8    Treat the image as a single word.\n"
       "  9    Treat the image as a single word in a circle.\n"
       " 10    Treat the image as a single character.\n"
-
-        //TODO: Consider publishing these modes.
-        #if 0
       " 11    Sparse text. Find as much text as possible in no"
       " particular order.\n"
       " 12    Sparse text with OSD.\n"
       " 13    Raw line. Treat the image as a single text line,\n"
-          "\t\t\tbypassing hacks that are Tesseract-specific.\n"
+      "\t\t\tbypassing hacks that are Tesseract-specific.\n";
-        #endif
+
-        ;
+  printf("%s", msg);
+}
+
+void PrintHelpForOEM() {
+  const char* msg =
+      "OCR Engine modes:\n"
+      "  0    Original Tesseract only.\n"
+      "  1    Neural nets LSTM only.\n"
+      "  2    Tesseract + LSTM.\n"
+      "  3    Default, based on what is available.\n";
 
   printf("%s", msg);
 }
@@ -136,32 +154,34 @@ void PrintHelpMessage(const char* program) {
       "  -l LANG[+LANG]        Specify language(s) used for OCR.\n"
       "  -c VAR=VALUE          Set value for config variables.\n"
       "                        Multiple -c arguments are allowed.\n"
-      "  -psm NUM              Specify page segmentation mode.\n"
+      "  --psm NUM             Specify page segmentation mode.\n"
-      "NOTE: These options must occur before any configfile.\n"
+      "  --oem NUM             Specify OCR Engine mode.\n"
-     ;
+      "NOTE: These options must occur before any configfile.\n";
 
   printf("\n%s\n", ocr_options);
   PrintHelpForPSM();
+  PrintHelpForOEM();
 
-  const char *single_options =
+  const char* single_options =
       "Single options:\n"
       "  -h, --help            Show this help message.\n"
       "  --help-psm            Show page segmentation modes.\n"
+      "  --help-oem            Show OCR Engine modes.\n"
       "  -v, --version         Show version information.\n"
       "  --list-langs          List available languages for tesseract engine.\n"
-      "  --print-parameters    Print tesseract parameters to stdout.\n"
+      "  --print-parameters    Print tesseract parameters.\n";
-      ;
 
   printf("\n%s", single_options);
 }
 
-void SetVariablesFromCLArgs(tesseract::TessBaseAPI* api, int argc, char** argv) {
+void SetVariablesFromCLArgs(tesseract::TessBaseAPI* api, int argc,
+                            char** argv) {
   char opt1[256], opt2[255];
   for (int i = 0; i < argc; i++) {
     if (strcmp(argv[i], "-c") == 0 && i + 1 < argc) {
       strncpy(opt1, argv[i + 1], 255);
       opt1[255] = '\0';
-      char *p = strchr(opt1, '=');
+      char* p = strchr(opt1, '=');
       if (!p) {
         fprintf(stderr, "Missing = in configvar assignment\n");
         exit(1);
@@ -215,25 +235,20 @@ void FixPageSegMode(tesseract::TessBaseAPI* api,
 }
 
 // NOTE: arg_i is used here to avoid ugly *i so many times in this function
-void ParseArgs(const int argc, char** argv,
+void ParseArgs(const int argc, char** argv, const char** lang,
-                  const char** lang,
+               const char** image, const char** outputbase,
-                  const char** image,
+               const char** datapath, bool* list_langs, bool* print_parameters,
-                  const char** outputbase,
-                  const char** datapath,
-                  bool* list_langs,
-                  bool* print_parameters,
                GenericVector<STRING>* vars_vec,
-                  GenericVector<STRING>* vars_values,
+               GenericVector<STRING>* vars_values, int* arg_i,
-                  int* arg_i,
+               tesseract::PageSegMode* pagesegmode,
-                  tesseract::PageSegMode* pagesegmode) {
+               tesseract::OcrEngineMode* enginemode) {
   if (argc == 1) {
     PrintHelpMessage(argv[0]);
     exit(0);
   }
 
   if (argc == 2) {
-    if ((strcmp(argv[1], "-h") == 0) ||
+    if ((strcmp(argv[1], "-h") == 0) || (strcmp(argv[1], "--help") == 0)) {
-         (strcmp(argv[1], "--help") == 0)) {
       PrintHelpMessage(argv[0]);
       exit(0);
     }
@@ -241,8 +256,11 @@ void ParseArgs(const int argc, char** argv,
       PrintHelpForPSM();
       exit(0);
     }
-    if ((strcmp(argv[1], "-v") == 0) ||
+    if ((strcmp(argv[1], "--help-oem") == 0)) {
-         (strcmp(argv[1], "--version") == 0)) {
+      PrintHelpForOEM();
+      exit(0);
+    }
+    if ((strcmp(argv[1], "-v") == 0) || (strcmp(argv[1], "--version") == 0)) {
       PrintVersionInfo();
       exit(0);
     }
@@ -269,8 +287,16 @@ void ParseArgs(const int argc, char** argv,
       noocr = true;
       *list_langs = true;
     } else if (strcmp(argv[i], "-psm") == 0 && i + 1 < argc) {
+      // The parameter -psm is deprecated and was replaced by --psm.
+      // It is still supported for compatibility reasons.
       *pagesegmode = static_cast<tesseract::PageSegMode>(atoi(argv[i + 1]));
       ++i;
+    } else if (strcmp(argv[i], "--psm") == 0 && i + 1 < argc) {
+      *pagesegmode = static_cast<tesseract::PageSegMode>(atoi(argv[i + 1]));
+      ++i;
+    } else if (strcmp(argv[i], "--oem") == 0 && i + 1 < argc) {
+      *enginemode = static_cast<tesseract::OcrEngineMode>(atoi(argv[i + 1]));
+      ++i;
     } else if (strcmp(argv[i], "--print-parameters") == 0) {
       noocr = true;
       *print_parameters = true;
@@ -298,10 +324,10 @@ void ParseArgs(const int argc, char** argv,
   }
 }
 
-void PreloadRenderers(tesseract::TessBaseAPI* api,
+void PreloadRenderers(
+    tesseract::TessBaseAPI* api,
     tesseract::PointerVector<tesseract::TessResultRenderer>* renderers,
-          tesseract::PageSegMode pagesegmode,
+    tesseract::PageSegMode pagesegmode, const char* outputbase) {
-          const char* outputbase) {
   if (pagesegmode == tesseract::PSM_OSD_ONLY) {
     renderers->push_back(new tesseract::TessOsdRenderer(outputbase));
   } else {
@@ -324,8 +350,10 @@ void PreloadRenderers(tesseract::TessBaseAPI* api,
 
     api->GetBoolVariable("tessedit_create_pdf", &b);
     if (b) {
-      renderers->push_back(new tesseract::TessPDFRenderer(outputbase,
+      bool textonly;
-                                                        api->GetDatapath()));
+      api->GetBoolVariable("textonly_pdf", &textonly);
+      renderers->push_back(new tesseract::TessPDFRenderer(
+          outputbase, api->GetDatapath(), textonly));
     }
 
     api->GetBoolVariable("tessedit_write_unlv", &b);
@@ -358,26 +386,36 @@ void PreloadRenderers(tesseract::TessBaseAPI* api,
  *  main()
  *
  **********************************************************************/
-int main(int argc, char **argv) {
+
+int main(int argc, char** argv) {
   const char* lang = "eng";
   const char* image = NULL;
   const char* outputbase = NULL;
   const char* datapath = NULL;
   bool list_langs = false;
   bool print_parameters = false;
-  GenericVector<STRING> vars_vec, vars_values;
   int arg_i = 1;
   tesseract::PageSegMode pagesegmode = tesseract::PSM_AUTO;
+  tesseract::OcrEngineMode enginemode = tesseract::OEM_DEFAULT;
+  /* main() calls functions like ParseArgs which call exit().
+   * This results in memory leaks if vars_vec and vars_values are
+   * declared as auto variables (destructor is not called then). */
+  static GenericVector<STRING> vars_vec;
+  static GenericVector<STRING> vars_values;
+
+#if !defined(DEBUG)
+  // Disable debugging and informational messages from Leptonica.
+  setMsgSeverity(L_SEVERITY_ERROR);
+#endif
 
 #if defined(HAVE_TIFFIO_H) && defined(_WIN32)
   /* Show libtiff warnings on console (not in GUI). */
   TIFFSetWarningHandler(Win32WarningHandler);
 #endif /* HAVE_TIFFIO_H &&  _WIN32 */
 
-  ParseArgs(argc, argv,
+  ParseArgs(argc, argv, &lang, &image, &outputbase, &datapath, &list_langs,
-          &lang, &image, &outputbase, &datapath,
+            &print_parameters, &vars_vec, &vars_values, &arg_i, &pagesegmode,
-          &list_langs, &print_parameters,
+            &enginemode);
-          &vars_vec, &vars_values, &arg_i, &pagesegmode);
 
   bool banner = false;
   if (outputbase != NULL && strcmp(outputbase, "-") &&
@@ -390,8 +428,8 @@ int main(int argc, char **argv) {
 
   api.SetOutputName(outputbase);
 
-  int init_failed = api.Init(datapath, lang, tesseract::OEM_DEFAULT,
+  int init_failed = api.Init(datapath, lang, enginemode, &(argv[arg_i]),
-                &(argv[arg_i]), argc - arg_i, &vars_vec, &vars_values, false);
+                             argc - arg_i, &vars_vec, &vars_values, false);
   if (init_failed) {
     fprintf(stderr, "Could not initialize tesseract.\n");
     exit(1);
@@ -433,7 +471,8 @@ int main(int argc, char **argv) {
     tesseract::PageIterator* it = api.AnalyseLayout();
     if (it) {
       it->Orientation(&orientation, &direction, &order, &deskew_angle);
-      tprintf("Orientation: %d\nWritingDirection: %d\nTextlineOrder: %d\n" \
+      tprintf(
+          "Orientation: %d\nWritingDirection: %d\nTextlineOrder: %d\n"
           "Deskew angle: %.4f\n",
           orientation, direction, order, deskew_angle);
     } else {
@@ -456,8 +495,6 @@ int main(int argc, char **argv) {
 
   tesseract::PointerVector<tesseract::TessResultRenderer> renderers;
 
-
-
   if (in_training_mode) {
     renderers.push_back(NULL);
   } else {

appveyor.yml

@@ -1,4 +1,4 @@
-os: Visual Studio 2015
+os: Visual Studio 2017
 
 platform:
   - Win32
@@ -7,19 +7,39 @@ platform:
 configuration:
   - Release
   
+# for curl
+install:
+  - set PATH=C:\Program Files\Git\mingw64\bin;%PATH%
+
 before_build:
-  - if %platform%==Win32 set generator=Visual Studio 14
+  - if %platform%==Win32 set generator=Visual Studio 15 2017
-  - if %platform%==Win64 set generator=Visual Studio 14 Win64
+  - if %platform%==Win64 set generator=Visual Studio 15 2017 Win64
   - if %platform%==Win32 set vcplatform=Win32
   - if %platform%==Win64 set vcplatform=x64
 
-  - curl -fsS -o cppan.zip https://cppan.org/client/cppan-master-Windows-client.zip
+  - curl -fsS -L -o cppan.zip https://cppan.org/client/cppan-master-Windows-client.zip
   - 7z x cppan.zip
   - set PATH=%PATH%;%cd%
 
+  - cppan # dummy run to create %USERPROFILE%\.cppan\cppan.yml
+  - ps: 'Add-Content $env:USERPROFILE\.cppan\cppan.yml "`n`nbuild_warning_level: 0`n"'
+  - ps: 'Add-Content $env:USERPROFILE\.cppan\cppan.yml "`n`nbuild_system_verbose: false`n"'
+  - ps: 'Add-Content $env:USERPROFILE\.cppan\cppan.yml "`n`nvar_check_jobs: 1`n"'
+  
 build_script:
-  - cppan
   - mkdir build
+  - mkdir build\bin
+  - mkdir build\bin\Release
   - cd build
-  - cmake .. -G "%generator%" -DSTATIC=1
+  #- cmd: 'echo local_settings: > cppan.yml'
-  - msbuild tesseract.sln /p:Platform=%vcplatform% /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll"
+  #- cmd: 'echo     generator: %generator% >> cppan.yml'
+  #- cmd: 'echo     use_shared_libs: true >> cppan.yml'
+  #- cppan --build ..
+  - cmake .. -G "%generator%" -DBUILD_TRAINING_TOOLS=Off -DAPPVEYOR=1
+  - cmake --build . --config Release > bin\Release\log.txt 2>&1
+
+artifacts:
+  - path: build\bin\Release
+  #- path: build
+    name: tesseract-$(APPVEYOR_BUILD_VERSION)
+

arch/Makefile.am

@@ -0,0 +1,38 @@
+AM_CPPFLAGS += -I$(top_srcdir)/ccutil -I$(top_srcdir)/viewer
+AUTOMAKE_OPTIONS = subdir-objects
+SUBDIRS =
+AM_CXXFLAGS =
+
+if VISIBILITY
+AM_CXXFLAGS += -fvisibility=hidden -fvisibility-inlines-hidden
+AM_CPPFLAGS += -DTESS_EXPORTS
+endif
+
+include_HEADERS = dotproductavx.h dotproductsse.h simddetect.h
+
+noinst_HEADERS =
+
+if !USING_MULTIPLELIBS
+noinst_LTLIBRARIES = libtesseract_avx.la libtesseract_sse.la
+noinst_LTLIBRARIES += libtesseract_arch.la
+else
+lib_LTLIBRARIES = libtesseract_avx.la libtesseract_sse.la
+lib_LTLIBRARIES += libtesseract_arch.la
+libtesseract_arch_la_LDFLAGS = -version-info $(GENERIC_LIBRARY_VERSION)
+libtesseract_avx_la_LDFLAGS = -version-info $(GENERIC_LIBRARY_VERSION)
+libtesseract_sse_la_LDFLAGS = -version-info $(GENERIC_LIBRARY_VERSION)
+endif
+
+if AVX_OPT
+libtesseract_avx_la_CXXFLAGS = -mavx
+endif
+if SSE41_OPT
+libtesseract_sse_la_CXXFLAGS = -msse4.1
+endif
+
+libtesseract_arch_la_SOURCES = simddetect.cpp
+
+libtesseract_avx_la_SOURCES = dotproductavx.cpp
+
+libtesseract_sse_la_SOURCES = dotproductsse.cpp
+

arch/dotproductavx.cpp

@@ -0,0 +1,112 @@
+///////////////////////////////////////////////////////////////////////
+// File:        dotproductavx.cpp
+// Description: Architecture-specific dot-product function.
+// Author:      Ray Smith
+// Created:     Wed Jul 22 10:48:05 PDT 2015
+//
+// (C) Copyright 2015, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#if !defined(__AVX__)
+// Implementation for non-avx archs.
+
+#include "dotproductavx.h"
+#include <stdio.h>
+#include <stdlib.h>
+
+namespace tesseract {
+double DotProductAVX(const double* u, const double* v, int n) {
+  fprintf(stderr, "DotProductAVX can't be used on Android\n");
+  abort();
+}
+}  // namespace tesseract
+
+#else  // !defined(__AVX__)
+// Implementation for avx capable archs.
+#include <immintrin.h>
+#include <stdint.h>
+#include "dotproductavx.h"
+#include "host.h"
+
+namespace tesseract {
+
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel AVX intrinsics to access the SIMD instruction set.
+double DotProductAVX(const double* u, const double* v, int n) {
+  int max_offset = n - 4;
+  int offset = 0;
+  // Accumulate a set of 4 sums in sum, by loading pairs of 4 values from u and
+  // v, and multiplying them together in parallel.
+  __m256d sum = _mm256_setzero_pd();
+  if (offset <= max_offset) {
+    offset = 4;
+    // Aligned load is reputedly faster but requires 32 byte aligned input.
+    if ((reinterpret_cast<const uintptr_t>(u) & 31) == 0 &&
+        (reinterpret_cast<const uintptr_t>(v) & 31) == 0) {
+      // Use aligned load.
+      __m256d floats1 = _mm256_load_pd(u);
+      __m256d floats2 = _mm256_load_pd(v);
+      // Multiply.
+      sum = _mm256_mul_pd(floats1, floats2);
+      while (offset <= max_offset) {
+        floats1 = _mm256_load_pd(u + offset);
+        floats2 = _mm256_load_pd(v + offset);
+        offset += 4;
+        __m256d product = _mm256_mul_pd(floats1, floats2);
+        sum = _mm256_add_pd(sum, product);
+      }
+    } else {
+      // Use unaligned load.
+      __m256d floats1 = _mm256_loadu_pd(u);
+      __m256d floats2 = _mm256_loadu_pd(v);
+      // Multiply.
+      sum = _mm256_mul_pd(floats1, floats2);
+      while (offset <= max_offset) {
+        floats1 = _mm256_loadu_pd(u + offset);
+        floats2 = _mm256_loadu_pd(v + offset);
+        offset += 4;
+        __m256d product = _mm256_mul_pd(floats1, floats2);
+        sum = _mm256_add_pd(sum, product);
+      }
+    }
+  }
+  // Add the 4 product sums together horizontally. Not so easy as with sse, as
+  // there is no add across the upper/lower 128 bit boundary, so permute to
+  // move the upper 128 bits to lower in another register.
+  __m256d sum2 = _mm256_permute2f128_pd(sum, sum, 1);
+  sum = _mm256_hadd_pd(sum, sum2);
+  sum = _mm256_hadd_pd(sum, sum);
+  double result;
+  // _mm256_extract_f64 doesn't exist, but resist the temptation to use an sse
+  // instruction, as that introduces a 70 cycle delay. All this casting is to
+  // fool the intrinsics into thinking we are extracting the bottom int64.
+  auto cast_sum = _mm256_castpd_si256(sum);
+  *(reinterpret_cast<inT64*>(&result)) =
+#if defined(_WIN32) || defined(__i386__)
+      // This is a very simple workaround that is activated
+      // for all platforms that do not have _mm256_extract_epi64.
+      // _mm256_extract_epi64(X, Y) == ((uint64_t*)&X)[Y]
+      ((uint64_t*)&cast_sum)[0]
+#else
+      _mm256_extract_epi64(cast_sum, 0)
+#endif
+      ;
+  while (offset < n) {
+    result += u[offset] * v[offset];
+    ++offset;
+  }
+  return result;
+}
+
+}  // namespace tesseract.
+
+#endif  // ANDROID_BUILD

arch/dotproductavx.h

@@ -0,0 +1,30 @@
+///////////////////////////////////////////////////////////////////////
+// File:        dotproductavx.h
+// Description: Architecture-specific dot-product function.
+// Author:      Ray Smith
+// Created:     Wed Jul 22 10:51:05 PDT 2015
+//
+// (C) Copyright 2015, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#ifndef TESSERACT_ARCH_DOTPRODUCTAVX_H_
+#define TESSERACT_ARCH_DOTPRODUCTAVX_H_
+
+namespace tesseract {
+
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel AVX intrinsics to access the SIMD instruction set.
+double DotProductAVX(const double* u, const double* v, int n);
+
+}  // namespace tesseract.
+
+#endif  // TESSERACT_ARCH_DOTPRODUCTAVX_H_

arch/dotproductsse.cpp

@@ -0,0 +1,141 @@
+///////////////////////////////////////////////////////////////////////
+// File:        dotproductsse.cpp
+// Description: Architecture-specific dot-product function.
+// Author:      Ray Smith
+// Created:     Wed Jul 22 10:57:45 PDT 2015
+//
+// (C) Copyright 2015, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#if !defined(__SSE4_1__)
+// This code can't compile with "-msse4.1", so use dummy stubs.
+
+#include "dotproductsse.h"
+#include <stdio.h>
+#include <stdlib.h>
+
+namespace tesseract {
+double DotProductSSE(const double* u, const double* v, int n) {
+  fprintf(stderr, "DotProductSSE can't be used on Android\n");
+  abort();
+}
+inT32 IntDotProductSSE(const inT8* u, const inT8* v, int n) {
+  fprintf(stderr, "IntDotProductSSE can't be used on Android\n");
+  abort();
+}
+}  // namespace tesseract
+
+#else  // !defined(__SSE4_1__)
+// Non-Android code here
+
+#include <emmintrin.h>
+#include <smmintrin.h>
+#include <stdint.h>
+#include "dotproductsse.h"
+#include "host.h"
+
+namespace tesseract {
+
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel SSE intrinsics to access the SIMD instruction set.
+double DotProductSSE(const double* u, const double* v, int n) {
+  int max_offset = n - 2;
+  int offset = 0;
+  // Accumulate a set of 2 sums in sum, by loading pairs of 2 values from u and
+  // v, and multiplying them together in parallel.
+  __m128d sum = _mm_setzero_pd();
+  if (offset <= max_offset) {
+    offset = 2;
+    // Aligned load is reputedly faster but requires 16 byte aligned input.
+    if ((reinterpret_cast<const uintptr_t>(u) & 15) == 0 &&
+        (reinterpret_cast<const uintptr_t>(v) & 15) == 0) {
+      // Use aligned load.
+      sum = _mm_load_pd(u);
+      __m128d floats2 = _mm_load_pd(v);
+      // Multiply.
+      sum = _mm_mul_pd(sum, floats2);
+      while (offset <= max_offset) {
+        __m128d floats1 = _mm_load_pd(u + offset);
+        floats2 = _mm_load_pd(v + offset);
+        offset += 2;
+        floats1 = _mm_mul_pd(floats1, floats2);
+        sum = _mm_add_pd(sum, floats1);
+      }
+    } else {
+      // Use unaligned load.
+      sum = _mm_loadu_pd(u);
+      __m128d floats2 = _mm_loadu_pd(v);
+      // Multiply.
+      sum = _mm_mul_pd(sum, floats2);
+      while (offset <= max_offset) {
+        __m128d floats1 = _mm_loadu_pd(u + offset);
+        floats2 = _mm_loadu_pd(v + offset);
+        offset += 2;
+        floats1 = _mm_mul_pd(floats1, floats2);
+        sum = _mm_add_pd(sum, floats1);
+      }
+    }
+  }
+  // Add the 2 sums in sum horizontally.
+  sum = _mm_hadd_pd(sum, sum);
+  // Extract the low result.
+  double result = _mm_cvtsd_f64(sum);
+  // Add on any left-over products.
+  while (offset < n) {
+    result += u[offset] * v[offset];
+    ++offset;
+  }
+  return result;
+}
+
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel SSE intrinsics to access the SIMD instruction set.
+inT32 IntDotProductSSE(const inT8* u, const inT8* v, int n) {
+  int max_offset = n - 8;
+  int offset = 0;
+  // Accumulate a set of 4 32-bit sums in sum, by loading 8 pairs of 8-bit
+  // values, extending to 16 bit, multiplying to make 32 bit results.
+  __m128i sum = _mm_setzero_si128();
+  if (offset <= max_offset) {
+    offset = 8;
+    __m128i packed1 = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(u));
+    __m128i packed2 = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(v));
+    sum = _mm_cvtepi8_epi16(packed1);
+    packed2 = _mm_cvtepi8_epi16(packed2);
+    // The magic _mm_add_epi16 is perfect here. It multiplies 8 pairs of 16 bit
+    // ints to make 32 bit results, which are then horizontally added in pairs
+    // to make 4 32 bit results that still fit in a 128 bit register.
+    sum = _mm_madd_epi16(sum, packed2);
+    while (offset <= max_offset) {
+      packed1 = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(u + offset));
+      packed2 = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(v + offset));
+      offset += 8;
+      packed1 = _mm_cvtepi8_epi16(packed1);
+      packed2 = _mm_cvtepi8_epi16(packed2);
+      packed1 = _mm_madd_epi16(packed1, packed2);
+      sum = _mm_add_epi32(sum, packed1);
+    }
+  }
+  // Sum the 4 packed 32 bit sums and extract the low result.
+  sum = _mm_hadd_epi32(sum, sum);
+  sum = _mm_hadd_epi32(sum, sum);
+  inT32 result = _mm_cvtsi128_si32(sum);
+  while (offset < n) {
+    result += u[offset] * v[offset];
+    ++offset;
+  }
+  return result;
+}
+
+}  // namespace tesseract.
+
+#endif  // ANDROID_BUILD

arch/dotproductsse.h

@@ -0,0 +1,35 @@
+///////////////////////////////////////////////////////////////////////
+// File:        dotproductsse.h
+// Description: Architecture-specific dot-product function.
+// Author:      Ray Smith
+// Created:     Wed Jul 22 10:57:05 PDT 2015
+//
+// (C) Copyright 2015, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#ifndef TESSERACT_ARCH_DOTPRODUCTSSE_H_
+#define TESSERACT_ARCH_DOTPRODUCTSSE_H_
+
+#include "host.h"
+
+namespace tesseract {
+
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel SSE intrinsics to access the SIMD instruction set.
+double DotProductSSE(const double* u, const double* v, int n);
+// Computes and returns the dot product of the n-vectors u and v.
+// Uses Intel SSE intrinsics to access the SIMD instruction set.
+inT32 IntDotProductSSE(const inT8* u, const inT8* v, int n);
+
+}  // namespace tesseract.
+
+#endif  // TESSERACT_ARCH_DOTPRODUCTSSE_H_

arch/simddetect.cpp

@@ -0,0 +1,68 @@
+///////////////////////////////////////////////////////////////////////
+// File:        simddetect.h
+// Description: Architecture detector.
+// Author:      Stefan Weil (based on code from Ray Smith)
+//
+// (C) Copyright 2014, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#include "simddetect.h"
+#include "tprintf.h"
+
+#undef X86_BUILD
+#if defined(__x86_64__) || defined(__i386__) || defined(_WIN32)
+#if !defined(ANDROID_BUILD)
+#define X86_BUILD 1
+#endif  // !ANDROID_BUILD
+#endif  // x86 target
+
+#if defined(X86_BUILD)
+#if defined(__GNUC__)
+#include <cpuid.h>
+#elif defined(_WIN32)
+#include <intrin.h>
+#endif
+#endif
+
+SIMDDetect SIMDDetect::detector;
+
+// If true, then AVX has been detected.
+bool SIMDDetect::avx_available_;
+// If true, then SSe4.1 has been detected.
+bool SIMDDetect::sse_available_;
+
+// Constructor.
+// Tests the architecture in a system-dependent way to detect AVX, SSE and
+// any other available SIMD equipment.
+// __GNUC__ is also defined by compilers that include GNU extensions such as
+// clang.
+SIMDDetect::SIMDDetect() {
+#if defined(X86_BUILD)
+#if defined(__GNUC__)
+  unsigned int eax, ebx, ecx, edx;
+  if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) != 0) {
+    sse_available_ = (ecx & 0x00080000) != 0;
+    avx_available_ = (ecx & 0x10000000) != 0;
+  }
+#elif defined(_WIN32)
+  int cpuInfo[4];
+  __cpuid(cpuInfo, 0);
+  if (cpuInfo[0] >= 1) {
+    __cpuid(cpuInfo, 1);
+    sse_available_ = (cpuInfo[2] & 0x00080000) != 0;
+    avx_available_ = (cpuInfo[2] & 0x10000000) != 0;
+  }
+#else
+#error "I don't know how to test for SIMD with this compiler"
+#endif
+#endif  // X86_BUILD
+}

arch/simddetect.h

@@ -0,0 +1,41 @@
+///////////////////////////////////////////////////////////////////////
+// File:        simddetect.h
+// Description: Architecture detector.
+// Author:      Stefan Weil (based on code from Ray Smith)
+//
+// (C) Copyright 2014, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#include "platform.h"
+
+// Architecture detector. Add code here to detect any other architectures for
+// SIMD-based faster dot product functions. Intended to be a single static
+// object, but it does no real harm to have more than one.
+class SIMDDetect {
+ public:
+  // Returns true if AVX is available on this system.
+  static inline bool IsAVXAvailable() { return detector.avx_available_; }
+  // Returns true if SSE4.1 is available on this system.
+  static inline bool IsSSEAvailable() { return detector.sse_available_; }
+
+ private:
+  // Constructor, must set all static member variables.
+  SIMDDetect();
+
+ private:
+  // Singleton.
+  static SIMDDetect detector;
+  // If true, then AVX has been detected.
+  static TESS_API bool avx_available_;
+  // If true, then SSe4.1 has been detected.
+  static TESS_API bool sse_available_;
+};

autogen.sh

@@ -1,4 +1,13 @@
 #!/bin/sh
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+# http://www.apache.org/licenses/LICENSE-2.0
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 
 # This is a simple script which is meant to help developers
 # better deal with the GNU autotools, specifically:
@@ -37,7 +46,20 @@ if [ "$1" = "clean" ]; then
     find . -iname "Makefile.in" -type f -exec rm '{}' +
 fi
 
-# create m4 directory if it not exists
+# Prevent any errors that might result from failing to properly invoke 
+# `libtoolize` or `glibtoolize,` whichever is present on your system, 
+# from occurring by testing for its existence and capturing the absolute path to 
+# its location for caching purposes prior to using it later on in 'Step 2:'  
+if command -v libtoolize >/dev/null 2>&1; then
+  LIBTOOLIZE="$(command -v libtoolize)"
+elif command -v glibtoolize >/dev/null 2>&1; then
+  LIBTOOLIZE="$(command -v glibtoolize)"
+else
+  echo "Unable to find a valid copy of libtoolize or glibtoolize in your PATH!"
+  bail_out
+fi
+
+# create m4 directory if it does not exist
 if [ ! -d m4 ];  then
     mkdir m4
 fi
@@ -61,9 +83,9 @@ aclocal -I config || bail_out
 
 # --- Step 2:
 
-echo "Running libtoolize"
+echo "Running $LIBTOOLIZE"
-libtoolize -f -c || glibtoolize -f -c || bail_out
+$LIBTOOLIZE -f -c || bail_out
-libtoolize --automake || glibtoolize --automake || bail_out
+$LIBTOOLIZE --automake || bail_out
 
 # --- Step 3: Generate config.h.in from:
 #             . configure.ac (look for AM_CONFIG_HEADER tag or AC_CONFIG_HEADER tag)

ccmain/Makefile.am

@@ -1,12 +1,14 @@
 AM_CPPFLAGS += \
     -DUSE_STD_NAMESPACE \
     -I$(top_srcdir)/ccutil -I$(top_srcdir)/ccstruct \
+    -I$(top_srcdir)/arch -I$(top_srcdir)/lstm \
     -I$(top_srcdir)/viewer \
     -I$(top_srcdir)/classify  -I$(top_srcdir)/dict \
     -I$(top_srcdir)/wordrec -I$(top_srcdir)/cutil \
     -I$(top_srcdir)/textord -I$(top_srcdir)/opencl
 
 AM_CPPFLAGS += $(OPENCL_CPPFLAGS)
+AM_CPPFLAGS += $(OPENMP_CXXFLAGS)
 
 if VISIBILITY
 AM_CPPFLAGS += -DTESS_EXPORTS \
@@ -33,18 +35,18 @@ libtesseract_main_la_LIBADD = \
     ../ccstruct/libtesseract_ccstruct.la \
     ../viewer/libtesseract_viewer.la \
     ../dict/libtesseract_dict.la \
+    ../arch/libtesseract_avx.la \
+    ../arch/libtesseract_sse.la \
+    ../lstm/libtesseract_lstm.la \
     ../classify/libtesseract_classify.la \
     ../cutil/libtesseract_cutil.la \
     ../opencl/libtesseract_opencl.la
-    if !NO_CUBE_BUILD
-        libtesseract_main_la_LIBADD += ../cube/libtesseract_cube.la
-    endif
 endif
 
 libtesseract_main_la_SOURCES = \
     adaptions.cpp applybox.cpp control.cpp  \
     docqual.cpp equationdetect.cpp fixspace.cpp fixxht.cpp \
-    ltrresultiterator.cpp \
+    linerec.cpp ltrresultiterator.cpp \
     osdetect.cpp output.cpp pageiterator.cpp pagesegmain.cpp \
     pagewalk.cpp par_control.cpp paragraphs.cpp paramsd.cpp pgedit.cpp recogtraining.cpp \
     reject.cpp resultiterator.cpp superscript.cpp \
@@ -52,12 +54,3 @@ libtesseract_main_la_SOURCES = \
     tfacepp.cpp thresholder.cpp \
     werdit.cpp
 
-if !NO_CUBE_BUILD
-AM_CPPFLAGS += \
-    -I$(top_srcdir)/neural_networks/runtime -I$(top_srcdir)/cube
-noinst_HEADERS += \
-    cube_reco_context.h cubeclassifier.h tesseract_cube_combiner.h
-libtesseract_main_la_SOURCES += \
-    cube_control.cpp cube_reco_context.cpp cubeclassifier.cpp \
-    tesseract_cube_combiner.cpp
-endif

ccmain/control.cpp

@@ -31,21 +31,22 @@
 #include <errno.h>
 #endif
 #include <ctype.h>
-#include "ocrclass.h"
+#include "callcpp.h"
-#include "werdit.h"
+#include "control.h"
+#include "docqual.h"
 #include "drawfx.h"
-#include "tessbox.h"
+#include "fixspace.h"
-#include "tessvars.h"
+#include "globals.h"
+#include "lstmrecognizer.h"
+#include "ocrclass.h"
+#include "output.h"
 #include "pgedit.h"
 #include "reject.h"
-#include "fixspace.h"
-#include "docqual.h"
-#include "control.h"
-#include "output.h"
-#include "callcpp.h"
-#include "globals.h"
 #include "sorthelper.h"
+#include "tessbox.h"
 #include "tesseractclass.h"
+#include "tessvars.h"
+#include "werdit.h"
 
 #define MIN_FONT_ROW_COUNT  8
 #define MAX_XHEIGHT_DIFF  3
@@ -73,7 +74,6 @@ void Tesseract::recog_pseudo_word(PAGE_RES* page_res,
   }
 }
 
-
 /**
  * Recognize a single word in interactive mode.
  *
@@ -85,7 +85,12 @@ BOOL8 Tesseract::recog_interactive(PAGE_RES_IT* pr_it) {
 
   WordData word_data(*pr_it);
   SetupWordPassN(2, &word_data);
+  // LSTM doesn't run on pass2, but we want to run pass2 for tesseract.
+  if (lstm_recognizer_ == NULL) {
     classify_word_and_language(2, pr_it, &word_data);
+  } else {
+    classify_word_and_language(1, pr_it, &word_data);
+  }
   if (tessedit_debug_quality_metrics) {
     WERD_RES* word_res = pr_it->word();
     word_char_quality(word_res, pr_it->row()->row, &char_qual, &good_char_qual);
@@ -188,8 +193,8 @@ void Tesseract::SetupWordPassN(int pass_n, WordData* word) {
       WERD_RES* word_res = new WERD_RES;
       word_res->InitForRetryRecognition(*word->word);
       word->lang_words.push_back(word_res);
-      // Cube doesn't get setup for pass2.
+      // LSTM doesn't get setup for pass2.
-      if (pass_n == 1 || lang_t->tessedit_ocr_engine_mode != OEM_CUBE_ONLY) {
+      if (pass_n == 1 || lang_t->tessedit_ocr_engine_mode != OEM_LSTM_ONLY) {
         word_res->SetupForRecognition(
               lang_t->unicharset, lang_t, BestPix(),
               lang_t->tessedit_ocr_engine_mode, NULL,
@@ -220,15 +225,13 @@ bool Tesseract::RecogAllWordsPassN(int pass_n, ETEXT_DESC* monitor,
         monitor->progress = 70 * w / words->size();
         if (monitor->progress_callback != NULL) {
           TBOX box = pr_it->word()->word->bounding_box();
-            (*monitor->progress_callback)(monitor->progress,
+          (*monitor->progress_callback)(monitor->progress, box.left(),
-                                          box.left(), box.right(),
+                                        box.right(), box.top(), box.bottom());
-                                          box.top(), box.bottom());
         }
       } else {
         monitor->progress = 70 + 30 * w / words->size();
-        if (monitor->progress_callback!=NULL) {
+        if (monitor->progress_callback != NULL) {
-                      (*monitor->progress_callback)(monitor->progress,
+          (*monitor->progress_callback)(monitor->progress, 0, 0, 0, 0);
-                                                    0, 0, 0, 0);
         }
       }
       if (monitor->deadline_exceeded() ||
@@ -253,7 +256,8 @@ bool Tesseract::RecogAllWordsPassN(int pass_n, ETEXT_DESC* monitor,
       pr_it->forward();
     ASSERT_HOST(pr_it->word() != NULL);
     bool make_next_word_fuzzy = false;
-    if (ReassignDiacritics(pass_n, pr_it, &make_next_word_fuzzy)) {
+    if (!AnyLSTMLang() &&
+        ReassignDiacritics(pass_n, pr_it, &make_next_word_fuzzy)) {
       // Needs to be setup again to see the new outlines in the chopped_word.
       SetupWordPassN(pass_n, word);
     }
@@ -384,9 +388,8 @@ bool Tesseract::recog_all_words(PAGE_RES* page_res,
     if (!RecogAllWordsPassN(2, monitor, &page_res_it, &words)) return false;
   }
 
-  // The next passes can only be run if tesseract has been used, as cube
+  // The next passes are only required for Tess-only.
-  // doesn't set all the necessary outputs in WERD_RES.
+  if (AnyTessLang() && !AnyLSTMLang()) {
-  if (AnyTessLang()) {
     // ****************** Pass 3 *******************
     // Fix fuzzy spaces.
     set_global_loc_code(LOC_FUZZY_SPACE);
@@ -402,15 +405,6 @@ bool Tesseract::recog_all_words(PAGE_RES* page_res,
     // ****************** Pass 5,6 *******************
     rejection_passes(page_res, monitor, target_word_box, word_config);
 
-#ifndef NO_CUBE_BUILD
-    // ****************** Pass 7 *******************
-    // Cube combiner.
-    // If cube is loaded and its combiner is present, run it.
-    if (tessedit_ocr_engine_mode == OEM_TESSERACT_CUBE_COMBINED) {
-      run_cube_combiner(page_res);
-    }
-#endif
-
     // ****************** Pass 8 *******************
     font_recognition_pass(page_res);
 
@@ -438,9 +432,14 @@ bool Tesseract::recog_all_words(PAGE_RES* page_res,
   for (page_res_it.restart_page(); page_res_it.word() != NULL;
        page_res_it.forward()) {
     WERD_RES* word = page_res_it.word();
-    if (word->best_choice == NULL || word->best_choice->length() == 0)
+    POLY_BLOCK* pb = page_res_it.block()->block != NULL
+                         ? page_res_it.block()->block->poly_block()
+                         : NULL;
+    if (word->best_choice == NULL || word->best_choice->length() == 0 ||
+        (word->best_choice->IsAllSpaces() && (pb == NULL || pb->IsText()))) {
       page_res_it.DeleteCurrentWord();
     }
+  }
 
   if (monitor != NULL) {
     monitor->progress = 100;
@@ -539,7 +538,7 @@ void Tesseract::bigram_correction_pass(PAGE_RES *page_res) {
         }
       }
     }
-    if (overrides_word1.size() >= 1) {
+    if (!overrides_word1.empty()) {
       // Excellent, we have some bigram matches.
       if (EqualIgnoringCaseAndTerminalPunct(*w_prev->best_choice,
                                             *overrides_word1[best_idx]) &&
@@ -755,16 +754,32 @@ void Tesseract::script_pos_pass(PAGE_RES* page_res) {
   }
 }
 
-// Factored helper considers the indexed word and updates all the pointed
+// Helper finds the gap between the index word and the next.
-// values.
+static void WordGap(const PointerVector<WERD_RES>& words, int index, int* right,
-static void EvaluateWord(const PointerVector<WERD_RES>& words, int index,
+                    int* next_left) {
-                         float* rating, float* certainty, bool* bad,
-                         bool* valid_permuter, int* right, int* next_left) {
   *right = -MAX_INT32;
   *next_left = MAX_INT32;
   if (index < words.size()) {
+    *right = words[index]->word->bounding_box().right();
+    if (index + 1 < words.size())
+      *next_left = words[index + 1]->word->bounding_box().left();
+  }
+}
+
+// Factored helper computes the rating, certainty, badness and validity of
+// the permuter of the words in [first_index, end_index).
+static void EvaluateWordSpan(const PointerVector<WERD_RES>& words,
+                             int first_index, int end_index, float* rating,
+                             float* certainty, bool* bad,
+                             bool* valid_permuter) {
+  if (end_index <= first_index) {
+    *bad = true;
+    *valid_permuter = false;
+  }
+  for (int index = first_index; index < end_index && index < words.size();
+       ++index) {
     WERD_CHOICE* choice = words[index]->best_choice;
-    if (choice == NULL) {
+    if (choice == nullptr) {
       *bad = true;
     } else {
       *rating += choice->rating();
@@ -772,12 +787,6 @@ static void EvaluateWord(const PointerVector<WERD_RES>& words, int index,
       if (!Dict::valid_word_permuter(choice->permuter(), false))
         *valid_permuter = false;
     }
-    *right = words[index]->word->bounding_box().right();
-    if (index + 1 < words.size())
-      *next_left = words[index + 1]->word->bounding_box().left();
-  } else {
-    *valid_permuter = false;
-    *bad = true;
   }
 }
 
@@ -802,24 +811,13 @@ static int SelectBestWords(double rating_ratio,
   while (b < best_words->size() || n < new_words->size()) {
     // Start of the current run in each.
     int start_b = b, start_n = n;
-    // Rating of the current run in each.
-    float b_rating = 0.0f, n_rating = 0.0f;
-    // Certainty of the current run in each.
-    float b_certainty = 0.0f, n_certainty = 0.0f;
-    // True if any word is missing its best choice.
-    bool b_bad = false, n_bad = false;
-    // True if all words have a valid permuter.
-    bool b_valid_permuter = true, n_valid_permuter = true;
-
     while (b < best_words->size() || n < new_words->size()) {
       int b_right = -MAX_INT32;
       int next_b_left = MAX_INT32;
-      EvaluateWord(*best_words, b, &b_rating, &b_certainty, &b_bad,
+      WordGap(*best_words, b, &b_right, &next_b_left);
-                   &b_valid_permuter, &b_right, &next_b_left);
       int n_right = -MAX_INT32;
       int next_n_left = MAX_INT32;
-      EvaluateWord(*new_words, n, &n_rating, &n_certainty, &n_bad,
+      WordGap(*new_words, n, &n_right, &next_n_left);
-                   &n_valid_permuter, &n_right, &next_n_left);
       if (MAX(b_right, n_right) < MIN(next_b_left, next_n_left)) {
         // The word breaks overlap. [start_b,b] and [start_n, n] match.
         break;
@@ -831,6 +829,20 @@ static int SelectBestWords(double rating_ratio,
       else
         ++n;
     }
+    // Rating of the current run in each.
+    float b_rating = 0.0f, n_rating = 0.0f;
+    // Certainty of the current run in each.
+    float b_certainty = 0.0f, n_certainty = 0.0f;
+    // True if any word is missing its best choice.
+    bool b_bad = false, n_bad = false;
+    // True if all words have a valid permuter.
+    bool b_valid_permuter = true, n_valid_permuter = true;
+    int end_b = b < best_words->size() ? b + 1 : b;
+    int end_n = n < new_words->size() ? n + 1 : n;
+    EvaluateWordSpan(*best_words, start_b, end_b, &b_rating, &b_certainty,
+                     &b_bad, &b_valid_permuter);
+    EvaluateWordSpan(*new_words, start_n, end_n, &n_rating, &n_certainty,
+                     &n_bad, &n_valid_permuter);
     bool new_better = false;
     if (!n_bad && (b_bad || (n_certainty > b_certainty &&
                              n_rating < b_rating) ||
@@ -838,7 +850,7 @@ static int SelectBestWords(double rating_ratio,
                              n_rating < b_rating * rating_ratio &&
                              n_certainty > b_certainty - certainty_margin))) {
       // New is better.
-      for (int i = start_n; i <= n; ++i) {
+      for (int i = start_n; i < end_n; ++i) {
         out_words.push_back((*new_words)[i]);
         (*new_words)[i] = NULL;
         ++num_new;
@@ -846,14 +858,12 @@ static int SelectBestWords(double rating_ratio,
       new_better = true;
     } else if (!b_bad) {
       // Current best is better.
-      for (int i = start_b; i <= b; ++i) {
+      for (int i = start_b; i < end_b; ++i) {
         out_words.push_back((*best_words)[i]);
         (*best_words)[i] = NULL;
         ++num_best;
       }
     }
-    int end_b = b < best_words->size() ? b + 1 : b;
-    int end_n = n < new_words->size() ? n + 1 : n;
     if (debug) {
       tprintf("%d new words %s than %d old words: r: %g v %g c: %g v %g"
               " valid dict: %d v %d\n",
@@ -876,10 +886,9 @@ static int SelectBestWords(double rating_ratio,
 // Returns positive if this recognizer found more new best words than the
 // number kept from best_words.
 int Tesseract::RetryWithLanguage(const WordData& word_data,
-                                 WordRecognizer recognizer,
+                                 WordRecognizer recognizer, bool debug,
                                  WERD_RES** in_word,
                                  PointerVector<WERD_RES>* best_words) {
-  bool debug = classify_debug_level || cube_debug_level;
   if (debug) {
     tprintf("Trying word using lang %s, oem %d\n",
             lang.string(), static_cast<int>(tessedit_ocr_engine_mode));
@@ -898,8 +907,7 @@ int Tesseract::RetryWithLanguage(const WordData& word_data,
       new_words[i]->DebugTopChoice("Lang result");
   }
   // Initial version is a bit of a hack based on better certainty and rating
-  // (to reduce false positives from cube) or a dictionary vs non-dictionary
+  // or a dictionary vs non-dictionary word.
-  // word.
   return SelectBestWords(classify_max_rating_ratio,
                          classify_max_certainty_margin,
                          debug, &new_words, best_words);
@@ -1283,7 +1291,8 @@ void Tesseract::classify_word_and_language(int pass_n, PAGE_RES_IT* pr_it,
   // Points to the best result. May be word or in lang_words.
   WERD_RES* word = word_data->word;
   clock_t start_t = clock();
-  if (classify_debug_level || cube_debug_level) {
+  bool debug = classify_debug_level > 0 || multilang_debug_level > 0;
+  if (debug) {
     tprintf("%s word with lang %s at:",
             word->done ? "Already done" : "Processing",
             most_recently_used_->lang.string());
@@ -1302,12 +1311,12 @@ void Tesseract::classify_word_and_language(int pass_n, PAGE_RES_IT* pr_it,
          most_recently_used_ != sub_langs_[sub]; ++sub) {}
   }
   most_recently_used_->RetryWithLanguage(
-      *word_data, recognizer, &word_data->lang_words[sub], &best_words);
+      *word_data, recognizer, debug, &word_data->lang_words[sub], &best_words);
   Tesseract* best_lang_tess = most_recently_used_;
   if (!WordsAcceptable(best_words)) {
     // Try all the other languages to see if they are any better.
     if (most_recently_used_ != this &&
-        this->RetryWithLanguage(*word_data, recognizer,
+        this->RetryWithLanguage(*word_data, recognizer, debug,
                                 &word_data->lang_words[sub_langs_.size()],
                                 &best_words) > 0) {
       best_lang_tess = this;
@@ -1315,7 +1324,7 @@ void Tesseract::classify_word_and_language(int pass_n, PAGE_RES_IT* pr_it,
     for (int i = 0; !WordsAcceptable(best_words) && i < sub_langs_.size();
          ++i) {
       if (most_recently_used_ != sub_langs_[i] &&
-          sub_langs_[i]->RetryWithLanguage(*word_data, recognizer,
+          sub_langs_[i]->RetryWithLanguage(*word_data, recognizer, debug,
                                            &word_data->lang_words[i],
                                            &best_words) > 0) {
         best_lang_tess = sub_langs_[i];
@@ -1357,12 +1366,26 @@ void Tesseract::classify_word_pass1(const WordData& word_data,
   BLOCK* block = word_data.block;
   prev_word_best_choice_ = word_data.prev_word != NULL
       ? word_data.prev_word->word->best_choice : NULL;
-#ifndef NO_CUBE_BUILD
+#ifndef ANDROID_BUILD
-  // If we only intend to run cube - run it and return.
+  if (tessedit_ocr_engine_mode == OEM_LSTM_ONLY ||
-  if (tessedit_ocr_engine_mode == OEM_CUBE_ONLY) {
+      tessedit_ocr_engine_mode == OEM_TESSERACT_LSTM_COMBINED) {
-    cube_word_pass1(block, row, *in_word);
+    if (!(*in_word)->odd_size || tessedit_ocr_engine_mode == OEM_LSTM_ONLY) {
+      LSTMRecognizeWord(*block, row, *in_word, out_words);
+      if (!out_words->empty())
+        return;  // Successful lstm recognition.
+    }
+    if (tessedit_ocr_engine_mode == OEM_LSTM_ONLY) {
+      // No fallback allowed, so use a fake.
+      (*in_word)->SetupFake(lstm_recognizer_->GetUnicharset());
       return;
     }
+    // Fall back to tesseract for failed words or odd words.
+    (*in_word)->SetupForRecognition(unicharset, this, BestPix(),
+                                    OEM_TESSERACT_ONLY, NULL,
+                                    classify_bln_numeric_mode,
+                                    textord_use_cjk_fp_model,
+                                    poly_allow_detailed_fx, row, block);
+  }
 #endif
   WERD_RES* word = *in_word;
   match_word_pass_n(1, word, row, block);
@@ -1497,11 +1520,7 @@ void Tesseract::classify_word_pass2(const WordData& word_data,
                                     WERD_RES** in_word,
                                     PointerVector<WERD_RES>* out_words) {
   // Return if we do not want to run Tesseract.
-  if (tessedit_ocr_engine_mode != OEM_TESSERACT_ONLY &&
+  if (tessedit_ocr_engine_mode == OEM_LSTM_ONLY) {
-      tessedit_ocr_engine_mode != OEM_TESSERACT_CUBE_COMBINED &&
-      word_data.word->best_choice != NULL)
-    return;
-  if (tessedit_ocr_engine_mode == OEM_CUBE_ONLY) {
     return;
   }
   ROW* row = word_data.row;
@@ -1886,7 +1905,7 @@ static void find_modal_font(           //good chars in word
  * Get the fonts for the word.
  */
 void Tesseract::set_word_fonts(WERD_RES *word) {
-  // Don't try to set the word fonts for a cube word, as the configs
+  // Don't try to set the word fonts for an lstm word, as the configs
   // will be meaningless.
   if (word->chopped_word == NULL) return;
   ASSERT_HOST(word->best_choice != NULL);

ccmain/cube_control.cpp

@@ -1,432 +0,0 @@
-/******************************************************************
- * File:        cube_control.cpp
- * Description: Tesseract class methods for invoking cube convolutional
- *              neural network word recognizer.
- * Author:      Raquel Romano
- * Created:     September 2009
- *
- **********************************************************************/
-
-// Include automatically generated configuration file if running autoconf.
-#ifdef HAVE_CONFIG_H
-#include "config_auto.h"
-#endif
-
-#include "allheaders.h"
-
-#include "cube_object.h"
-#include "cube_reco_context.h"
-#include "tesseractclass.h"
-#include "tesseract_cube_combiner.h"
-
-namespace tesseract {
-
-/**
- * @name convert_prob_to_tess_certainty
- *
- * Normalize a probability in the range [0.0, 1.0] to a tesseract
- * certainty in the range [-20.0, 0.0]
- */
-static float convert_prob_to_tess_certainty(float prob) {
-  return (prob - 1.0) * 20.0;
-}
-
-/**
- * @name char_box_to_tbox
- *
- * Create a TBOX from a character bounding box. If nonzero, the
- * x_offset accounts for any additional padding of the word box that
- * should be taken into account.
- *
- */
-TBOX char_box_to_tbox(Box* char_box, TBOX word_box, int x_offset) {
-  l_int32 left;
-  l_int32 top;
-  l_int32 width;
-  l_int32 height;
-  l_int32 right;
-  l_int32 bottom;
-
-  boxGetGeometry(char_box, &left, &top, &width, &height);
-  left += word_box.left() - x_offset;
-  right = left + width;
-  top = word_box.bottom() + word_box.height() - top;
-  bottom = top - height;
-  return TBOX(left, bottom, right, top);
-}
-
-/**
- * @name extract_cube_state
- *
- * Extract CharSamp objects and character bounding boxes from the
- * CubeObject's state. The caller should free both structres.
- *
- */
-bool Tesseract::extract_cube_state(CubeObject* cube_obj,
-                                   int* num_chars,
-                                   Boxa** char_boxes,
-                                   CharSamp*** char_samples) {
-  if (!cube_obj) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (extract_cube_state): Invalid cube object "
-              "passed to extract_cube_state\n");
-    }
-    return false;
-  }
-
-  // Note that the CubeObject accessors return either the deslanted or
-  // regular objects search object or beam search object, whichever
-  // was used in the last call to Recognize()
-  CubeSearchObject* cube_search_obj = cube_obj->SrchObj();
-  if (!cube_search_obj) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (Extract_cube_state): Could not retrieve "
-              "cube's search object in extract_cube_state.\n");
-    }
-    return false;
-  }
-  BeamSearch *beam_search_obj = cube_obj->BeamObj();
-  if (!beam_search_obj) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (Extract_cube_state): Could not retrieve "
-              "cube's beam search object in extract_cube_state.\n");
-    }
-    return false;
-  }
-
-  // Get the character samples and bounding boxes by backtracking
-  // through the beam search path
-  int best_node_index = beam_search_obj->BestPresortedNodeIndex();
-  *char_samples = beam_search_obj->BackTrack(
-      cube_search_obj, best_node_index, num_chars, NULL, char_boxes);
-  if (!*char_samples)
-    return false;
-  return true;
-}
-
-/**
- * @name create_cube_box_word
- *
- * Fill the given BoxWord with boxes from character bounding
- * boxes. The char_boxes have local coordinates w.r.t. the
- * word bounding box, i.e., the left-most character bbox of each word
- * has (0,0) left-top coord, but the BoxWord must be defined in page
- * coordinates.
- */
-bool Tesseract::create_cube_box_word(Boxa *char_boxes,
-                                     int num_chars,
-                                     TBOX word_box,
-                                     BoxWord* box_word) {
-  if (!box_word) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (create_cube_box_word): Invalid box_word.\n");
-    }
-    return false;
-  }
-
-  // Find the x-coordinate of left-most char_box, which could be
-  // nonzero if the word image was padded before recognition took place.
-  int x_offset = -1;
-  for (int i = 0; i < num_chars; ++i) {
-    Box* char_box = boxaGetBox(char_boxes, i, L_CLONE);
-    if (x_offset < 0 || char_box->x < x_offset) {
-      x_offset = char_box->x;
-    }
-    boxDestroy(&char_box);
-  }
-
-  for (int i = 0; i < num_chars; ++i) {
-    Box* char_box = boxaGetBox(char_boxes, i, L_CLONE);
-    TBOX tbox = char_box_to_tbox(char_box, word_box, x_offset);
-    boxDestroy(&char_box);
-    box_word->InsertBox(i, tbox);
-  }
-  return true;
-}
-
-/**
- * @name init_cube_objects
- *
- * Instantiates Tesseract object's CubeRecoContext and TesseractCubeCombiner.
- * Returns false if cube context could not be created or if load_combiner is
- * true, but the combiner could not be loaded.
- */
-bool Tesseract::init_cube_objects(bool load_combiner,
-                                  TessdataManager *tessdata_manager) {
-  ASSERT_HOST(cube_cntxt_ == NULL);
-  ASSERT_HOST(tess_cube_combiner_ == NULL);
-
-  // Create the cube context object
-  cube_cntxt_ = CubeRecoContext::Create(this, tessdata_manager, &unicharset);
-  if (cube_cntxt_ == NULL) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (Tesseract::init_cube_objects()): Failed to "
-              "instantiate CubeRecoContext\n");
-    }
-    return false;
-  }
-
-  // Create the combiner object and load the combiner net for target languages.
-  if (load_combiner) {
-    tess_cube_combiner_ = new tesseract::TesseractCubeCombiner(cube_cntxt_);
-    if (!tess_cube_combiner_ || !tess_cube_combiner_->LoadCombinerNet()) {
-      delete cube_cntxt_;
-      cube_cntxt_ = NULL;
-      if (tess_cube_combiner_ != NULL) {
-        delete tess_cube_combiner_;
-        tess_cube_combiner_ = NULL;
-      }
-      if (cube_debug_level > 0)
-        tprintf("Cube ERROR (Failed to instantiate TesseractCubeCombiner\n");
-      return false;
-    }
-  }
-  return true;
-}
-
-/**
- * @name run_cube_combiner
- *
- * Iterates through tesseract's results and calls cube on each word,
- * combining the results with the existing tesseract result.
- */
-void Tesseract::run_cube_combiner(PAGE_RES *page_res) {
-  if (page_res == NULL || tess_cube_combiner_ == NULL)
-    return;
-  PAGE_RES_IT page_res_it(page_res);
-  // Iterate through the word results and call cube on each word.
-  for (page_res_it.restart_page(); page_res_it.word () != NULL;
-       page_res_it.forward()) {
-    BLOCK* block = page_res_it.block()->block;
-    if (block->poly_block() != NULL && !block->poly_block()->IsText())
-      continue;  // Don't deal with non-text blocks.
-    WERD_RES* word = page_res_it.word();
-    // Skip cube entirely if tesseract's certainty is greater than threshold.
-    int combiner_run_thresh = convert_prob_to_tess_certainty(
-        cube_cntxt_->Params()->CombinerRunThresh());
-    if (word->best_choice->certainty() >= combiner_run_thresh) {
-      continue;
-    }
-    // Use the same language as Tesseract used for the word.
-    Tesseract* lang_tess = word->tesseract;
-
-    // Setup a trial WERD_RES in which to classify with cube.
-    WERD_RES cube_word;
-    cube_word.InitForRetryRecognition(*word);
-    cube_word.SetupForRecognition(lang_tess->unicharset, this, BestPix(),
-                                  OEM_CUBE_ONLY,
-                                  NULL, false, false, false,
-                                  page_res_it.row()->row,
-                                  page_res_it.block()->block);
-    CubeObject *cube_obj = lang_tess->cube_recognize_word(
-        page_res_it.block()->block, &cube_word);
-    if (cube_obj != NULL)
-      lang_tess->cube_combine_word(cube_obj, &cube_word, word);
-    delete cube_obj;
-  }
-}
-
-/**
- * @name cube_word_pass1
- *
- * Recognizes a single word using (only) cube. Compatible with
- * Tesseract's classify_word_pass1/classify_word_pass2.
- */
-void Tesseract::cube_word_pass1(BLOCK* block, ROW *row, WERD_RES *word) {
-  CubeObject *cube_obj = cube_recognize_word(block, word);
-  delete cube_obj;
-}
-
-/**
- * @name cube_recognize_word
- *
- * Cube recognizer to recognize a single word as with classify_word_pass1
- * but also returns the cube object in case the combiner is needed.
- */
-CubeObject* Tesseract::cube_recognize_word(BLOCK* block, WERD_RES* word) {
-  if (!cube_binary_ || !cube_cntxt_) {
-    if (cube_debug_level > 0 && !cube_binary_)
-      tprintf("Tesseract::run_cube(): NULL binary image.\n");
-    word->SetupFake(unicharset);
-    return NULL;
-  }
-  TBOX word_box = word->word->bounding_box();
-  if (block != NULL && (block->re_rotation().x() != 1.0f ||
-        block->re_rotation().y() != 0.0f)) {
-    // TODO(rays) We have to rotate the bounding box to get the true coords.
-    // This will be achieved in the future via DENORM.
-    // In the mean time, cube can't process this word.
-    if (cube_debug_level > 0) {
-      tprintf("Cube can't process rotated word at:");
-      word_box.print();
-    }
-    word->SetupFake(unicharset);
-    return NULL;
-  }
-  CubeObject* cube_obj = new tesseract::CubeObject(
-      cube_cntxt_, cube_binary_, word_box.left(),
-      pixGetHeight(cube_binary_) - word_box.top(),
-      word_box.width(), word_box.height());
-  if (!cube_recognize(cube_obj, block, word)) {
-    delete cube_obj;
-    return NULL;
-  }
-  return cube_obj;
-}
-
-/**
- * @name cube_combine_word
- *
- * Combines the cube and tesseract results for a single word, leaving the
- * result in tess_word.
- */
-void Tesseract::cube_combine_word(CubeObject* cube_obj, WERD_RES* cube_word,
-                                  WERD_RES* tess_word) {
-  float combiner_prob = tess_cube_combiner_->CombineResults(tess_word,
-                                                            cube_obj);
-  // If combiner probability is greater than tess/cube combiner
-  // classifier threshold, i.e. tesseract wins, then just return the
-  // tesseract result unchanged, as the combiner knows nothing about how
-  // correct the answer is. If cube and tesseract agree, then improve the
-  // scores before returning.
-  WERD_CHOICE* tess_best = tess_word->best_choice;
-  WERD_CHOICE* cube_best = cube_word->best_choice;
-  if (cube_debug_level || classify_debug_level) {
-    tprintf("Combiner prob = %g vs threshold %g\n",
-            combiner_prob, cube_cntxt_->Params()->CombinerClassifierThresh());
-  }
-  if (combiner_prob >=
-      cube_cntxt_->Params()->CombinerClassifierThresh()) {
-    if (tess_best->unichar_string() == cube_best->unichar_string()) {
-      // Cube and tess agree, so improve the scores.
-      tess_best->set_rating(tess_best->rating() / 2);
-      tess_best->set_certainty(tess_best->certainty() / 2);
-    }
-    return;
-  }
-  // Cube wins.
-  // It is better for the language combiner to have all tesseract scores,
-  // so put them in the cube result.
-  cube_best->set_rating(tess_best->rating());
-  cube_best->set_certainty(tess_best->certainty());
-  if (cube_debug_level || classify_debug_level) {
-    tprintf("Cube INFO: tesseract result replaced by cube: %s -> %s\n",
-            tess_best->unichar_string().string(),
-            cube_best->unichar_string().string());
-  }
-  tess_word->ConsumeWordResults(cube_word);
-}
-
-/**
- * @name cube_recognize
- *
- * Call cube on the current word, and write the result to word.
- * Sets up a fake result and returns false if something goes wrong.
- */
-bool Tesseract::cube_recognize(CubeObject *cube_obj, BLOCK* block,
-                               WERD_RES *word) {
-  // Run cube
-  WordAltList *cube_alt_list = cube_obj->RecognizeWord();
-  if (!cube_alt_list || cube_alt_list->AltCount() <= 0) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube returned nothing for word at:");
-      word->word->bounding_box().print();
-    }
-    word->SetupFake(unicharset);
-    return false;
-  }
-
-  // Get cube's best result and its probability, mapped to tesseract's
-  // certainty range
-  char_32 *cube_best_32 = cube_alt_list->Alt(0);
-  double cube_prob = CubeUtils::Cost2Prob(cube_alt_list->AltCost(0));
-  float cube_certainty = convert_prob_to_tess_certainty(cube_prob);
-  string cube_best_str;
-  CubeUtils::UTF32ToUTF8(cube_best_32, &cube_best_str);
-
-  // Retrieve Cube's character bounding boxes and CharSamples,
-  // corresponding to the most recent call to RecognizeWord().
-  Boxa *char_boxes = NULL;
-  CharSamp **char_samples = NULL;;
-  int num_chars;
-  if (!extract_cube_state(cube_obj, &num_chars, &char_boxes, &char_samples)
-      && cube_debug_level > 0) {
-    tprintf("Cube WARNING (Tesseract::cube_recognize): Cannot extract "
-            "cube state.\n");
-    word->SetupFake(unicharset);
-    return false;
-  }
-
-  // Convert cube's character bounding boxes to a BoxWord.
-  BoxWord cube_box_word;
-  TBOX tess_word_box = word->word->bounding_box();
-  if (word->denorm.block() != NULL)
-    tess_word_box.rotate(word->denorm.block()->re_rotation());
-  bool box_word_success = create_cube_box_word(char_boxes, num_chars,
-                                               tess_word_box,
-                                               &cube_box_word);
-  boxaDestroy(&char_boxes);
-  if (!box_word_success) {
-    if (cube_debug_level > 0) {
-      tprintf("Cube WARNING (Tesseract::cube_recognize): Could not "
-              "create cube BoxWord\n");
-    }
-    word->SetupFake(unicharset);
-    return false;
-  }
-
-  // Fill tesseract result's fields with cube results
-  fill_werd_res(cube_box_word, cube_best_str.c_str(), word);
-
-  // Create cube's best choice.
-  BLOB_CHOICE** choices = new BLOB_CHOICE*[num_chars];
-  for (int i = 0; i < num_chars; ++i) {
-    UNICHAR_ID uch_id =
-        cube_cntxt_->CharacterSet()->UnicharID(char_samples[i]->StrLabel());
-    choices[i] = new BLOB_CHOICE(uch_id, -cube_certainty, cube_certainty,
-                                 -1, 0.0f, 0.0f, 0.0f, BCC_STATIC_CLASSIFIER);
-  }
-  word->FakeClassifyWord(num_chars, choices);
-  // within a word, cube recognizes the word in reading order.
-  word->best_choice->set_unichars_in_script_order(true);
-  delete [] choices;
-  delete [] char_samples;
-
-  // Some sanity checks
-  ASSERT_HOST(word->best_choice->length() == word->reject_map.length());
-
-  if (cube_debug_level || classify_debug_level) {
-    tprintf("Cube result: %s r=%g, c=%g\n",
-            word->best_choice->unichar_string().string(),
-            word->best_choice->rating(),
-            word->best_choice->certainty());
-  }
-  return true;
-}
-
-/**
- * @name fill_werd_res
- *
- * Fill Tesseract's word result fields with cube's.
- *
- */
-void Tesseract::fill_werd_res(const BoxWord& cube_box_word,
-                              const char* cube_best_str,
-                              WERD_RES* tess_werd_res) {
-  delete tess_werd_res->box_word;
-  tess_werd_res->box_word = new BoxWord(cube_box_word);
-  tess_werd_res->box_word->ClipToOriginalWord(tess_werd_res->denorm.block(),
-                                              tess_werd_res->word);
-  // Fill text and remaining fields
-  tess_werd_res->word->set_text(cube_best_str);
-  tess_werd_res->tess_failed = FALSE;
-  tess_werd_res->tess_accepted = tess_acceptable_word(tess_werd_res);
-  // There is no output word, so we can' call AdaptableWord, but then I don't
-  // think we need to. Fudge the result with accepted.
-  tess_werd_res->tess_would_adapt = tess_werd_res->tess_accepted;
-
-  // Set word to done, i.e., ignore all of tesseract's tests for rejection
-  tess_werd_res->done = tess_werd_res->tess_accepted;
-}
-
-}  // namespace tesseract

ccmain/cube_reco_context.cpp

@@ -1,208 +0,0 @@
-/**********************************************************************
- * File:        cube_reco_context.cpp
- * Description: Implementation of the Cube Recognition Context Class
- * Author:    Ahmad Abdulkader
- * Created:   2007
- *
- * (C) Copyright 2008, Google Inc.
- ** Licensed under the Apache License, Version 2.0 (the "License");
- ** you may not use this file except in compliance with the License.
- ** You may obtain a copy of the License at
- ** http://www.apache.org/licenses/LICENSE-2.0
- ** Unless required by applicable law or agreed to in writing, software
- ** distributed under the License is distributed on an "AS IS" BASIS,
- ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- ** See the License for the specific language governing permissions and
- ** limitations under the License.
- *
- **********************************************************************/
-
-#include <string>
-#include <limits.h>
-
-#include "cube_reco_context.h"
-
-#include "classifier_factory.h"
-#include "cube_tuning_params.h"
-#include "dict.h"
-#include "feature_bmp.h"
-#include "tessdatamanager.h"
-#include "tesseractclass.h"
-#include "tess_lang_model.h"
-
-namespace tesseract {
-
-/**
- * Instantiate a CubeRecoContext object using a Tesseract object.
- * CubeRecoContext will not take ownership of tess_obj, but will
- * record the pointer to it and will make use of various Tesseract
- * components (language model, flags, etc). Thus the caller should
- * keep tess_obj alive so long as the instantiated CubeRecoContext is used.
- */
-CubeRecoContext::CubeRecoContext(Tesseract *tess_obj) {
-  tess_obj_ = tess_obj;
-  lang_ = "";
-  loaded_ = false;
-  lang_mod_ = NULL;
-  params_ = NULL;
-  char_classifier_ = NULL;
-  char_set_ = NULL;
-  word_size_model_ = NULL;
-  char_bigrams_ = NULL;
-  word_unigrams_ = NULL;
-  noisy_input_ = false;
-  size_normalization_ = false;
-}
-
-CubeRecoContext::~CubeRecoContext() {
-  if (char_classifier_ != NULL) {
-    delete char_classifier_;
-    char_classifier_ = NULL;
-  }
-
-  if (word_size_model_ != NULL) {
-    delete word_size_model_;
-    word_size_model_ = NULL;
-  }
-
-  if (char_set_ != NULL) {
-    delete char_set_;
-    char_set_ = NULL;
-  }
-
-  if (char_bigrams_ != NULL) {
-    delete char_bigrams_;
-    char_bigrams_ = NULL;
-  }
-
-  if (word_unigrams_ != NULL) {
-    delete word_unigrams_;
-    word_unigrams_ = NULL;
-  }
-
-  if (lang_mod_ != NULL) {
-    delete lang_mod_;
-    lang_mod_ = NULL;
-  }
-
-  if (params_ != NULL) {
-    delete params_;
-    params_ = NULL;
-  }
-}
-
-/**
- * Returns the path of the data files by looking up the TESSDATA_PREFIX
- * environment variable and appending a "tessdata" directory to it
- */
-bool CubeRecoContext::GetDataFilePath(string *path) const {
-  *path = tess_obj_->datadir.string();
-  return true;
-}
-
-/**
- * The object initialization function that loads all the necessary
- * components of a RecoContext.  TessdataManager is used to load the
- * data from [lang].traineddata file.  If TESSDATA_CUBE_UNICHARSET
- * component is present, Cube will be instantiated with the unicharset
- * specified in this component and the corresponding dictionary
- * (TESSDATA_CUBE_SYSTEM_DAWG), and will map Cube's unicharset to
- * Tesseract's. Otherwise, TessdataManager will assume that Cube will
- * be using Tesseract's unicharset and dawgs, and will load the
- * unicharset from the TESSDATA_UNICHARSET component and will load the
- * dawgs from TESSDATA_*_DAWG components.
- */
-bool CubeRecoContext::Load(TessdataManager *tessdata_manager,
-                           UNICHARSET *tess_unicharset) {
-  ASSERT_HOST(tess_obj_ != NULL);
-  tess_unicharset_ = tess_unicharset;
-  string data_file_path;
-
-  // Get the data file path.
-  if (GetDataFilePath(&data_file_path) == false) {
-    fprintf(stderr, "Unable to get data file path\n");
-    return false;
-  }
-
-  // Get the language from the Tesseract object.
-  lang_ = tess_obj_->lang.string();
-
-  // Create the char set.
-  if ((char_set_ =
-       CharSet::Create(tessdata_manager, tess_unicharset)) == NULL) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Load): unable to load "
-            "CharSet\n");
-    return false;
-  }
-  // Create the language model.
-  string lm_file_name = data_file_path + lang_ + ".cube.lm";
-  string lm_params;
-  if (!CubeUtils::ReadFileToString(lm_file_name, &lm_params)) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Load): unable to read cube "
-            "language model params from %s\n", lm_file_name.c_str());
-    return false;
-  }
-  lang_mod_ = new TessLangModel(lm_params, data_file_path,
-                                tess_obj_->getDict().load_system_dawg,
-                                tessdata_manager, this);
-  if (lang_mod_ == NULL) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Load): unable to create "
-            "TessLangModel\n");
-    return false;
-  }
-
-  // Create the optional char bigrams object.
-  char_bigrams_ = CharBigrams::Create(data_file_path, lang_);
-
-  // Create the optional word unigrams object.
-  word_unigrams_ = WordUnigrams::Create(data_file_path, lang_);
-
-  // Create the optional size model.
-  word_size_model_ = WordSizeModel::Create(data_file_path, lang_,
-    char_set_, Contextual());
-
-  // Load tuning params.
-  params_ = CubeTuningParams::Create(data_file_path, lang_);
-  if (params_ == NULL) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Load): unable to read "
-            "CubeTuningParams from %s\n", data_file_path.c_str());
-    return false;
-  }
-
-  // Create the char classifier.
-  char_classifier_ = CharClassifierFactory::Create(data_file_path, lang_,
-                                                   lang_mod_, char_set_,
-                                                   params_);
-  if (char_classifier_ == NULL) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Load): unable to load "
-            "CharClassifierFactory object from %s\n", data_file_path.c_str());
-    return false;
-  }
-
-  loaded_ = true;
-
-  return true;
-}
-
-/** Creates a CubeRecoContext object using a tesseract object */
-CubeRecoContext * CubeRecoContext::Create(Tesseract *tess_obj,
-                                          TessdataManager *tessdata_manager,
-                                          UNICHARSET *tess_unicharset) {
-  // create the object
-  CubeRecoContext *cntxt = new CubeRecoContext(tess_obj);
-  if (cntxt == NULL) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Create): unable to create "
-            "CubeRecoContext object\n");
-    return NULL;
-  }
-  // load the necessary components
-  if (cntxt->Load(tessdata_manager, tess_unicharset) == false) {
-    fprintf(stderr, "Cube ERROR (CubeRecoContext::Create): unable to init "
-            "CubeRecoContext object\n");
-    delete cntxt;
-    return NULL;
-  }
-  // success
-  return cntxt;
-}
-}  // tesseract}

ccmain/cube_reco_context.h

@@ -1,157 +0,0 @@
-/**********************************************************************
- * File:        cube_reco_context.h
- * Description: Declaration of the Cube Recognition Context Class
- * Author:    Ahmad Abdulkader
- * Created:   2007
- *
- * (C) Copyright 2008, Google Inc.
- ** Licensed under the Apache License, Version 2.0 (the "License");
- ** you may not use this file except in compliance with the License.
- ** You may obtain a copy of the License at
- ** http://www.apache.org/licenses/LICENSE-2.0
- ** Unless required by applicable law or agreed to in writing, software
- ** distributed under the License is distributed on an "AS IS" BASIS,
- ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- ** See the License for the specific language governing permissions and
- ** limitations under the License.
- *
- **********************************************************************/
-
-// The CubeRecoContext class abstracts the Cube OCR Engine. Typically a process
-// (or a thread) would create one CubeRecoContext object per language.
-// The CubeRecoContext object also provides methods to get and set the
-// different attribues of the Cube OCR Engine.
-
-#ifndef CUBE_RECO_CONTEXT_H
-#define CUBE_RECO_CONTEXT_H
-
-#include <string>
-#include "neural_net.h"
-#include "lang_model.h"
-#include "classifier_base.h"
-#include "feature_base.h"
-#include "char_set.h"
-#include "word_size_model.h"
-#include "char_bigrams.h"
-#include "word_unigrams.h"
-
-namespace tesseract {
-
-class Tesseract;
-class TessdataManager;
-
-class CubeRecoContext {
- public:
-  // Reading order enum type
-  enum ReadOrder {
-   L2R,
-   R2L
-  };
-
-  // Instantiate using a Tesseract object
-  CubeRecoContext(Tesseract *tess_obj);
-
-  ~CubeRecoContext();
-
-  // accessor functions
-  inline const string & Lang() const { return lang_; }
-  inline CharSet *CharacterSet() const { return char_set_; }
-  const UNICHARSET *TessUnicharset() const { return tess_unicharset_; }
-  inline CharClassifier *Classifier() const { return char_classifier_; }
-  inline WordSizeModel *SizeModel() const { return word_size_model_; }
-  inline CharBigrams *Bigrams() const { return char_bigrams_; }
-  inline WordUnigrams *WordUnigramsObj() const { return word_unigrams_; }
-  inline TuningParams *Params() const { return params_; }
-  inline LangModel *LangMod() const { return lang_mod_; }
-
-  // the reading order of the language
-  inline ReadOrder ReadingOrder() const {
-    return ((lang_ == "ara") ? R2L : L2R);
-  }
-
-  // does the language support case
-  inline bool HasCase() const {
-    return (lang_ != "ara" && lang_ != "hin");
-  }
-
-  inline bool Cursive() const {
-    return (lang_ == "ara");
-  }
-
-  inline bool HasItalics() const {
-    return (lang_ != "ara" && lang_ != "hin");
-  }
-
-  inline bool Contextual() const {
-    return (lang_ == "ara");
-  }
-
-  // RecoContext runtime flags accessor functions
-  inline bool SizeNormalization() const { return size_normalization_; }
-  inline bool NoisyInput() const { return noisy_input_; }
-  inline bool OOD() const { return lang_mod_->OOD(); }
-  inline bool Numeric() const { return lang_mod_->Numeric(); }
-  inline bool WordList() const { return lang_mod_->WordList(); }
-  inline bool Punc() const { return lang_mod_->Punc(); }
-  inline bool CaseSensitive() const {
-    return char_classifier_->CaseSensitive();
-  }
-
-  inline void SetSizeNormalization(bool size_normalization) {
-    size_normalization_ = size_normalization;
-  }
-  inline void SetNoisyInput(bool noisy_input) {
-    noisy_input_ = noisy_input;
-  }
-  inline void SetOOD(bool ood_enabled) {
-    lang_mod_->SetOOD(ood_enabled);
-  }
-  inline void SetNumeric(bool numeric_enabled) {
-    lang_mod_->SetNumeric(numeric_enabled);
-  }
-  inline void SetWordList(bool word_list_enabled) {
-    lang_mod_->SetWordList(word_list_enabled);
-  }
-  inline void SetPunc(bool punc_enabled) {
-    lang_mod_->SetPunc(punc_enabled);
-  }
-  inline void SetCaseSensitive(bool case_sensitive) {
-    char_classifier_->SetCaseSensitive(case_sensitive);
-  }
-  inline tesseract::Tesseract *TesseractObject() const {
-    return tess_obj_;
-  }
-
-  // Returns the path of the data files
-  bool GetDataFilePath(string *path) const;
-  // Creates a CubeRecoContext object using a tesseract object. Data
-  // files are loaded via the tessdata_manager, and the tesseract
-  // unicharset is provided in order to map Cube's unicharset to
-  // Tesseract's in the case where the two unicharsets differ.
-  static CubeRecoContext *Create(Tesseract *tess_obj,
-                                 TessdataManager *tessdata_manager,
-                                 UNICHARSET *tess_unicharset);
-
- private:
-  bool loaded_;
-  string lang_;
-  CharSet *char_set_;
-  UNICHARSET *tess_unicharset_;
-  WordSizeModel *word_size_model_;
-  CharClassifier *char_classifier_;
-  CharBigrams *char_bigrams_;
-  WordUnigrams *word_unigrams_;
-  TuningParams *params_;
-  LangModel *lang_mod_;
-  Tesseract *tess_obj_;  // CubeRecoContext does not own this pointer
-  bool size_normalization_;
-  bool noisy_input_;
-
-  // Loads and initialized all the necessary components of a
-  // CubeRecoContext. See .cpp for more details.
-  bool Load(TessdataManager *tessdata_manager,
-            UNICHARSET *tess_unicharset);
-};
-}
-
-#endif  // CUBE_RECO_CONTEXT_H

ccmain/cubeclassifier.cpp

@@ -1,134 +0,0 @@
-// Copyright 2011 Google Inc. All Rights Reserved.
-// Author: rays@google.com (Ray Smith)
-///////////////////////////////////////////////////////////////////////
-// File:        cubeclassifier.cpp
-// Description: Cube implementation of a ShapeClassifier.
-// Author:      Ray Smith
-// Created:     Wed Nov 23 10:39:45 PST 2011
-//
-// (C) Copyright 2011, Google Inc.
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-///////////////////////////////////////////////////////////////////////
-
-#include "cubeclassifier.h"
-
-#include "char_altlist.h"
-#include "char_set.h"
-#include "cube_object.h"
-#include "cube_reco_context.h"
-#include "tessclassifier.h"
-#include "tesseractclass.h"
-#include "trainingsample.h"
-#include "unicharset.h"
-
-namespace tesseract {
-
-CubeClassifier::CubeClassifier(tesseract::Tesseract* tesseract)
-    : cube_cntxt_(tesseract->GetCubeRecoContext()),
-      shape_table_(*tesseract->shape_table()) {
-}
-CubeClassifier::~CubeClassifier() {
-}
-
-/// Classifies the given [training] sample, writing to results.
-/// See ShapeClassifier for a full description.
-int CubeClassifier::UnicharClassifySample(
-    const TrainingSample& sample, Pix* page_pix, int debug,
-    UNICHAR_ID keep_this, GenericVector<UnicharRating>* results) {
-  results->clear();
-  if (page_pix == NULL) return 0;
-
-  ASSERT_HOST(cube_cntxt_ != NULL);
-  const TBOX& char_box = sample.bounding_box();
-  CubeObject* cube_obj = new tesseract::CubeObject(
-      cube_cntxt_, page_pix, char_box.left(),
-      pixGetHeight(page_pix) - char_box.top(),
-      char_box.width(), char_box.height());
-  CharAltList* alt_list = cube_obj->RecognizeChar();
-  if (alt_list != NULL) {
-    alt_list->Sort();
-    CharSet* char_set = cube_cntxt_->CharacterSet();
-    for (int i = 0; i < alt_list->AltCount(); ++i) {
-      // Convert cube representation to a shape_id.
-      int alt_id = alt_list->Alt(i);
-      int unichar_id = char_set->UnicharID(char_set->ClassString(alt_id));
-      if (unichar_id >= 0)
-        results->push_back(UnicharRating(unichar_id, alt_list->AltProb(i)));
-    }
-    delete alt_list;
-  }
-  delete cube_obj;
-  return results->size();
-}
-
-/** Provides access to the ShapeTable that this classifier works with. */
-const ShapeTable* CubeClassifier::GetShapeTable() const {
-  return &shape_table_;
-}
-
-CubeTessClassifier::CubeTessClassifier(tesseract::Tesseract* tesseract)
-    : cube_cntxt_(tesseract->GetCubeRecoContext()),
-      shape_table_(*tesseract->shape_table()),
-      pruner_(new TessClassifier(true, tesseract)) {
-}
-CubeTessClassifier::~CubeTessClassifier() {
-  delete pruner_;
-}
-
-/// Classifies the given [training] sample, writing to results.
-/// See ShapeClassifier for a full description.
-int CubeTessClassifier::UnicharClassifySample(
-    const TrainingSample& sample, Pix* page_pix, int debug,
-    UNICHAR_ID keep_this, GenericVector<UnicharRating>* results) {
-  int num_results = pruner_->UnicharClassifySample(sample, page_pix, debug,
-                                                   keep_this, results);
-  if (page_pix == NULL) return num_results;
-
-  ASSERT_HOST(cube_cntxt_ != NULL);
-  const TBOX& char_box = sample.bounding_box();
-  CubeObject* cube_obj = new tesseract::CubeObject(
-      cube_cntxt_, page_pix, char_box.left(),
-      pixGetHeight(page_pix) - char_box.top(),
-      char_box.width(), char_box.height());
-  CharAltList* alt_list = cube_obj->RecognizeChar();
-  CharSet* char_set = cube_cntxt_->CharacterSet();
-  if (alt_list != NULL) {
-    for (int r = 0; r < num_results; ++r) {
-      // Get the best cube probability of the unichar in the result.
-      double best_prob = 0.0;
-      for (int i = 0; i < alt_list->AltCount(); ++i) {
-        int alt_id = alt_list->Alt(i);
-        int unichar_id = char_set->UnicharID(char_set->ClassString(alt_id));
-        if (unichar_id == (*results)[r].unichar_id &&
-            alt_list->AltProb(i) > best_prob) {
-          best_prob = alt_list->AltProb(i);
-        }
-      }
-      (*results)[r].rating = best_prob;
-    }
-    delete alt_list;
-    // Re-sort by rating.
-    results->sort(&UnicharRating::SortDescendingRating);
-  }
-  delete cube_obj;
-  return results->size();
-}
-
-/** Provides access to the ShapeTable that this classifier works with. */
-const ShapeTable* CubeTessClassifier::GetShapeTable() const {
-  return &shape_table_;
-}
-
-}  // namespace tesseract
-
-
-

ccmain/cubeclassifier.h

@@ -1,80 +0,0 @@
-// Copyright 2011 Google Inc. All Rights Reserved.
-// Author: rays@google.com (Ray Smith)
-///////////////////////////////////////////////////////////////////////
-// File:        cubeclassifier.h
-// Description: Cube implementation of a ShapeClassifier.
-// Author:      Ray Smith
-// Created:     Wed Nov 23 10:36:32 PST 2011
-//
-// (C) Copyright 2011, Google Inc.
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-// http://www.apache.org/licenses/LICENSE-2.0
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-//
-///////////////////////////////////////////////////////////////////////
-
-#ifndef THIRD_PARTY_TESSERACT_CCMAIN_CUBECLASSIFIER_H_
-#define THIRD_PARTY_TESSERACT_CCMAIN_CUBECLASSIFIER_H_
-
-#include "shapeclassifier.h"
-
-namespace tesseract {
-
-class Classify;
-class CubeRecoContext;
-class ShapeTable;
-class TessClassifier;
-class Tesseract;
-class TrainingSample;
-struct UnicharRating;
-
-// Cube implementation of a ShapeClassifier.
-class CubeClassifier : public ShapeClassifier {
- public:
-  explicit CubeClassifier(Tesseract* tesseract);
-  virtual ~CubeClassifier();
-
-  // Classifies the given [training] sample, writing to results.
-  // See ShapeClassifier for a full description.
-  virtual int UnicharClassifySample(const TrainingSample& sample, Pix* page_pix,
-                                    int debug, UNICHAR_ID keep_this,
-                                    GenericVector<UnicharRating>* results);
-  // Provides access to the ShapeTable that this classifier works with.
-  virtual const ShapeTable* GetShapeTable() const;
-
- private:
-  // Cube objects.
-  CubeRecoContext* cube_cntxt_;
-  const ShapeTable& shape_table_;
-};
-
-// Combination of Tesseract class pruner with scoring by cube.
-class CubeTessClassifier : public ShapeClassifier {
- public:
-  explicit CubeTessClassifier(Tesseract* tesseract);
-  virtual ~CubeTessClassifier();
-
-  // Classifies the given [training] sample, writing to results.
-  // See ShapeClassifier for a full description.
-  virtual int UnicharClassifySample(const TrainingSample& sample, Pix* page_pix,
-                                    int debug, UNICHAR_ID keep_this,
-                                    GenericVector<UnicharRating>* results);
-  // Provides access to the ShapeTable that this classifier works with.
-  virtual const ShapeTable* GetShapeTable() const;
-
- private:
-  // Cube objects.
-  CubeRecoContext* cube_cntxt_;
-  const ShapeTable& shape_table_;
-  TessClassifier* pruner_;
-};
-
-}  // namespace tesseract
-
-#endif /* THIRD_PARTY_TESSERACT_CCMAIN_CUBECLASSIFIER_H_ */

ccmain/docqual.cpp

@@ -98,8 +98,8 @@ void Tesseract::word_char_quality(WERD_RES *word,
                                   ROW *row,
                                   inT16 *match_count,
                                   inT16 *accepted_match_count) {
-  if (word->bln_boxes == NULL ||
+  if (word->bln_boxes == NULL || word->rebuild_word == NULL ||
-    word->rebuild_word == NULL || word->rebuild_word->blobs.empty()) {
+      word->rebuild_word->blobs.empty()) {
     *match_count = 0;
     *accepted_match_count = 0;
     return;
@@ -132,7 +132,7 @@ inT16 Tesseract::count_outline_errs(char c, inT16 outline_count) {
   int expected_outline_count;
 
   if (STRING (outlines_odd).contains (c))
-    return 0;                    //Don't use this char
+    return 0;  // Don't use this char
   else if (STRING (outlines_2).contains (c))
     expected_outline_count = 2;
   else
@@ -151,7 +151,6 @@ void Tesseract::quality_based_rejection(PAGE_RES_IT &page_res_it,
   }
 }
 
-
 /*************************************************************************
  * unrej_good_quality_words()
  * Accept potential rejects in words which pass the following checks:
@@ -403,7 +402,6 @@ void Tesseract::doc_and_block_rejection(  //reject big chunks
 
 }  // namespace tesseract
 
-
 /*************************************************************************
  * reject_whole_page()
  * Don't believe any of it - set the reject map to 00..00 in all words

ccmain/equationdetect.cpp

@@ -624,10 +624,6 @@ void EquationDetect::IdentifySeedParts() {
 }
 
 float EquationDetect::ComputeForegroundDensity(const TBOX& tbox) {
-#if LIBLEPT_MINOR_VERSION < 69 && LIBLEPT_MAJOR_VERSION <= 1
-  // This will disable the detector because no seed will be identified.
-  return 1.0f;
-#else
   Pix *pix_bi = lang_tesseract_->pix_binary();
   int pix_height = pixGetHeight(pix_bi);
   Box* box = boxCreate(tbox.left(), pix_height - tbox.top(),
@@ -639,7 +635,6 @@ float EquationDetect::ComputeForegroundDensity(const TBOX& tbox) {
   boxDestroy(&box);
 
   return fract;
-#endif
 }
 
 bool EquationDetect::CheckSeedFgDensity(const float density_th,

ccmain/equationdetect.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_EQUATIONDETECT_H__
+#ifndef TESSERACT_CCMAIN_EQUATIONDETECT_H_
-#define TESSERACT_CCMAIN_EQUATIONDETECT_H__
+#define TESSERACT_CCMAIN_EQUATIONDETECT_H_
 
 #include "blobbox.h"
 #include "equationdetectbase.h"

ccmain/fixspace.cpp

@@ -211,7 +211,6 @@ void Tesseract::match_current_words(WERD_RES_LIST &words, ROW *row,
   }
 }
 
-
 /**
  * @name eval_word_spacing()
  * The basic measure is the number of characters in contextually confirmed

ccmain/linerec.cpp

@@ -0,0 +1,333 @@
+///////////////////////////////////////////////////////////////////////
+// File:        linerec.cpp
+// Description: Top-level line-based recognition module for Tesseract.
+// Author:      Ray Smith
+// Created:     Thu May 02 09:47:06 PST 2013
+//
+// (C) Copyright 2013, Google Inc.
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+///////////////////////////////////////////////////////////////////////
+
+#include "tesseractclass.h"
+
+#include "allheaders.h"
+#include "boxread.h"
+#include "imagedata.h"
+#ifndef ANDROID_BUILD
+#include "lstmrecognizer.h"
+#include "recodebeam.h"
+#endif
+#include "ndminx.h"
+#include "pageres.h"
+#include "tprintf.h"
+
+namespace tesseract {
+
+// Arbitarary penalty for non-dictionary words.
+// TODO(rays) How to learn this?
+const float kNonDictionaryPenalty = 5.0f;
+// Scale factor to make certainty more comparable to Tesseract.
+const float kCertaintyScale = 7.0f;
+// Worst acceptable certainty for a dictionary word.
+const float kWorstDictCertainty = -25.0f;
+
+// Generates training data for training a line recognizer, eg LSTM.
+// Breaks the page into lines, according to the boxes, and writes them to a
+// serialized DocumentData based on output_basename.
+void Tesseract::TrainLineRecognizer(const STRING& input_imagename,
+                                    const STRING& output_basename,
+                                    BLOCK_LIST *block_list) {
+  STRING lstmf_name = output_basename + ".lstmf";
+  DocumentData images(lstmf_name);
+  if (applybox_page > 0) {
+    // Load existing document for the previous pages.
+    if (!images.LoadDocument(lstmf_name.string(), "eng", 0, 0, NULL)) {
+      tprintf("Failed to read training data from %s!\n", lstmf_name.string());
+      return;
+    }
+  }
+  GenericVector<TBOX> boxes;
+  GenericVector<STRING> texts;
+  // Get the boxes for this page, if there are any.
+  if (!ReadAllBoxes(applybox_page, false, input_imagename, &boxes, &texts, NULL,
+                    NULL) ||
+      boxes.empty()) {
+    tprintf("Failed to read boxes from %s\n", input_imagename.string());
+    return;
+  }
+  TrainFromBoxes(boxes, texts, block_list, &images);
+  images.Shuffle();
+  if (!images.SaveDocument(lstmf_name.string(), NULL)) {
+    tprintf("Failed to write training data to %s!\n", lstmf_name.string());
+  }
+}
+
+// Generates training data for training a line recognizer, eg LSTM.
+// Breaks the boxes into lines, normalizes them, converts to ImageData and
+// appends them to the given training_data.
+void Tesseract::TrainFromBoxes(const GenericVector<TBOX>& boxes,
+                               const GenericVector<STRING>& texts,
+                               BLOCK_LIST *block_list,
+                               DocumentData* training_data) {
+  int box_count = boxes.size();
+  // Process all the text lines in this page, as defined by the boxes.
+  int end_box = 0;
+  // Don't let \t, which marks newlines in the box file, get into the line
+  // content, as that makes the line unusable in training.
+  while (end_box < texts.size() && texts[end_box] == "\t") ++end_box;
+  for (int start_box = end_box; start_box < box_count; start_box = end_box) {
+    // Find the textline of boxes starting at start and their bounding box.
+    TBOX line_box = boxes[start_box];
+    STRING line_str = texts[start_box];
+    for (end_box = start_box + 1; end_box < box_count && texts[end_box] != "\t";
+         ++end_box) {
+      line_box += boxes[end_box];
+      line_str += texts[end_box];
+    }
+    // Find the most overlapping block.
+    BLOCK* best_block = NULL;
+    int best_overlap = 0;
+    BLOCK_IT b_it(block_list);
+    for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
+      BLOCK* block = b_it.data();
+      if (block->poly_block() != NULL && !block->poly_block()->IsText())
+        continue;  // Not a text block.
+      TBOX block_box = block->bounding_box();
+      block_box.rotate(block->re_rotation());
+      if (block_box.major_overlap(line_box)) {
+        TBOX overlap_box = line_box.intersection(block_box);
+        if (overlap_box.area() > best_overlap) {
+          best_overlap = overlap_box.area();
+          best_block = block;
+        }
+      }
+    }
+    ImageData* imagedata = NULL;
+    if (best_block == NULL) {
+      tprintf("No block overlapping textline: %s\n", line_str.string());
+    } else {
+      imagedata = GetLineData(line_box, boxes, texts, start_box, end_box,
+                              *best_block);
+    }
+    if (imagedata != NULL)
+      training_data->AddPageToDocument(imagedata);
+    // Don't let \t, which marks newlines in the box file, get into the line
+    // content, as that makes the line unusable in training.
+    while (end_box < texts.size() && texts[end_box] == "\t") ++end_box;
+  }
+}
+
+// Returns an Imagedata containing the image of the given box,
+// and ground truth boxes/truth text if available in the input.
+// The image is not normalized in any way.
+ImageData* Tesseract::GetLineData(const TBOX& line_box,
+                                  const GenericVector<TBOX>& boxes,
+                                  const GenericVector<STRING>& texts,
+                                  int start_box, int end_box,
+                                  const BLOCK& block) {
+  TBOX revised_box;
+  ImageData* image_data = GetRectImage(line_box, block, kImagePadding,
+                                       &revised_box);
+  if (image_data == NULL) return NULL;
+  image_data->set_page_number(applybox_page);
+  // Copy the boxes and shift them so they are relative to the image.
+  FCOORD block_rotation(block.re_rotation().x(), -block.re_rotation().y());
+  ICOORD shift = -revised_box.botleft();
+  GenericVector<TBOX> line_boxes;
+  GenericVector<STRING> line_texts;
+  for (int b = start_box; b < end_box; ++b) {
+    TBOX box = boxes[b];
+    box.rotate(block_rotation);
+    box.move(shift);
+    line_boxes.push_back(box);
+    line_texts.push_back(texts[b]);
+  }
+  GenericVector<int> page_numbers;
+  page_numbers.init_to_size(line_boxes.size(), applybox_page);
+  image_data->AddBoxes(line_boxes, line_texts, page_numbers);
+  return image_data;
+}
+
+// Helper gets the image of a rectangle, using the block.re_rotation() if
+// needed to get to the image, and rotating the result back to horizontal
+// layout. (CJK characters will be on their left sides) The vertical text flag
+// is set in the returned ImageData if the text was originally vertical, which
+// can be used to invoke a different CJK recognition engine. The revised_box
+// is also returned to enable calculation of output bounding boxes.
+ImageData* Tesseract::GetRectImage(const TBOX& box, const BLOCK& block,
+                                   int padding, TBOX* revised_box) const {
+  TBOX wbox = box;
+  wbox.pad(padding, padding);
+  *revised_box = wbox;
+  // Number of clockwise 90 degree rotations needed to get back to tesseract
+  // coords from the clipped image.
+  int num_rotations = 0;
+  if (block.re_rotation().y() > 0.0f)
+    num_rotations = 1;
+  else if (block.re_rotation().x() < 0.0f)
+    num_rotations = 2;
+  else if (block.re_rotation().y() < 0.0f)
+    num_rotations = 3;
+  // Handle two cases automatically: 1 the box came from the block, 2 the box
+  // came from a box file, and refers to the image, which the block may not.
+  if (block.bounding_box().major_overlap(*revised_box))
+    revised_box->rotate(block.re_rotation());
+  // Now revised_box always refers to the image.
+  // BestPix is never colormapped, but may be of any depth.
+  Pix* pix = BestPix();
+  int width = pixGetWidth(pix);
+  int height = pixGetHeight(pix);
+  TBOX image_box(0, 0, width, height);
+  // Clip to image bounds;
+  *revised_box &= image_box;
+  if (revised_box->null_box()) return NULL;
+  Box* clip_box = boxCreate(revised_box->left(), height - revised_box->top(),
+                            revised_box->width(), revised_box->height());
+  Pix* box_pix = pixClipRectangle(pix, clip_box, NULL);
+  if (box_pix == NULL) return NULL;
+  boxDestroy(&clip_box);
+  if (num_rotations > 0) {
+    Pix* rot_pix = pixRotateOrth(box_pix, num_rotations);
+    pixDestroy(&box_pix);
+    box_pix = rot_pix;
+  }
+  // Convert sub-8-bit images to 8 bit.
+  int depth = pixGetDepth(box_pix);
+  if (depth < 8) {
+    Pix* grey;
+    grey = pixConvertTo8(box_pix, false);
+    pixDestroy(&box_pix);
+    box_pix = grey;
+  }
+  bool vertical_text = false;
+  if (num_rotations > 0) {
+    // Rotated the clipped revised box back to internal coordinates.
+    FCOORD rotation(block.re_rotation().x(), -block.re_rotation().y());
+    revised_box->rotate(rotation);
+    if (num_rotations != 2)
+      vertical_text = true;
+  }
+  return new ImageData(vertical_text, box_pix);
+}
+
+#ifndef ANDROID_BUILD
+// Recognizes a word or group of words, converting to WERD_RES in *words.
+// Analogous to classify_word_pass1, but can handle a group of words as well.
+void Tesseract::LSTMRecognizeWord(const BLOCK& block, ROW *row, WERD_RES *word,
+                                  PointerVector<WERD_RES>* words) {
+  TBOX word_box = word->word->bounding_box();
+  // Get the word image - no frills.
+  if (tessedit_pageseg_mode == PSM_SINGLE_WORD ||
+      tessedit_pageseg_mode == PSM_RAW_LINE) {
+    // In single word mode, use the whole image without any other row/word
+    // interpretation.
+    word_box = TBOX(0, 0, ImageWidth(), ImageHeight());
+  } else {
+    float baseline = row->base_line((word_box.left() + word_box.right()) / 2);
+    if (baseline + row->descenders() < word_box.bottom())
+      word_box.set_bottom(baseline + row->descenders());
+    if (baseline + row->x_height() + row->ascenders() > word_box.top())
+      word_box.set_top(baseline + row->x_height() + row->ascenders());
+  }
+  ImageData* im_data = GetRectImage(word_box, block, kImagePadding, &word_box);
+  if (im_data == NULL) return;
+  lstm_recognizer_->RecognizeLine(*im_data, true, classify_debug_level > 0,
+                                  kWorstDictCertainty / kCertaintyScale,
+                                  lstm_use_matrix, &unicharset, word_box, 2.0,
+                                  false, words);
+  delete im_data;
+  SearchWords(words);
+}
+
+// Apply segmentation search to the given set of words, within the constraints
+// of the existing ratings matrix. If there is already a best_choice on a word
+// leaves it untouched and just sets the done/accepted etc flags.
+void Tesseract::SearchWords(PointerVector<WERD_RES>* words) {
+  // Run the segmentation search on the network outputs and make a BoxWord
+  // for each of the output words.
+  // If we drop a word as junk, then there is always a space in front of the
+  // next.
+  const Dict* stopper_dict = lstm_recognizer_->GetDict();
+  if (stopper_dict == nullptr) stopper_dict = &getDict();
+  bool any_nonspace_delimited = false;
+  for (int w = 0; w < words->size(); ++w) {
+    WERD_RES* word = (*words)[w];
+    if (word->best_choice != nullptr &&
+        word->best_choice->ContainsAnyNonSpaceDelimited()) {
+      any_nonspace_delimited = true;
+      break;
+    }
+  }
+  for (int w = 0; w < words->size(); ++w) {
+    WERD_RES* word = (*words)[w];
+    if (word->best_choice == NULL) {
+      // If we are using the beam search, the unicharset had better match!
+      word->SetupWordScript(unicharset);
+      WordSearch(word);
+    } else if (word->best_choice->unicharset() == &unicharset &&
+               !lstm_recognizer_->IsRecoding()) {
+      // We set up the word without using the dictionary, so set the permuter
+      // now, but we can only do it because the unicharsets match.
+      word->best_choice->set_permuter(
+          getDict().valid_word(*word->best_choice, true));
+    }
+    if (word->best_choice == NULL) {
+      // It is a dud.
+      word->SetupFake(lstm_recognizer_->GetUnicharset());
+    } else {
+      // Set the best state.
+      for (int i = 0; i < word->best_choice->length(); ++i) {
+        int length = word->best_choice->state(i);
+        word->best_state.push_back(length);
+      }
+      word->reject_map.initialise(word->best_choice->length());
+      word->tess_failed = false;
+      word->tess_accepted = true;
+      word->tess_would_adapt = false;
+      word->done = true;
+      word->tesseract = this;
+      float word_certainty = MIN(word->space_certainty,
+                                 word->best_choice->certainty());
+      word_certainty *= kCertaintyScale;
+      // Arbitrary ding factor for non-dictionary words.
+      if (!lstm_recognizer_->IsRecoding() &&
+          !Dict::valid_word_permuter(word->best_choice->permuter(), true))
+        word_certainty -= kNonDictionaryPenalty;
+      if (getDict().stopper_debug_level >= 1) {
+        tprintf("Best choice certainty=%g, space=%g, scaled=%g, final=%g\n",
+                word->best_choice->certainty(), word->space_certainty,
+                MIN(word->space_certainty, word->best_choice->certainty()) *
+                    kCertaintyScale,
+                word_certainty);
+        word->best_choice->print();
+      }
+      word->best_choice->set_certainty(word_certainty);
+      // Discard words that are impossibly bad, but allow a bit more for
+      // dictionary words, and keep bad words in non-space-delimited langs.
+      if (word_certainty >= RecodeBeamSearch::kMinCertainty ||
+          any_nonspace_delimited ||
+          (word_certainty >= kWorstDictCertainty &&
+           Dict::valid_word_permuter(word->best_choice->permuter(), true))) {
+        word->tess_accepted = stopper_dict->AcceptableResult(word);
+      } else {
+        if (getDict().stopper_debug_level >= 1) {
+          tprintf("Deleting word with certainty %g\n", word_certainty);
+          word->best_choice->print();
+        }
+        // It is a dud.
+        word->SetupFake(lstm_recognizer_->GetUnicharset());
+      }
+    }
+  }
+}
+#endif  // ANDROID_BUILD
+
+}  // namespace tesseract.

ccmain/ltrresultiterator.cpp

@@ -145,11 +145,10 @@ float LTRResultIterator::Confidence(PageIteratorLevel level) const {
   return 0.0f;
 }
 
-void LTRResultIterator::RowAttributes(float* row_height,
+void LTRResultIterator::RowAttributes(float* row_height, float* descenders,
-                                      float* descenders,
                                       float* ascenders) const {
-    *row_height = it_->row()->row->x_height() + it_->row()-> row->ascenders()
+  *row_height = it_->row()->row->x_height() + it_->row()->row->ascenders() -
-                  - it_->row()->row->descenders();
+                it_->row()->row->descenders();
   *descenders = it_->row()->row->descenders();
   *ascenders = it_->row()->row->ascenders();
 }
@@ -221,6 +220,12 @@ bool LTRResultIterator::WordIsFromDictionary() const {
          permuter == USER_DAWG_PERM;
 }
 
+// Returns the number of blanks before the current word.
+int LTRResultIterator::BlanksBeforeWord() const {
+  if (it_->word() == NULL) return 1;
+  return it_->word()->word->space();
+}
+
 // Returns true if the current word is numeric.
 bool LTRResultIterator::WordIsNumeric() const {
   if (it_->word() == NULL) return false;  // Already at the end!

ccmain/ltrresultiterator.h

@@ -18,8 +18,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H__
+#ifndef TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H_
-#define TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H__
+#define TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H_
 
 #include "platform.h"
 #include "pageiterator.h"
@@ -92,8 +92,7 @@ class TESS_API LTRResultIterator : public PageIterator {
   float Confidence(PageIteratorLevel level) const;
 
   // Returns the attributes of the current row.
-  void RowAttributes(float* row_height,
+  void RowAttributes(float* row_height, float* descenders,
-                     float* descenders,
                      float* ascenders) const;
 
   // ============= Functions that refer to words only ============.
@@ -125,6 +124,9 @@ class TESS_API LTRResultIterator : public PageIterator {
   // Returns true if the current word was found in a dictionary.
   bool WordIsFromDictionary() const;
 
+  // Returns the number of blanks before the current word.
+  int BlanksBeforeWord() const;
+
   // Returns true if the current word is numeric.
   bool WordIsNumeric() const;
 
@@ -216,4 +218,4 @@ class ChoiceIterator {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H__
+#endif  // TESSERACT_CCMAIN_LTR_RESULT_ITERATOR_H_

ccmain/mutableiterator.h

@@ -18,8 +18,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_MUTABLEITERATOR_H__
+#ifndef TESSERACT_CCMAIN_MUTABLEITERATOR_H_
-#define TESSERACT_CCMAIN_MUTABLEITERATOR_H__
+#define TESSERACT_CCMAIN_MUTABLEITERATOR_H_
 
 #include "resultiterator.h"
 
@@ -61,4 +61,4 @@ class MutableIterator : public ResultIterator {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_MUTABLEITERATOR_H__
+#endif  // TESSERACT_CCMAIN_MUTABLEITERATOR_H_

ccmain/osdetect.cpp

@@ -164,13 +164,19 @@ void remove_nontext_regions(tesseract::Tesseract *tess, BLOCK_LIST *blocks,
   int vertical_y = 1;
   tesseract::TabVector_LIST v_lines;
   tesseract::TabVector_LIST h_lines;
-  int resolution = (kMinCredibleResolution > pixGetXRes(pix)) ?
+  int resolution;
-      kMinCredibleResolution : pixGetXRes(pix);
+  if (kMinCredibleResolution > pixGetXRes(pix)) {
+    resolution = kMinCredibleResolution;
+    tprintf("Warning. Invalid resolution %d dpi. Using %d instead.\n",
+            pixGetXRes(pix), resolution);
+  } else {
+    resolution = pixGetXRes(pix);
+  }
 
   tesseract::LineFinder::FindAndRemoveLines(resolution, false, pix,
                                             &vertical_x, &vertical_y,
                                             NULL, &v_lines, &h_lines);
-  Pix* im_pix = tesseract::ImageFind::FindImages(pix);
+  Pix* im_pix = tesseract::ImageFind::FindImages(pix, nullptr);
   if (im_pix != NULL) {
     pixSubtract(pix, pix, im_pix);
     pixDestroy(&im_pix);

ccmain/osdetect.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_OSDETECT_H__
+#ifndef TESSERACT_CCMAIN_OSDETECT_H_
-#define TESSERACT_CCMAIN_OSDETECT_H__
+#define TESSERACT_CCMAIN_OSDETECT_H_
 
 #include "strngs.h"
 #include "unicharset.h"
@@ -135,4 +135,4 @@ bool os_detect_blob(BLOBNBOX* bbox, OrientationDetector* o,
 // applied for the text to be upright (readable).
 TESS_API int OrientationIdToValue(const int& id);
 
-#endif  // TESSERACT_CCMAIN_OSDETECT_H__
+#endif  // TESSERACT_CCMAIN_OSDETECT_H_

ccmain/output.cpp

@@ -78,17 +78,15 @@ void Tesseract::output_pass(  //Tess output pass //send to api
   while (page_res_it.word () != NULL) {
     check_debug_pt (page_res_it.word (), 120);
 
-	if (target_word_box)
+    if (target_word_box) {
-	{
+      TBOX current_word_box = page_res_it.word()->word->bounding_box();
-
+      FCOORD center_pt(
-		TBOX current_word_box=page_res_it.word ()->word->bounding_box();
+          (current_word_box.right() + current_word_box.left()) / 2,
-		FCOORD center_pt((current_word_box.right()+current_word_box.left())/2,(current_word_box.bottom()+current_word_box.top())/2);
+          (current_word_box.bottom() + current_word_box.top()) / 2);
-		if (!target_word_box->contains(center_pt))
+      if (!target_word_box->contains(center_pt)) {
-		{
+        page_res_it.forward();
-			page_res_it.forward ();
         continue;
       }
-
     }
     if (tessedit_write_block_separators &&
     block_of_last_word != page_res_it.block ()) {
@@ -337,7 +335,7 @@ void Tesseract::set_unlv_suspects(WERD_RES *word_res) {
   rating_per_ch = word.rating() / word_res->reject_map.length();
 
   if (rating_per_ch >= suspect_rating_per_ch)
-    return;                      //Don't touch bad ratings
+    return;  // Don't touch bad ratings
 
   if ((word_res->tess_accepted) || (rating_per_ch < suspect_accept_rating)) {
     /* Unreject any Tess Acceptable word - but NOT tess reject chs*/

ccmain/pageiterator.cpp

@@ -87,7 +87,7 @@ const PageIterator& PageIterator::operator=(const PageIterator& src) {
   rect_top_ = src.rect_top_;
   rect_width_ = src.rect_width_;
   rect_height_ = src.rect_height_;
-  if (it_ != NULL) delete it_;
+  delete it_;
   it_ = new PAGE_RES_IT(*src.it_);
   BeginWord(src.blob_index_);
   return *this;
@@ -597,10 +597,8 @@ void PageIterator::BeginWord(int offset) {
     }
     word_ = NULL;
     // We will be iterating the box_word.
-    if (cblob_it_ != NULL) {
     delete cblob_it_;
     cblob_it_ = NULL;
-    }
   } else {
     // No recognition yet, so a "symbol" is a cblob.
     word_ = word_res->word;

ccmain/pageiterator.h

@@ -18,8 +18,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_PAGEITERATOR_H__
+#ifndef TESSERACT_CCMAIN_PAGEITERATOR_H_
-#define TESSERACT_CCMAIN_PAGEITERATOR_H__
+#define TESSERACT_CCMAIN_PAGEITERATOR_H_
 
 #include "publictypes.h"
 #include "platform.h"
@@ -361,4 +361,4 @@ class TESS_API PageIterator {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_PAGEITERATOR_H__
+#endif  // TESSERACT_CCMAIN_PAGEITERATOR_H_

ccmain/pagesegmain.cpp

@@ -18,9 +18,6 @@
  **********************************************************************/
 
 #ifdef _WIN32
-#ifndef __GNUC__
-#include <windows.h>
-#endif  // __GNUC__
 #ifndef unlink
 #include <io.h>
 #endif
@@ -40,6 +37,7 @@
 #include "blobbox.h"
 #include "blread.h"
 #include "colfind.h"
+#include "debugpixa.h"
 #include "equationdetect.h"
 #include "imagefind.h"
 #include "linefind.h"
@@ -179,28 +177,6 @@ int Tesseract::SegmentPage(const STRING* input_file, BLOCK_LIST* blocks,
   return auto_page_seg_ret_val;
 }
 
-// Helper writes a grey image to a file for use by scrollviewer.
-// Normally for speed we don't display the image in the layout debug windows.
-// If textord_debug_images is true, we draw the image as a background to some
-// of the debug windows. printable determines whether these
-// images are optimized for printing instead of screen display.
-static void WriteDebugBackgroundImage(bool printable, Pix* pix_binary) {
-  Pix* grey_pix = pixCreate(pixGetWidth(pix_binary),
-                            pixGetHeight(pix_binary), 8);
-  // Printable images are light grey on white, but for screen display
-  // they are black on dark grey so the other colors show up well.
-  if (printable) {
-    pixSetAll(grey_pix);
-    pixSetMasked(grey_pix, pix_binary, 192);
-  } else {
-    pixSetAllArbitrary(grey_pix, 64);
-    pixSetMasked(grey_pix, pix_binary, 0);
-  }
-  AlignedBlob::IncrementDebugPix();
-  pixWrite(AlignedBlob::textord_debug_pix().string(), grey_pix, IFF_PNG);
-  pixDestroy(&grey_pix);
-}
-
 /**
  * Auto page segmentation. Divide the page image into blocks of uniform
  * text linespacing and images.
@@ -229,9 +205,6 @@ int Tesseract::AutoPageSeg(PageSegMode pageseg_mode, BLOCK_LIST* blocks,
                            TO_BLOCK_LIST* to_blocks,
                            BLOBNBOX_LIST* diacritic_blobs, Tesseract* osd_tess,
                            OSResults* osr) {
-  if (textord_debug_images) {
-    WriteDebugBackgroundImage(textord_debug_printable, pix_binary_);
-  }
   Pix* photomask_pix = NULL;
   Pix* musicmask_pix = NULL;
   // The blocks made by the ColumnFinder. Moved to blocks before return.
@@ -253,9 +226,10 @@ int Tesseract::AutoPageSeg(PageSegMode pageseg_mode, BLOCK_LIST* blocks,
     if (equ_detect_) {
       finder->SetEquationDetect(equ_detect_);
     }
-    result = finder->FindBlocks(
+    result = finder->FindBlocks(pageseg_mode, scaled_color_, scaled_factor_,
-        pageseg_mode, scaled_color_, scaled_factor_, to_block, photomask_pix,
+                                to_block, photomask_pix, pix_thresholds_,
-        pix_thresholds_, pix_grey_, &found_blocks, diacritic_blobs, to_blocks);
+                                pix_grey_, &pixa_debug_, &found_blocks,
+                                diacritic_blobs, to_blocks);
     if (result >= 0)
       finder->GetDeskewVectors(&deskew_, &reskew_);
     delete finder;
@@ -268,11 +242,6 @@ int Tesseract::AutoPageSeg(PageSegMode pageseg_mode, BLOCK_LIST* blocks,
   BLOCK_IT block_it(blocks);
   // Move the found blocks to the input/output blocks.
   block_it.add_list_after(&found_blocks);
-
-  if (textord_debug_images) {
-    // The debug image is no longer needed so delete it.
-    unlink(AlignedBlob::textord_debug_pix().string());
-  }
   return result;
 }
 
@@ -314,19 +283,21 @@ ColumnFinder* Tesseract::SetupPageSegAndDetectOrientation(
 
   ASSERT_HOST(pix_binary_ != NULL);
   if (tessedit_dump_pageseg_images) {
-    pixWrite("tessinput.png", pix_binary_, IFF_PNG);
+    pixa_debug_.AddPix(pix_binary_, "PageSegInput");
   }
   // Leptonica is used to find the rule/separator lines in the input.
   LineFinder::FindAndRemoveLines(source_resolution_,
                                  textord_tabfind_show_vlines, pix_binary_,
                                  &vertical_x, &vertical_y, music_mask_pix,
                                  &v_lines, &h_lines);
-  if (tessedit_dump_pageseg_images)
+  if (tessedit_dump_pageseg_images) {
-    pixWrite("tessnolines.png", pix_binary_, IFF_PNG);
+    pixa_debug_.AddPix(pix_binary_, "NoLines");
+  }
   // Leptonica is used to find a mask of the photo regions in the input.
-  *photo_mask_pix = ImageFind::FindImages(pix_binary_);
+  *photo_mask_pix = ImageFind::FindImages(pix_binary_, &pixa_debug_);
-  if (tessedit_dump_pageseg_images)
+  if (tessedit_dump_pageseg_images) {
-    pixWrite("tessnoimages.png", pix_binary_, IFF_PNG);
+    pixa_debug_.AddPix(pix_binary_, "NoImages");
+  }
   if (!PSM_COL_FIND_ENABLED(pageseg_mode)) v_lines.clear();
 
   // The rest of the algorithm uses the usual connected components.
@@ -412,7 +383,8 @@ ColumnFinder* Tesseract::SetupPageSegAndDetectOrientation(
                   "Don't rotate.\n", osd_margin);
           osd_orientation = 0;
         } else {
-          tprintf("OSD: Weak margin (%.2f) for %d blob text block, "
+          tprintf(
+              "OSD: Weak margin (%.2f) for %d blob text block, "
               "but using orientation anyway: %d\n",
               osd_margin, osd_blobs.length(), osd_orientation);
         }

ccmain/par_control.cpp

@@ -18,9 +18,9 @@
 ///////////////////////////////////////////////////////////////////////
 
 #include "tesseractclass.h"
-#ifdef OPENMP
+#ifdef _OPENMP
 #include <omp.h>
-#endif  // OPENMP
+#endif  // _OPENMP
 
 namespace tesseract {
 
@@ -53,7 +53,9 @@ void Tesseract::PrerecAllWordsPar(const GenericVector<WordData>& words) {
   }
   // Pre-classify all the blobs.
   if (tessedit_parallelize > 1) {
-    #pragma omp parallel for num_threads(10)
+#ifdef _OPENMP
+#pragma omp parallel for num_threads(10)
+#endif  // _OPENMP
     for (int b = 0; b < blobs.size(); ++b) {
       *blobs[b].choices =
           blobs[b].tesseract->classify_blob(blobs[b].blob, "par", White, NULL);

ccmain/paragraphs.cpp

@@ -2052,7 +2052,7 @@ void ConvertHypothesizedModelRunsToParagraphs(
     bool single_line_paragraph = false;
     SetOfModels models;
     rows[start].NonNullHypotheses(&models);
-    if (models.size() > 0) {
+    if (!models.empty()) {
       model = models[0];
       if (rows[start].GetLineType(model) != LT_BODY)
         single_line_paragraph = true;
@@ -2113,6 +2113,7 @@ void ConvertHypothesizedModelRunsToParagraphs(
       if ((*row_owners)[row] != NULL) {
         tprintf("Memory leak! ConvertHypothesizeModelRunsToParagraphs() called "
                 "more than once!\n");
+        delete (*row_owners)[row];
       }
       (*row_owners)[row] = p;
     }
@@ -2189,17 +2190,17 @@ void LeftoverSegments(const GenericVector<RowScratchRegisters> &rows,
     SetOfModels models_w_crowns;
     rows[i].StrongHypotheses(&models);
     rows[i].NonNullHypotheses(&models_w_crowns);
-    if (models.empty() && models_w_crowns.size() > 0) {
+    if (models.empty() && !models_w_crowns.empty()) {
       // Crown paragraph.  Is it followed by a modeled line?
       for (int end = i + 1; end < rows.size(); end++) {
         SetOfModels end_models;
         SetOfModels strong_end_models;
         rows[end].NonNullHypotheses(&end_models);
         rows[end].StrongHypotheses(&strong_end_models);
-        if (end_models.size() == 0) {
+        if (end_models.empty()) {
           needs_fixing = true;
           break;
-        } else if (strong_end_models.size() > 0) {
+        } else if (!strong_end_models.empty()) {
           needs_fixing = false;
           break;
         }
@@ -2484,7 +2485,7 @@ void InitializeRowInfo(bool after_recognition,
   info->ltr = ltr >= rtl;
   info->has_leaders = num_leaders > 3;
   info->num_words = werds.size();
-  if (werds.size() > 0) {
+  if (!werds.empty()) {
     WERD_RES *lword = werds[0], *rword = werds[werds.size() - 1];
     info->lword_text = lword->best_choice->unichar_string().string();
     info->rword_text = rword->best_choice->unichar_string().string();
@@ -2537,7 +2538,7 @@ void DetectParagraphs(int debug_level,
 
   // If we're called before text recognition, we might not have
   // tight block bounding boxes, so trim by the minimum on each side.
-  if (row_infos.size() > 0) {
+  if (!row_infos.empty()) {
     int min_lmargin = row_infos[0].pix_ldistance;
     int min_rmargin = row_infos[0].pix_rdistance;
     for (int i = 1; i < row_infos.size(); i++) {

ccmain/paramsd.cpp

@@ -329,13 +329,19 @@ void ParamsEditor::WriteParams(char *filename,
     fclose(fp);
     sprintf (msg_str, "Overwrite file " "%s" "? (Y/N)", filename);
     int a = sv_window_->ShowYesNoDialog(msg_str);
-    if (a == 'n') { return; }  // don't write
+    if (a == 'n') {
+      return;
+    }  // don't write
   }
 
 
   fp = fopen (filename, "wb");  // can we write to it?
   if (fp == NULL) {
-    sv_window_->AddMessage("Can't write to file " "%s" "", filename);
+    sv_window_->AddMessage(
+        "Can't write to file "
+        "%s"
+        "",
+        filename);
     return;
   }
 

ccmain/paramsd.h

@@ -19,14 +19,12 @@
 //
 // Tesseract parameter editor is used to edit all the parameters used
 // within tesseract from the ui.
+#ifndef TESSERACT_CCMAIN_PARAMSD_H_
+#define TESSERACT_CCMAIN_PARAMSD_H_
+
 #ifndef GRAPHICS_DISABLED
-#ifndef VARABLED_H
-#define VARABLED_H
 
 #include "elst.h"
-#ifndef NO_CUBE_BUILD
-#include "scrollview.h"
-#endif
 #include "params.h"
 #include "tesseractclass.h"
 
@@ -122,5 +120,5 @@ class ParamsEditor : public SVEventHandler {
   ScrollView* sv_window_;
 };
 
-#endif
+#endif  // GRAPHICS_DISABLED
-#endif
+#endif  // TESSERACT_CCMAIN_PARAMSD_H_

ccmain/pgedit.cpp

@@ -191,7 +191,7 @@ ScrollView* bln_word_window_handle() {  // return handle
  */
 
 void build_image_window(int width, int height) {
-  if (image_win != NULL) { delete image_win; }
+  delete image_win;
   image_win = new ScrollView(editor_image_win_name.string(),
                              editor_image_xpos, editor_image_ypos,
                              width + 1,

ccmain/resultiterator.h

@@ -19,8 +19,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_RESULT_ITERATOR_H__
+#ifndef TESSERACT_CCMAIN_RESULT_ITERATOR_H_
-#define TESSERACT_CCMAIN_RESULT_ITERATOR_H__
+#define TESSERACT_CCMAIN_RESULT_ITERATOR_H_
 
 #include "platform.h"
 #include "ltrresultiterator.h"
@@ -241,4 +241,4 @@ class TESS_API ResultIterator : public LTRResultIterator {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_RESULT_ITERATOR_H__
+#endif  // TESSERACT_CCMAIN_RESULT_ITERATOR_H_

ccmain/tessedit.cpp

@@ -40,11 +40,14 @@
 #include "efio.h"
 #include "danerror.h"
 #include "globals.h"
+#ifndef ANDROID_BUILD
+#include "lstmrecognizer.h"
+#endif
 #include "tesseractclass.h"
 #include "params.h"
 
 #define VARDIR        "configs/" /*variables files */
-                                 //config under api
+                                 // config under api
 #define API_CONFIG      "configs/api_config"
 
 ETEXT_DESC *global_monitor = NULL;  // progress monitor
@@ -89,8 +92,8 @@ bool Tesseract::init_tesseract_lang_data(
     const char *arg0, const char *textbase, const char *language,
     OcrEngineMode oem, char **configs, int configs_size,
     const GenericVector<STRING> *vars_vec,
-    const GenericVector<STRING> *vars_values,
+    const GenericVector<STRING> *vars_values, bool set_only_non_debug_params,
-    bool set_only_non_debug_params) {
+    TessdataManager *mgr) {
   // Set the basename, compute the data directory.
   main_setup(arg0, textbase);
 
@@ -102,20 +105,39 @@ bool Tesseract::init_tesseract_lang_data(
 
   // Initialize TessdataManager.
   STRING tessdata_path = language_data_path_prefix + kTrainedDataSuffix;
-  if (!tessdata_manager.Init(tessdata_path.string(),
+  if (!mgr->is_loaded() && !mgr->Init(tessdata_path.string())) {
-                             tessdata_manager_debug_level)) {
+    // Try without tessdata.
+    m_data_sub_dir.set_value("");
+    main_setup(arg0, textbase);
+    language_data_path_prefix = datadir;
+    language_data_path_prefix += lang;
+    language_data_path_prefix += ".";
+    tessdata_path = language_data_path_prefix + kTrainedDataSuffix;
+    if (!mgr->Init(tessdata_path.string())) {
+      tprintf("Error opening data file %s\n", tessdata_path.string());
+      tprintf(
+          "Please make sure the TESSDATA_PREFIX environment variable is set"
+          " to your \"tessdata\" directory.\n");
       return false;
     }
+  }
+  if (oem == OEM_DEFAULT) {
+    // Set the engine mode from availability, which can then be overidden by
+    // the config file when we read it below.
+    if (!mgr->IsLSTMAvailable()) {
+      tessedit_ocr_engine_mode.set_value(OEM_TESSERACT_ONLY);
+    } else if (!mgr->IsBaseAvailable()) {
+      tessedit_ocr_engine_mode.set_value(OEM_LSTM_ONLY);
+    } else {
+      tessedit_ocr_engine_mode.set_value(OEM_TESSERACT_LSTM_COMBINED);
+    }
+  }
 
   // If a language specific config file (lang.config) exists, load it in.
-  if (tessdata_manager.SeekToStart(TESSDATA_LANG_CONFIG)) {
+  TFile fp;
-    ParamUtils::ReadParamsFromFp(
+  if (mgr->GetComponent(TESSDATA_LANG_CONFIG, &fp)) {
-        tessdata_manager.GetDataFilePtr(),
+    ParamUtils::ReadParamsFromFp(SET_PARAM_CONSTRAINT_NONE, &fp,
-        tessdata_manager.GetEndOffset(TESSDATA_LANG_CONFIG),
+                                 this->params());
-        SET_PARAM_CONSTRAINT_NONE, this->params());
-    if (tessdata_manager_debug_level) {
-      tprintf("Loaded language config file\n");
-    }
   }
 
   SetParamConstraint set_params_constraint = set_only_non_debug_params ?
@@ -145,10 +167,6 @@ bool Tesseract::init_tesseract_lang_data(
     if (params_file != NULL) {
       ParamUtils::PrintParams(params_file, this->params());
       fclose(params_file);
-      if (tessdata_manager_debug_level > 0) {
-        tprintf("Wrote parameters to %s\n",
-                tessedit_write_params_to_file.string());
-      }
     } else {
       tprintf("Failed to open %s for writing params.\n",
               tessedit_write_params_to_file.string());
@@ -157,30 +175,48 @@ bool Tesseract::init_tesseract_lang_data(
 
   // Determine which ocr engine(s) should be loaded and used for recognition.
   if (oem != OEM_DEFAULT) tessedit_ocr_engine_mode.set_value(oem);
-  if (tessdata_manager_debug_level) {
-    tprintf("Loading Tesseract/Cube with tessedit_ocr_engine_mode %d\n",
-            static_cast<int>(tessedit_ocr_engine_mode));
-  }
 
   // If we are only loading the config file (and so not planning on doing any
   // recognition) then there's nothing else do here.
   if (tessedit_init_config_only) {
-    if (tessdata_manager_debug_level) {
-      tprintf("Returning after loading config file\n");
-    }
     return true;
   }
 
+// The various OcrEngineMode settings (see publictypes.h) determine which
+// engine-specific data files need to be loaded.
+// If LSTM_ONLY is requested, the base Tesseract files are *Not* required.
+#ifndef ANDROID_BUILD
+  if (tessedit_ocr_engine_mode == OEM_LSTM_ONLY ||
+      tessedit_ocr_engine_mode == OEM_TESSERACT_LSTM_COMBINED) {
+    if (mgr->swap()) {
+      tprintf("Error: LSTM requested on big-endian hardware!!\n");
+      tprintf("Big-endian not yet supported! Loading tesseract.\n");
+      tessedit_ocr_engine_mode.set_value(OEM_TESSERACT_ONLY);
+    } else if (mgr->GetComponent(TESSDATA_LSTM, &fp)) {
+      lstm_recognizer_ = new LSTMRecognizer;
+      ASSERT_HOST(lstm_recognizer_->DeSerialize(mgr->swap(), &fp));
+      if (lstm_use_matrix) lstm_recognizer_->LoadDictionary(language, mgr);
+    } else {
+      tprintf("Error: LSTM requested, but not present!! Loading tesseract.\n");
+      tessedit_ocr_engine_mode.set_value(OEM_TESSERACT_ONLY);
+    }
+  }
+#endif
+
   // Load the unicharset
-  if (!tessdata_manager.SeekToStart(TESSDATA_UNICHARSET) ||
+  if (tessedit_ocr_engine_mode == OEM_LSTM_ONLY) {
-      !unicharset.load_from_file(tessdata_manager.GetDataFilePtr())) {
+    // Avoid requiring a unicharset when we aren't running base tesseract.
+#ifndef ANDROID_BUILD
+    unicharset.CopyFrom(lstm_recognizer_->GetUnicharset());
+#endif
+  } else if (!mgr->GetComponent(TESSDATA_UNICHARSET, &fp) ||
+             !unicharset.load_from_file(&fp, false)) {
     return false;
   }
   if (unicharset.size() > MAX_NUM_CLASSES) {
     tprintf("Error: Size of unicharset is greater than MAX_NUM_CLASSES\n");
     return false;
   }
-  if (tessdata_manager_debug_level) tprintf("Loaded unicharset\n");
   right_to_left_ = unicharset.major_right_to_left();
 
   // Setup initial unichar ambigs table and read universal ambigs.
@@ -189,33 +225,11 @@ bool Tesseract::init_tesseract_lang_data(
   unichar_ambigs.InitUnicharAmbigs(unicharset, use_ambigs_for_adaption);
   unichar_ambigs.LoadUniversal(encoder_unicharset, &unicharset);
 
-  if (!tessedit_ambigs_training &&
+  if (!tessedit_ambigs_training && mgr->GetComponent(TESSDATA_AMBIGS, &fp)) {
-      tessdata_manager.SeekToStart(TESSDATA_AMBIGS)) {
+    unichar_ambigs.LoadUnicharAmbigs(encoder_unicharset, &fp,
-    TFile ambigs_file;
+                                     ambigs_debug_level,
-    ambigs_file.Open(tessdata_manager.GetDataFilePtr(),
+                                     use_ambigs_for_adaption, &unicharset);
-                     tessdata_manager.GetEndOffset(TESSDATA_AMBIGS) + 1);
-    unichar_ambigs.LoadUnicharAmbigs(
-        encoder_unicharset,
-        &ambigs_file,
-        ambigs_debug_level, use_ambigs_for_adaption, &unicharset);
-    if (tessdata_manager_debug_level) tprintf("Loaded ambigs\n");
   }
-
-  // The various OcrEngineMode settings (see publictypes.h) determine which
-  // engine-specific data files need to be loaded. Currently everything needs
-  // the base tesseract data, which supplies other useful information, but
-  // alternative engines, such as cube and LSTM are optional.
-#ifndef NO_CUBE_BUILD
-  if (tessedit_ocr_engine_mode == OEM_CUBE_ONLY) {
-    ASSERT_HOST(init_cube_objects(false, &tessdata_manager));
-    if (tessdata_manager_debug_level)
-      tprintf("Loaded Cube w/out combiner\n");
-  } else if (tessedit_ocr_engine_mode == OEM_TESSERACT_CUBE_COMBINED) {
-    ASSERT_HOST(init_cube_objects(true, &tessdata_manager));
-    if (tessdata_manager_debug_level)
-      tprintf("Loaded Cube with combiner\n");
-  }
-#endif
   // Init ParamsModel.
   // Load pass1 and pass2 weights (for now these two sets are the same, but in
   // the future separate sets of weights can be generated).
@@ -223,15 +237,12 @@ bool Tesseract::init_tesseract_lang_data(
       p < ParamsModel::PTRAIN_NUM_PASSES; ++p) {
     language_model_->getParamsModel().SetPass(
         static_cast<ParamsModel::PassEnum>(p));
-    if (tessdata_manager.SeekToStart(TESSDATA_PARAMS_MODEL)) {
+    if (mgr->GetComponent(TESSDATA_PARAMS_MODEL, &fp)) {
-      if (!language_model_->getParamsModel().LoadFromFp(
+      if (!language_model_->getParamsModel().LoadFromFp(lang.string(), &fp)) {
-          lang.string(), tessdata_manager.GetDataFilePtr(),
-          tessdata_manager.GetEndOffset(TESSDATA_PARAMS_MODEL))) {
         return false;
       }
     }
   }
-  if (tessdata_manager_debug_level) language_model_->getParamsModel().Print();
 
   return true;
 }
@@ -276,8 +287,6 @@ void Tesseract::ParseLanguageString(const char* lang_str,
     remains = next;
     // Check whether lang_code is already in the target vector and add.
     if (!IsStrInList(lang_code, *target)) {
-      if (tessdata_manager_debug_level)
-        tprintf("Adding language '%s' to list\n", lang_code.string());
       target->push_back(lang_code);
     }
   }
@@ -287,12 +296,13 @@ void Tesseract::ParseLanguageString(const char* lang_str,
 // string and recursively any additional languages required by any language
 // traineddata file (via tessedit_load_sublangs in its config) that is loaded.
 // See init_tesseract_internal for args.
-int Tesseract::init_tesseract(
+int Tesseract::init_tesseract(const char *arg0, const char *textbase,
-    const char *arg0, const char *textbase, const char *language,
+                              const char *language, OcrEngineMode oem,
-    OcrEngineMode oem, char **configs, int configs_size,
+                              char **configs, int configs_size,
                               const GenericVector<STRING> *vars_vec,
                               const GenericVector<STRING> *vars_values,
-    bool set_only_non_debug_params) {
+                              bool set_only_non_debug_params,
+                              TessdataManager *mgr) {
   GenericVector<STRING> langs_to_load;
   GenericVector<STRING> langs_not_to_load;
   ParseLanguageString(language, &langs_to_load, &langs_not_to_load);
@@ -314,15 +324,15 @@ int Tesseract::init_tesseract(
       }
 
       int result = tess_to_init->init_tesseract_internal(
-          arg0, textbase, lang_str, oem, configs, configs_size,
+          arg0, textbase, lang_str, oem, configs, configs_size, vars_vec,
-          vars_vec, vars_values, set_only_non_debug_params);
+          vars_values, set_only_non_debug_params, mgr);
+      // Forget that language, but keep any reader we were given.
+      mgr->Clear();
 
       if (!loaded_primary) {
         if (result < 0) {
           tprintf("Failed loading language '%s'\n", lang_str);
         } else {
-          if (tessdata_manager_debug_level)
-            tprintf("Loaded language '%s' as main language\n", lang_str);
           ParseLanguageString(tess_to_init->tessedit_load_sublangs.string(),
                               &langs_to_load, &langs_not_to_load);
           loaded_primary = true;
@@ -332,8 +342,6 @@ int Tesseract::init_tesseract(
           tprintf("Failed loading language '%s'\n", lang_str);
           delete tess_to_init;
         } else {
-          if (tessdata_manager_debug_level)
-            tprintf("Loaded language '%s' as secondary language\n", lang_str);
           sub_langs_.push_back(tess_to_init);
           // Add any languages that this language requires
           ParseLanguageString(tess_to_init->tessedit_load_sublangs.string(),
@@ -358,16 +366,11 @@ int Tesseract::init_tesseract(
             this->language_model_->getParamsModel());
       }
       tprintf("Using params model of the primary language\n");
-      if (tessdata_manager_debug_level)  {
-        this->language_model_->getParamsModel().Print();
-      }
     } else {
       this->language_model_->getParamsModel().Clear();
       for (int s = 0; s < sub_langs_.size(); ++s) {
         sub_langs_[s]->language_model_->getParamsModel().Clear();
       }
-      if (tessdata_manager_debug_level)
-        tprintf("Using default language params\n");
     }
   }
 
@@ -391,33 +394,26 @@ int Tesseract::init_tesseract(
 // in vars_vec.
 // If set_only_init_params is true, then only the initialization variables
 // will be set.
-int Tesseract::init_tesseract_internal(
+int Tesseract::init_tesseract_internal(const char *arg0, const char *textbase,
-    const char *arg0, const char *textbase, const char *language,
+                                       const char *language, OcrEngineMode oem,
-    OcrEngineMode oem, char **configs, int configs_size,
+                                       char **configs, int configs_size,
                                        const GenericVector<STRING> *vars_vec,
                                        const GenericVector<STRING> *vars_values,
-    bool set_only_non_debug_params) {
+                                       bool set_only_non_debug_params,
+                                       TessdataManager *mgr) {
   if (!init_tesseract_lang_data(arg0, textbase, language, oem, configs,
                                 configs_size, vars_vec, vars_values,
-                                set_only_non_debug_params)) {
+                                set_only_non_debug_params, mgr)) {
     return -1;
   }
   if (tessedit_init_config_only) {
-    tessdata_manager.End();
     return 0;
   }
-  // If only Cube will be used, skip loading Tesseract classifier's
+  // If only LSTM will be used, skip loading Tesseract classifier's
-  // pre-trained templates.
+  // pre-trained templates and dictionary.
-  bool init_tesseract_classifier =
+  bool init_tesseract = tessedit_ocr_engine_mode != OEM_LSTM_ONLY;
-    (tessedit_ocr_engine_mode == OEM_TESSERACT_ONLY ||
+  program_editup(textbase, init_tesseract ? mgr : nullptr,
-     tessedit_ocr_engine_mode == OEM_TESSERACT_CUBE_COMBINED);
+                 init_tesseract ? mgr : nullptr);
-  // If only Cube will be used and if it has its own Unicharset,
-  // skip initializing permuter and loading Tesseract Dawgs.
-  bool init_dict =
-    !(tessedit_ocr_engine_mode == OEM_CUBE_ONLY &&
-      tessdata_manager.SeekToStart(TESSDATA_CUBE_UNICHARSET));
-  program_editup(textbase, init_tesseract_classifier, init_dict);
-  tessdata_manager.End();
   return 0;                      //Normal exit
 }
 
@@ -462,14 +458,14 @@ void Tesseract::SetupUniversalFontIds() {
 }
 
 // init the LM component
-int Tesseract::init_tesseract_lm(const char *arg0,
+int Tesseract::init_tesseract_lm(const char *arg0, const char *textbase,
-                   const char *textbase,
+                                 const char *language, TessdataManager *mgr) {
-                   const char *language) {
   if (!init_tesseract_lang_data(arg0, textbase, language, OEM_TESSERACT_ONLY,
-                                NULL, 0, NULL, NULL, false))
+                                NULL, 0, NULL, NULL, false, mgr))
     return -1;
-  getDict().Load(Dict::GlobalDawgCache());
+  getDict().SetupForLoad(Dict::GlobalDawgCache());
-  tessdata_manager.End();
+  getDict().Load(lang, mgr);
+  getDict().FinishLoad();
   return 0;
 }
 

ccmain/tesseract_cube_combiner.cpp

@@ -1,306 +0,0 @@
-/**********************************************************************
- * File:        tesseract_cube_combiner.h
- * Description: Declaration of the Tesseract & Cube results combiner Class
- * Author:    Ahmad Abdulkader
- * Created:   2008
- *
- * (C) Copyright 2008, Google Inc.
- ** Licensed under the Apache License, Version 2.0 (the "License");
- ** you may not use this file except in compliance with the License.
- ** You may obtain a copy of the License at
- ** http://www.apache.org/licenses/LICENSE-2.0
- ** Unless required by applicable law or agreed to in writing, software
- ** distributed under the License is distributed on an "AS IS" BASIS,
- ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- ** See the License for the specific language governing permissions and
- ** limitations under the License.
- *
- **********************************************************************/
-
-// The TesseractCubeCombiner class provides the functionality of combining
-// the recognition results of Tesseract and Cube at the word level
-
-#include <algorithm>
-#include <string>
-#include <vector>
-#include <wctype.h>
-
-#include "tesseract_cube_combiner.h"
-
-#include "cube_object.h"
-#include "cube_reco_context.h"
-#include "cube_utils.h"
-#include "neural_net.h"
-#include "tesseractclass.h"
-#include "word_altlist.h"
-
-namespace tesseract {
-
-TesseractCubeCombiner::TesseractCubeCombiner(CubeRecoContext *cube_cntxt) {
-  cube_cntxt_ = cube_cntxt;
-  combiner_net_ = NULL;
-}
-
-TesseractCubeCombiner::~TesseractCubeCombiner() {
-  if (combiner_net_ != NULL) {
-    delete combiner_net_;
-    combiner_net_ = NULL;
-  }
-}
-
-bool TesseractCubeCombiner::LoadCombinerNet() {
-  ASSERT_HOST(cube_cntxt_);
-  // Compute the path of the combiner net
-  string data_path;
-  cube_cntxt_->GetDataFilePath(&data_path);
-  string net_file_name =  data_path + cube_cntxt_->Lang() +
-                          ".tesseract_cube.nn";
-
-  // Return false if file does not exist
-  FILE *fp = fopen(net_file_name.c_str(), "rb");
-  if (fp == NULL)
-    return false;
-  else
-    fclose(fp);
-
-  // Load and validate net
-  combiner_net_ = NeuralNet::FromFile(net_file_name);
-  if (combiner_net_ == NULL) {
-    tprintf("Could not read combiner net file %s", net_file_name.c_str());
-    return false;
-  } else if (combiner_net_->out_cnt() != 2) {
-    tprintf("Invalid combiner net file %s! Output count != 2\n",
-            net_file_name.c_str());
-    delete combiner_net_;
-    combiner_net_ = NULL;
-    return false;
-  }
-  return true;
-}
-
-// Normalize a UTF-8 string. Converts the UTF-8 string to UTF32 and optionally
-// strips punc and/or normalizes case and then converts back
-string TesseractCubeCombiner::NormalizeString(const string &str,
-                                              bool remove_punc,
-                                              bool norm_case) {
-  // convert to UTF32
-  string_32 str32;
-  CubeUtils::UTF8ToUTF32(str.c_str(), &str32);
-  // strip punc and normalize
-  string_32 new_str32;
-  for (int idx = 0; idx < str32.length(); idx++) {
-    // if no punc removal is required or not a punctuation character
-    if (!remove_punc || iswpunct(str32[idx]) == 0) {
-      char_32 norm_char = str32[idx];
-      // normalize case if required
-      if (norm_case && iswalpha(norm_char)) {
-        norm_char = towlower(norm_char);
-      }
-      new_str32.push_back(norm_char);
-    }
-  }
-  // convert back to UTF8
-  string new_str;
-  CubeUtils::UTF32ToUTF8(new_str32.c_str(), &new_str);
-  return new_str;
-}
-
-// Compares 2 strings optionally ignoring punctuation
-int TesseractCubeCombiner::CompareStrings(const string &str1,
-                                          const string &str2,
-                                          bool ignore_punc,
-                                          bool ignore_case) {
-  if (!ignore_punc && !ignore_case) {
-    return str1.compare(str2);
-  }
-  string norm_str1 = NormalizeString(str1, ignore_punc, ignore_case);
-  string norm_str2 = NormalizeString(str2, ignore_punc, ignore_case);
-  return norm_str1.compare(norm_str2);
-}
-
-// Check if a string is a valid Tess dict word or not
-bool TesseractCubeCombiner::ValidWord(const string &str) {
-  return (cube_cntxt_->TesseractObject()->getDict().valid_word(str.c_str())
-          > 0);
-}
-
-// Public method for computing the combiner features. The agreement
-// output parameter will be true if both answers are identical,
-// and false otherwise.
-bool TesseractCubeCombiner::ComputeCombinerFeatures(const string &tess_str,
-                                                    int tess_confidence,
-                                                    CubeObject *cube_obj,
-                                                    WordAltList *cube_alt_list,
-                                                    vector<double> *features,
-                                                    bool *agreement) {
-  features->clear();
-  *agreement = false;
-  if (cube_alt_list == NULL || cube_alt_list->AltCount() <= 0)
-    return false;
-
-  // Get Cube's best string; return false if empty
-  char_32 *cube_best_str32 = cube_alt_list->Alt(0);
-  if (cube_best_str32 == NULL || CubeUtils::StrLen(cube_best_str32) < 1)
-    return false;
-  string cube_best_str;
-  int cube_best_cost = cube_alt_list->AltCost(0);
-  int cube_best_bigram_cost = 0;
-  bool cube_best_bigram_cost_valid = true;
-  if (cube_cntxt_->Bigrams())
-    cube_best_bigram_cost = cube_cntxt_->Bigrams()->
-        Cost(cube_best_str32, cube_cntxt_->CharacterSet());
-  else
-    cube_best_bigram_cost_valid = false;
-  CubeUtils::UTF32ToUTF8(cube_best_str32, &cube_best_str);
-
-  // Get Tesseract's UTF32 string
-  string_32 tess_str32;
-  CubeUtils::UTF8ToUTF32(tess_str.c_str(), &tess_str32);
-
-  // Compute agreement flag
-  *agreement = (tess_str.compare(cube_best_str) == 0);
-
-  // Get Cube's second best string; if empty, return false
-  char_32 *cube_next_best_str32;
-  string cube_next_best_str;
-  int cube_next_best_cost = WORST_COST;
-  if (cube_alt_list->AltCount() > 1) {
-    cube_next_best_str32 = cube_alt_list->Alt(1);
-    if (cube_next_best_str32 == NULL ||
-        CubeUtils::StrLen(cube_next_best_str32) == 0) {
-      return false;
-    }
-    cube_next_best_cost = cube_alt_list->AltCost(1);
-    CubeUtils::UTF32ToUTF8(cube_next_best_str32, &cube_next_best_str);
-  }
-  // Rank of Tesseract's top result in Cube's alternate list
-  int tess_rank = 0;
-  for (tess_rank = 0; tess_rank < cube_alt_list->AltCount(); tess_rank++) {
-    string alt_str;
-    CubeUtils::UTF32ToUTF8(cube_alt_list->Alt(tess_rank), &alt_str);
-    if (alt_str == tess_str)
-      break;
-  }
-
-  // Cube's cost for tesseract's result. Note that this modifies the
-  // state of cube_obj, including its alternate list by calling RecognizeWord()
-  int tess_cost = cube_obj->WordCost(tess_str.c_str());
-  // Cube's bigram cost of Tesseract's string
-  int tess_bigram_cost = 0;
-  int tess_bigram_cost_valid = true;
-  if (cube_cntxt_->Bigrams())
-    tess_bigram_cost = cube_cntxt_->Bigrams()->
-        Cost(tess_str32.c_str(), cube_cntxt_->CharacterSet());
-  else
-    tess_bigram_cost_valid = false;
-
-  // Tesseract confidence
-  features->push_back(tess_confidence);
-  // Cube cost of Tesseract string
-  features->push_back(tess_cost);
-  // Cube Rank of Tesseract string
-  features->push_back(tess_rank);
-  // length of Tesseract OCR string
-  features->push_back(tess_str.length());
-  // Tesseract OCR string in dictionary
-  features->push_back(ValidWord(tess_str));
-  if (tess_bigram_cost_valid) {
-    // bigram cost of Tesseract string
-    features->push_back(tess_bigram_cost);
-  }
-  // Cube tess_cost of Cube best string
-  features->push_back(cube_best_cost);
-  // Cube tess_cost of Cube next best string
-  features->push_back(cube_next_best_cost);
-  // length of Cube string
-  features->push_back(cube_best_str.length());
-  // Cube string in dictionary
-  features->push_back(ValidWord(cube_best_str));
-  if (cube_best_bigram_cost_valid) {
-    // bigram cost of Cube string
-    features->push_back(cube_best_bigram_cost);
-  }
-  // case-insensitive string comparison, including punctuation
-  int compare_nocase_punc = CompareStrings(cube_best_str,
-                                           tess_str, false, true);
-  features->push_back(compare_nocase_punc == 0);
-  // case-sensitive string comparison, ignoring punctuation
-  int compare_case_nopunc = CompareStrings(cube_best_str,
-                                           tess_str, true, false);
-  features->push_back(compare_case_nopunc == 0);
-  // case-insensitive string comparison, ignoring punctuation
-  int compare_nocase_nopunc = CompareStrings(cube_best_str,
-                                             tess_str, true, true);
-  features->push_back(compare_nocase_nopunc == 0);
-  return true;
-}
-
-// The CubeObject parameter is used for 2 purposes: 1) to retrieve
-// cube's alt list, and 2) to compute cube's word cost for the
-// tesseract result. The call to CubeObject::WordCost() modifies
-// the object's alternate list, so previous state will be lost.
-float TesseractCubeCombiner::CombineResults(WERD_RES *tess_res,
-                                            CubeObject *cube_obj) {
-  // If no combiner is loaded or the cube object is undefined,
-  // tesseract wins with probability 1.0
-  if (combiner_net_ == NULL || cube_obj == NULL) {
-    tprintf("Cube WARNING (TesseractCubeCombiner::CombineResults): "
-            "Cube objects not initialized; defaulting to Tesseract\n");
-    return 1.0;
-  }
-
-  // Retrieve the alternate list from the CubeObject's current state.
-  // If the alt list empty, tesseract wins with probability 1.0
-  WordAltList *cube_alt_list = cube_obj->AlternateList();
-  if (cube_alt_list == NULL)
-    cube_alt_list = cube_obj->RecognizeWord();
-  if (cube_alt_list == NULL || cube_alt_list->AltCount() <= 0) {
-    tprintf("Cube WARNING (TesseractCubeCombiner::CombineResults): "
-            "Cube returned no results; defaulting to Tesseract\n");
-    return 1.0;
-  }
-  return CombineResults(tess_res, cube_obj, cube_alt_list);
-}
-
-// The alt_list parameter is expected to have been extracted from the
-// CubeObject that recognized the word to be combined. The cube_obj
-// parameter passed may be either same instance or a separate instance to
-// be used only by the combiner. In both cases, its alternate
-// list will be modified by an internal call to RecognizeWord().
-float TesseractCubeCombiner::CombineResults(WERD_RES *tess_res,
-                                            CubeObject *cube_obj,
-                                            WordAltList *cube_alt_list) {
-  // If no combiner is loaded or the cube object is undefined, or the
-  // alt list is empty, tesseract wins with probability 1.0
-  if (combiner_net_ == NULL || cube_obj == NULL ||
-      cube_alt_list == NULL || cube_alt_list->AltCount() <= 0) {
-    tprintf("Cube WARNING (TesseractCubeCombiner::CombineResults): "
-            "Cube result cannot be retrieved; defaulting to Tesseract\n");
-    return 1.0;
-  }
-
-  // Tesseract result string, tesseract confidence, and cost of
-  // tesseract result according to cube
-  string tess_str = tess_res->best_choice->unichar_string().string();
-  // Map certainty [-20.0, 0.0] to confidence [0, 100]
-  int tess_confidence = MIN(100, MAX(1, static_cast<int>(
-      100 + (5 * tess_res->best_choice->certainty()))));
-
-  // Compute the combiner features. If feature computation fails or
-  // answers are identical, tesseract wins with probability 1.0
-  vector<double> features;
-  bool agreement;
-  bool combiner_success = ComputeCombinerFeatures(tess_str, tess_confidence,
-                                                  cube_obj, cube_alt_list,
-                                                  &features, &agreement);
-  if (!combiner_success || agreement)
-    return 1.0;
-
-  // Classify combiner feature vector and return output (probability
-  // of tesseract class).
-  double net_out[2];
-  if (!combiner_net_->FeedForward(&features[0], net_out))
-    return 1.0;
-  return net_out[1];
-}
-}

ccmain/tesseract_cube_combiner.h

@@ -1,103 +0,0 @@
-/**********************************************************************
- * File:        tesseract_cube_combiner.h
- * Description: Declaration of the Tesseract & Cube results combiner Class
- * Author:    Ahmad Abdulkader
- * Created:   2008
- *
- * (C) Copyright 2008, Google Inc.
- ** Licensed under the Apache License, Version 2.0 (the "License");
- ** you may not use this file except in compliance with the License.
- ** You may obtain a copy of the License at
- ** http://www.apache.org/licenses/LICENSE-2.0
- ** Unless required by applicable law or agreed to in writing, software
- ** distributed under the License is distributed on an "AS IS" BASIS,
- ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- ** See the License for the specific language governing permissions and
- ** limitations under the License.
- *
- **********************************************************************/
-
-// The TesseractCubeCombiner class provides the functionality of combining
-// the recognition results of Tesseract and Cube at the word level
-
-#ifndef TESSERACT_CCMAIN_TESSERACT_CUBE_COMBINER_H
-#define TESSERACT_CCMAIN_TESSERACT_CUBE_COMBINER_H
-
-#include <string>
-#include <vector>
-#include "pageres.h"
-
-#ifdef _WIN32
-#include <windows.h>
-using namespace std;
-#endif
-
-#ifdef USE_STD_NAMESPACE
-using std::string;
-using std::vector;
-#endif
-
-namespace tesseract {
-
-class CubeObject;
-class NeuralNet;
-class CubeRecoContext;
-class WordAltList;
-
-class TesseractCubeCombiner {
- public:
-  explicit TesseractCubeCombiner(CubeRecoContext *cube_cntxt);
-  virtual ~TesseractCubeCombiner();
-
-  // There are 2 public methods for combining the results of tesseract
-  // and cube. Both return the probability that the Tesseract result is
-  // correct. The difference between the two interfaces is in how the
-  // passed-in CubeObject is used.
-
-  // The CubeObject parameter is used for 2 purposes: 1) to retrieve
-  // cube's alt list, and 2) to compute cube's word cost for the
-  // tesseract result. Both uses may modify the state of the
-  // CubeObject (including the BeamSearch state) with a call to
-  // RecognizeWord().
-  float CombineResults(WERD_RES *tess_res, CubeObject *cube_obj);
-
-  // The alt_list parameter is expected to have been extracted from the
-  // CubeObject that recognized the word to be combined. The cube_obj
-  // parameter passed in is a separate instance to be used only by
-  // the combiner.
-  float CombineResults(WERD_RES *tess_res, CubeObject *cube_obj,
-                       WordAltList *alt_list);
-
-  // Public method for computing the combiner features. The agreement
-  // output parameter will be true if both answers are identical,
-  // false otherwise. Modifies the cube_alt_list, so no assumptions
-  // should be made about its state upon return.
-  bool ComputeCombinerFeatures(const string &tess_res,
-                               int tess_confidence,
-                               CubeObject *cube_obj,
-                               WordAltList *cube_alt_list,
-                               vector<double> *features,
-                               bool *agreement);
-
-  // Is the word valid according to Tesseract's language model
-  bool ValidWord(const string &str);
-
-  // Loads the combiner neural network from file, using cube_cntxt_
-  // to find path.
-  bool LoadCombinerNet();
- private:
-  // Normalize a UTF-8 string. Converts the UTF-8 string to UTF32 and optionally
-  // strips punc and/or normalizes case and then converts back
-  string NormalizeString(const string &str, bool remove_punc, bool norm_case);
-
-  // Compares 2 strings after optionally normalizing them and or stripping
-  // punctuation
-  int CompareStrings(const string &str1, const string &str2, bool ignore_punc,
-                     bool norm_case);
-
-  NeuralNet *combiner_net_;  // pointer to the combiner NeuralNet object
-  CubeRecoContext *cube_cntxt_;  // used for language ID and data paths
-};
-}
-
-#endif  // TESSERACT_CCMAIN_TESSERACT_CUBE_COMBINER_H

ccmain/tesseractclass.cpp

@@ -42,14 +42,11 @@
 #include "tesseractclass.h"
 
 #include "allheaders.h"
-#ifndef NO_CUBE_BUILD
-#include "cube_reco_context.h"
-#endif
 #include "edgblob.h"
 #include "equationdetect.h"
 #include "globals.h"
-#ifndef NO_CUBE_BUILD
+#ifndef ANDROID_BUILD
-#include "tesseract_cube_combiner.h"
+#include "lstmrecognizer.h"
 #endif
 
 namespace tesseract {
@@ -65,6 +62,9 @@ Tesseract::Tesseract()
                   "Generate training data from boxed chars", this->params()),
       BOOL_MEMBER(tessedit_make_boxes_from_boxes, false,
                   "Generate more boxes from boxed chars", this->params()),
+      BOOL_MEMBER(tessedit_train_line_recognizer, false,
+                  "Break input into lines and remap boxes if present",
+                  this->params()),
       BOOL_MEMBER(tessedit_dump_pageseg_images, false,
                   "Dump intermediate images made during page segmentation",
                   this->params()),
@@ -76,11 +76,10 @@ Tesseract::Tesseract()
           " 5=line, 6=word, 7=char"
           " (Values from PageSegMode enum in publictypes.h)",
           this->params()),
-      INT_INIT_MEMBER(tessedit_ocr_engine_mode, tesseract::OEM_TESSERACT_ONLY,
+      INT_INIT_MEMBER(tessedit_ocr_engine_mode, tesseract::OEM_DEFAULT,
-                      "Which OCR engine(s) to run (Tesseract, Cube, both)."
+                      "Which OCR engine(s) to run (Tesseract, LSTM, both)."
-                      " Defaults to loading and running only Tesseract"
+                      " Defaults to loading and running the most accurate"
-                      " (no Cube,no combiner)."
+                      " available.",
-                      " Values from OcrEngineMode enum in tesseractclass.h)",
                       this->params()),
       STRING_MEMBER(tessedit_char_blacklist, "",
                     "Blacklist of chars not to recognize", this->params()),
@@ -215,13 +214,16 @@ Tesseract::Tesseract()
       BOOL_MEMBER(test_pt, false, "Test for point", this->params()),
       double_MEMBER(test_pt_x, 99999.99, "xcoord", this->params()),
       double_MEMBER(test_pt_y, 99999.99, "ycoord", this->params()),
+      INT_MEMBER(multilang_debug_level, 0, "Print multilang debug info.",
+                 this->params()),
       INT_MEMBER(paragraph_debug_level, 0, "Print paragraph debug info.",
                  this->params()),
       BOOL_MEMBER(paragraph_text_based, true,
                   "Run paragraph detection on the post-text-recognition "
                   "(more accurate)",
                   this->params()),
-      INT_MEMBER(cube_debug_level, 0, "Print cube debug info.", this->params()),
+      BOOL_MEMBER(lstm_use_matrix, 1,
+                  "Use ratings matrix/beam search with lstm", this->params()),
       STRING_MEMBER(outlines_odd, "%| ", "Non standard number of outlines",
                     this->params()),
       STRING_MEMBER(outlines_2, "ij!?%\":;", "Non standard number of outlines",
@@ -265,7 +267,7 @@ Tesseract::Tesseract()
                   this->params()),
       BOOL_MEMBER(tessedit_debug_quality_metrics, false,
                   "Output data to debug file", this->params()),
-      BOOL_MEMBER(bland_unrej, false, "unrej potential with no chekcs",
+      BOOL_MEMBER(bland_unrej, false, "unrej potential with no checks",
                   this->params()),
       double_MEMBER(quality_rowrej_pc, 1.1,
                     "good_quality_doc gte good char limit", this->params()),
@@ -389,6 +391,9 @@ Tesseract::Tesseract()
                   this->params()),
       BOOL_MEMBER(tessedit_create_pdf, false, "Write .pdf output file",
                   this->params()),
+      BOOL_MEMBER(textonly_pdf, false,
+                  "Create PDF with only one invisible text layer",
+                  this->params()),
       STRING_MEMBER(unrecognised_char, "|",
                     "Output char for unidentified blobs", this->params()),
       INT_MEMBER(suspect_level, 99, "Suspect marker level", this->params()),
@@ -398,8 +403,8 @@ Tesseract::Tesseract()
                  "Don't suspect dict wds longer than this", this->params()),
       BOOL_MEMBER(suspect_constrain_1Il, false, "UNLV keep 1Il chars rejected",
                   this->params()),
-      double_MEMBER(suspect_rating_per_ch, 999.9, "Don't touch bad rating limit",
+      double_MEMBER(suspect_rating_per_ch, 999.9,
-                    this->params()),
+                    "Don't touch bad rating limit", this->params()),
       double_MEMBER(suspect_accept_rating, -999.9, "Accept good rating limit",
                     this->params()),
       BOOL_MEMBER(tessedit_minimal_rejection, false,
@@ -452,7 +457,7 @@ Tesseract::Tesseract()
                   this->params()),
       INT_MEMBER(tessedit_page_number, -1,
                  "-1 -> All pages"
-                 " , else specifc page to process",
+                 " , else specific page to process",
                  this->params()),
       BOOL_MEMBER(tessedit_write_images, false,
                   "Capture the image from the IPE", this->params()),
@@ -461,10 +466,6 @@ Tesseract::Tesseract()
       STRING_MEMBER(file_type, ".tif", "Filename extension", this->params()),
       BOOL_MEMBER(tessedit_override_permuter, true, "According to dict_word",
                   this->params()),
-      INT_MEMBER(tessdata_manager_debug_level, 0,
-                 "Debug level for"
-                 " TessdataManager functions.",
-                 this->params()),
       STRING_MEMBER(tessedit_load_sublangs, "",
                     "List of languages to load with this one", this->params()),
       BOOL_MEMBER(tessedit_use_primary_params_model, false,
@@ -512,7 +513,6 @@ Tesseract::Tesseract()
                     "Page separator (default is form feed control character)",
                     this->params()),
 
-
       // The following parameters were deprecated and removed from their
       // original
       // locations. The parameters are temporarily kept here to give Tesseract
@@ -604,8 +604,8 @@ Tesseract::Tesseract()
 
       backup_config_file_(NULL),
       pix_binary_(NULL),
-      cube_binary_(NULL),
       pix_grey_(NULL),
+      pix_original_(NULL),
       pix_thresholds_(NULL),
       source_resolution_(0),
       textord_(this),
@@ -616,33 +616,28 @@ Tesseract::Tesseract()
       reskew_(1.0f, 0.0f),
       most_recently_used_(this),
       font_table_size_(0),
-#ifndef NO_CUBE_BUILD
+      equ_detect_(NULL),
-      cube_cntxt_(NULL),
+#ifndef ANDROID_BUILD
-      tess_cube_combiner_(NULL),
+      lstm_recognizer_(NULL),
 #endif
-      equ_detect_(NULL) {
+      train_line_page_num_(0) {
 }
 
 Tesseract::~Tesseract() {
   Clear();
+  pixDestroy(&pix_original_);
   end_tesseract();
   sub_langs_.delete_data_pointers();
-#ifndef NO_CUBE_BUILD
+#ifndef ANDROID_BUILD
-  // Delete cube objects.
+  delete lstm_recognizer_;
-  if (cube_cntxt_ != NULL) {
+  lstm_recognizer_ = NULL;
-    delete cube_cntxt_;
-    cube_cntxt_ = NULL;
-  }
-  if (tess_cube_combiner_ != NULL) {
-    delete tess_cube_combiner_;
-    tess_cube_combiner_ = NULL;
-  }
 #endif
 }
 
 void Tesseract::Clear() {
+  STRING debug_name = imagebasename + "_debug.pdf";
+  pixa_debug_.WritePDF(debug_name.string());
   pixDestroy(&pix_binary_);
-  pixDestroy(&cube_binary_);
   pixDestroy(&pix_grey_);
   pixDestroy(&pix_thresholds_);
   pixDestroy(&scaled_color_);
@@ -692,8 +687,6 @@ void Tesseract::SetBlackAndWhitelist() {
 // page segmentation.
 void Tesseract::PrepareForPageseg() {
   textord_.set_use_cjk_fp_model(textord_use_cjk_fp_model);
-  pixDestroy(&cube_binary_);
-  cube_binary_ = pixClone(pix_binary());
   // Find the max splitter strategy over all langs.
   ShiroRekhaSplitter::SplitStrategy max_pageseg_strategy =
       static_cast<ShiroRekhaSplitter::SplitStrategy>(
@@ -704,9 +697,6 @@ void Tesseract::PrepareForPageseg() {
         static_cast<inT32>(sub_langs_[i]->pageseg_devanagari_split_strategy));
     if (pageseg_strategy > max_pageseg_strategy)
       max_pageseg_strategy = pageseg_strategy;
-    // Clone the cube image to all the sub langs too.
-    pixDestroy(&sub_langs_[i]->cube_binary_);
-    sub_langs_[i]->cube_binary_ = pixClone(pix_binary());
     pixDestroy(&sub_langs_[i]->pix_binary_);
     sub_langs_[i]->pix_binary_ = pixClone(pix_binary());
   }
@@ -714,7 +704,7 @@ void Tesseract::PrepareForPageseg() {
   // the newly splitted image.
   splitter_.set_orig_pix(pix_binary());
   splitter_.set_pageseg_split_strategy(max_pageseg_strategy);
-  if (splitter_.Split(true)) {
+  if (splitter_.Split(true, &pixa_debug_)) {
     ASSERT_HOST(splitter_.splitted_image());
     pixDestroy(&pix_binary_);
     pix_binary_ = pixClone(splitter_.splitted_image());
@@ -743,7 +733,7 @@ void Tesseract::PrepareForTessOCR(BLOCK_LIST* block_list,
   splitter_.set_segmentation_block_list(block_list);
   splitter_.set_ocr_split_strategy(max_ocr_strategy);
   // Run the splitter for OCR
-  bool split_for_ocr = splitter_.Split(false);
+  bool split_for_ocr = splitter_.Split(false, &pixa_debug_);
   // Restore pix_binary to the binarized original pix for future reference.
   ASSERT_HOST(splitter_.orig_pix());
   pixDestroy(&pix_binary_);

ccmain/tesseractclass.h

@@ -23,22 +23,22 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_TESSERACTCLASS_H__
+#ifndef TESSERACT_CCMAIN_TESSERACTCLASS_H_
-#define TESSERACT_CCMAIN_TESSERACTCLASS_H__
+#define TESSERACT_CCMAIN_TESSERACTCLASS_H_
 
 #include "allheaders.h"
 #include "control.h"
-#include "docqual.h"
+#include "debugpixa.h"
 #include "devanagari_processing.h"
+#include "docqual.h"
 #include "genericvector.h"
-#include "params.h"
 #include "ocrclass.h"
+#include "params.h"
 #include "textord.h"
 #include "wordrec.h"
 
 class BLOB_CHOICE_LIST_CLIST;
 class BLOCK_LIST;
-class CharSamp;
 struct OSResults;
 class PAGE_RES;
 class PAGE_RES_IT;
@@ -77,8 +77,7 @@ class WERD_RES;
 //             WordRec (wordrec/wordrec.h)
 //                         ^       Members include: WERD*, DENORM*
 //        Tesseract (ccmain/tesseractclass.h)
-//                                 Members include: Pix*, CubeRecoContext*,
+//                                 Members include: Pix*
-//                                 TesseractCubeCombiner*
 //
 // Other important classes:
 //
@@ -97,16 +96,11 @@ class WERD_RES;
 namespace tesseract {
 
 class ColumnFinder;
-#ifndef NO_CUBE_BUILD
+class DocumentData;
-class CubeLineObject;
-class CubeObject;
-class CubeRecoContext;
-#endif
 class EquationDetect;
+class ImageData;
+class LSTMRecognizer;
 class Tesseract;
-#ifndef NO_CUBE_BUILD
-class TesseractCubeCombiner;
-#endif
 
 // A collection of various variables for statistics and debugging.
 struct TesseractStats {
@@ -189,7 +183,7 @@ class Tesseract : public Wordrec {
   }
   // Destroy any existing pix and return a pointer to the pointer.
   Pix** mutable_pix_binary() {
-    Clear();
+    pixDestroy(&pix_binary_);
     return &pix_binary_;
   }
   Pix* pix_binary() const {
@@ -202,16 +196,24 @@ class Tesseract : public Wordrec {
     pixDestroy(&pix_grey_);
     pix_grey_ = grey_pix;
   }
-  // Returns a pointer to a Pix representing the best available image of the
+  Pix* pix_original() const { return pix_original_; }
-  // page. The image will be 8-bit grey if the input was grey or color. Note
+  // Takes ownership of the given original_pix.
-  // that in grey 0 is black and 255 is white. If the input was binary, then
+  void set_pix_original(Pix* original_pix) {
-  // the returned Pix will be binary. Note that here black is 1 and white is 0.
+    pixDestroy(&pix_original_);
-  // To tell the difference pixGetDepth() will return 8 or 1.
+    pix_original_ = original_pix;
-  // In either case, the return value is a borrowed Pix, and should not be
+    // Clone to sublangs as well.
-  // deleted or pixDestroyed.
+    for (int i = 0; i < sub_langs_.size(); ++i)
-  Pix* BestPix() const {
+      sub_langs_[i]->set_pix_original(original_pix ? pixClone(original_pix)
-    return pix_grey_ != NULL ? pix_grey_ : pix_binary_;
+                                                   : nullptr);
   }
+  // Returns a pointer to a Pix representing the best available (original) image
+  // of the page. Can be of any bit depth, but never color-mapped, as that has
+  // always been dealt with. Note that in grey and color, 0 is black and 255 is
+  // white. If the input was binary, then black is 1 and white is 0.
+  // To tell the difference pixGetDepth() will return 32, 8 or 1.
+  // In any case, the return value is a borrowed Pix, and should not be
+  // deleted or pixDestroyed.
+  Pix* BestPix() const { return pix_original_; }
   void set_pix_thresholds(Pix* thresholds) {
     pixDestroy(&pix_thresholds_);
     pix_thresholds_ = thresholds;
@@ -254,11 +256,19 @@ class Tesseract : public Wordrec {
   Tesseract* get_sub_lang(int index) const {
     return sub_langs_[index];
   }
-  // Returns true if any language uses Tesseract (as opposed to cube).
+  // Returns true if any language uses Tesseract (as opposed to LSTM).
   bool AnyTessLang() const {
-    if (tessedit_ocr_engine_mode != OEM_CUBE_ONLY) return true;
+    if (tessedit_ocr_engine_mode != OEM_LSTM_ONLY) return true;
     for (int i = 0; i < sub_langs_.size(); ++i) {
-      if (sub_langs_[i]->tessedit_ocr_engine_mode != OEM_CUBE_ONLY)
+      if (sub_langs_[i]->tessedit_ocr_engine_mode != OEM_LSTM_ONLY) return true;
+    }
+    return false;
+  }
+  // Returns true if any language uses the LSTM.
+  bool AnyLSTMLang() const {
+    if (tessedit_ocr_engine_mode != OEM_TESSERACT_ONLY) return true;
+    for (int i = 0; i < sub_langs_.size(); ++i) {
+      if (sub_langs_[i]->tessedit_ocr_engine_mode != OEM_TESSERACT_ONLY)
         return true;
     }
     return false;
@@ -293,6 +303,46 @@ class Tesseract : public Wordrec {
   // par_control.cpp
   void PrerecAllWordsPar(const GenericVector<WordData>& words);
 
+  //// linerec.cpp
+  // Generates training data for training a line recognizer, eg LSTM.
+  // Breaks the page into lines, according to the boxes, and writes them to a
+  // serialized DocumentData based on output_basename.
+  void TrainLineRecognizer(const STRING& input_imagename,
+                           const STRING& output_basename,
+                           BLOCK_LIST *block_list);
+  // Generates training data for training a line recognizer, eg LSTM.
+  // Breaks the boxes into lines, normalizes them, converts to ImageData and
+  // appends them to the given training_data.
+  void TrainFromBoxes(const GenericVector<TBOX>& boxes,
+                      const GenericVector<STRING>& texts,
+                      BLOCK_LIST *block_list,
+                      DocumentData* training_data);
+
+  // Returns an Imagedata containing the image of the given textline,
+  // and ground truth boxes/truth text if available in the input.
+  // The image is not normalized in any way.
+  ImageData* GetLineData(const TBOX& line_box,
+                         const GenericVector<TBOX>& boxes,
+                         const GenericVector<STRING>& texts,
+                         int start_box, int end_box,
+                         const BLOCK& block);
+  // Helper gets the image of a rectangle, using the block.re_rotation() if
+  // needed to get to the image, and rotating the result back to horizontal
+  // layout. (CJK characters will be on their left sides) The vertical text flag
+  // is set in the returned ImageData if the text was originally vertical, which
+  // can be used to invoke a different CJK recognition engine. The revised_box
+  // is also returned to enable calculation of output bounding boxes.
+  ImageData* GetRectImage(const TBOX& box, const BLOCK& block, int padding,
+                          TBOX* revised_box) const;
+  // Recognizes a word or group of words, converting to WERD_RES in *words.
+  // Analogous to classify_word_pass1, but can handle a group of words as well.
+  void LSTMRecognizeWord(const BLOCK& block, ROW *row, WERD_RES *word,
+                         PointerVector<WERD_RES>* words);
+  // Apply segmentation search to the given set of words, within the constraints
+  // of the existing ratings matrix. If there is already a best_choice on a word
+  // leaves it untouched and just sets the done/accepted etc flags.
+  void SearchWords(PointerVector<WERD_RES>* words);
+
   //// control.h /////////////////////////////////////////////////////////
   bool ProcessTargetWord(const TBOX& word_box, const TBOX& target_word_box,
                          const char* word_config, int pass);
@@ -324,9 +374,8 @@ class Tesseract : public Wordrec {
   // Helper to recognize the word using the given (language-specific) tesseract.
   // Returns positive if this recognizer found more new best words than the
   // number kept from best_words.
-  int RetryWithLanguage(const WordData& word_data,
+  int RetryWithLanguage(const WordData& word_data, WordRecognizer recognizer,
-                        WordRecognizer recognizer,
+                        bool debug, WERD_RES** in_word,
-                        WERD_RES** in_word,
                         PointerVector<WERD_RES>* best_words);
   // Moves good-looking "noise"/diacritics from the reject list to the main
   // blob list on the current word. Returns true if anything was done, and
@@ -428,34 +477,6 @@ class Tesseract : public Wordrec {
                              int *left_ok,
                              int *right_ok) const;
 
-  //// cube_control.cpp ///////////////////////////////////////////////////
-#ifndef NO_CUBE_BUILD
-  bool init_cube_objects(bool load_combiner,
-                         TessdataManager *tessdata_manager);
-  // Iterates through tesseract's results and calls cube on each word,
-  // combining the results with the existing tesseract result.
-  void run_cube_combiner(PAGE_RES *page_res);
-  // Recognizes a single word using (only) cube. Compatible with
-  // Tesseract's classify_word_pass1/classify_word_pass2.
-  void cube_word_pass1(BLOCK* block, ROW *row, WERD_RES *word);
-  // Cube recognizer to recognize a single word as with classify_word_pass1
-  // but also returns the cube object in case the combiner is needed.
-  CubeObject* cube_recognize_word(BLOCK* block, WERD_RES* word);
-  // Combines the cube and tesseract results for a single word, leaving the
-  // result in tess_word.
-  void cube_combine_word(CubeObject* cube_obj, WERD_RES* cube_word,
-                        WERD_RES* tess_word);
-  // Call cube on the current word, and write the result to word.
-  // Sets up a fake result  and returns false if something goes wrong.
-  bool cube_recognize(CubeObject *cube_obj, BLOCK* block, WERD_RES *word);
-  void fill_werd_res(const BoxWord& cube_box_word,
-                     const char* cube_best_str,
-                     WERD_RES* tess_werd_res);
-  bool extract_cube_state(CubeObject* cube_obj, int* num_chars,
-                          Boxa** char_boxes, CharSamp*** char_samples);
-  bool create_cube_box_word(Boxa *char_boxes, int num_chars,
-                            TBOX word_box, BoxWord* box_word);
-#endif
   //// output.h //////////////////////////////////////////////////////////
 
   void output_pass(PAGE_RES_IT &page_res_it, const TBOX *target_word_box);
@@ -475,20 +496,17 @@ class Tesseract : public Wordrec {
   // string and recursively any additional languages required by any language
   // traineddata file (via tessedit_load_sublangs in its config) that is loaded.
   // See init_tesseract_internal for args.
-  int init_tesseract(const char *arg0,
+  int init_tesseract(const char* arg0, const char* textbase,
-                     const char *textbase,
+                     const char* language, OcrEngineMode oem, char** configs,
-                     const char *language,
+                     int configs_size, const GenericVector<STRING>* vars_vec,
-                     OcrEngineMode oem,
+                     const GenericVector<STRING>* vars_values,
-                     char **configs,
+                     bool set_only_init_params, TessdataManager* mgr);
-                     int configs_size,
-                     const GenericVector<STRING> *vars_vec,
-                     const GenericVector<STRING> *vars_values,
-                     bool set_only_init_params);
   int init_tesseract(const char *datapath,
                      const char *language,
                      OcrEngineMode oem) {
-    return init_tesseract(datapath, NULL, language, oem,
+    TessdataManager mgr;
-                          NULL, 0, NULL, NULL, false);
+    return init_tesseract(datapath, NULL, language, oem, NULL, 0, NULL, NULL,
+                          false, &mgr);
   }
   // Common initialization for a single language.
   // arg0 is the datapath for the tessdata directory, which could be the
@@ -506,36 +524,30 @@ class Tesseract : public Wordrec {
   // in vars_vec.
   // If set_only_init_params is true, then only the initialization variables
   // will be set.
-  int init_tesseract_internal(const char *arg0,
+  int init_tesseract_internal(const char* arg0, const char* textbase,
-                              const char *textbase,
+                              const char* language, OcrEngineMode oem,
-                              const char *language,
+                              char** configs, int configs_size,
-                              OcrEngineMode oem,
+                              const GenericVector<STRING>* vars_vec,
-                              char **configs,
+                              const GenericVector<STRING>* vars_values,
-                              int configs_size,
+                              bool set_only_init_params, TessdataManager* mgr);
-                              const GenericVector<STRING> *vars_vec,
-                              const GenericVector<STRING> *vars_values,
-                              bool set_only_init_params);
 
   // Set the universal_id member of each font to be unique among all
   // instances of the same font loaded.
   void SetupUniversalFontIds();
 
-  int init_tesseract_lm(const char *arg0,
+  int init_tesseract_lm(const char* arg0, const char* textbase,
-                        const char *textbase,
+                        const char* language, TessdataManager* mgr);
-                        const char *language);
 
   void recognize_page(STRING& image_name);
   void end_tesseract();
 
-  bool init_tesseract_lang_data(const char *arg0,
+  bool init_tesseract_lang_data(const char* arg0, const char* textbase,
-                                const char *textbase,
+                                const char* language, OcrEngineMode oem,
-                                const char *language,
+                                char** configs, int configs_size,
-                                OcrEngineMode oem,
+                                const GenericVector<STRING>* vars_vec,
-                                char **configs,
+                                const GenericVector<STRING>* vars_values,
-                                int configs_size,
+                                bool set_only_init_params,
-                                const GenericVector<STRING> *vars_vec,
+                                TessdataManager* mgr);
-                                const GenericVector<STRING> *vars_values,
-                                bool set_only_init_params);
 
   void ParseLanguageString(const char* lang_str,
                            GenericVector<STRING>* to_load,
@@ -783,16 +795,17 @@ class Tesseract : public Wordrec {
              "Generate training data from boxed chars");
   BOOL_VAR_H(tessedit_make_boxes_from_boxes, false,
              "Generate more boxes from boxed chars");
+  BOOL_VAR_H(tessedit_train_line_recognizer, false,
+             "Break input into lines and remap boxes if present");
   BOOL_VAR_H(tessedit_dump_pageseg_images, false,
              "Dump intermediate images made during page segmentation");
   INT_VAR_H(tessedit_pageseg_mode, PSM_SINGLE_BLOCK,
             "Page seg mode: 0=osd only, 1=auto+osd, 2=auto, 3=col, 4=block,"
             " 5=line, 6=word, 7=char"
             " (Values from PageSegMode enum in publictypes.h)");
-  INT_VAR_H(tessedit_ocr_engine_mode, tesseract::OEM_TESSERACT_ONLY,
+  INT_VAR_H(tessedit_ocr_engine_mode, tesseract::OEM_DEFAULT,
-            "Which OCR engine(s) to run (Tesseract, Cube, both). Defaults"
+            "Which OCR engine(s) to run (Tesseract, LSTM, both). Defaults"
-            " to loading and running only Tesseract (no Cube, no combiner)."
+            " to loading and running the most accurate available.");
-            " (Values from OcrEngineMode enum in tesseractclass.h)");
   STRING_VAR_H(tessedit_char_blacklist, "",
                "Blacklist of chars not to recognize");
   STRING_VAR_H(tessedit_char_whitelist, "",
@@ -886,11 +899,12 @@ class Tesseract : public Wordrec {
   BOOL_VAR_H(test_pt, false, "Test for point");
   double_VAR_H(test_pt_x, 99999.99, "xcoord");
   double_VAR_H(test_pt_y, 99999.99, "ycoord");
+  INT_VAR_H(multilang_debug_level, 0, "Print multilang debug info.");
   INT_VAR_H(paragraph_debug_level, 0, "Print paragraph debug info.");
   BOOL_VAR_H(paragraph_text_based, true,
              "Run paragraph detection on the post-text-recognition "
              "(more accurate)");
-  INT_VAR_H(cube_debug_level, 1, "Print cube debug info.");
+  BOOL_VAR_H(lstm_use_matrix, 1, "Use ratings matrix/beam searct with lstm");
   STRING_VAR_H(outlines_odd, "%| ", "Non standard number of outlines");
   STRING_VAR_H(outlines_2, "ij!?%\":;", "Non standard number of outlines");
   BOOL_VAR_H(docqual_excuse_outline_errs, false,
@@ -926,7 +940,7 @@ class Tesseract : public Wordrec {
   BOOL_VAR_H(tessedit_debug_doc_rejection, false, "Page stats");
   BOOL_VAR_H(tessedit_debug_quality_metrics, false,
              "Output data to debug file");
-  BOOL_VAR_H(bland_unrej, false, "unrej potential with no chekcs");
+  BOOL_VAR_H(bland_unrej, false, "unrej potential with no checks");
   double_VAR_H(quality_rowrej_pc, 1.1,
                "good_quality_doc gte good char limit");
   BOOL_VAR_H(unlv_tilde_crunching, true,
@@ -1005,13 +1019,14 @@ class Tesseract : public Wordrec {
   BOOL_VAR_H(tessedit_create_hocr, false, "Write .html hOCR output file");
   BOOL_VAR_H(tessedit_create_tsv, false, "Write .tsv output file");
   BOOL_VAR_H(tessedit_create_pdf, false, "Write .pdf output file");
+  BOOL_VAR_H(textonly_pdf, false,
+             "Create PDF with only one invisible text layer");
   STRING_VAR_H(unrecognised_char, "|",
                "Output char for unidentified blobs");
   INT_VAR_H(suspect_level, 99, "Suspect marker level");
   INT_VAR_H(suspect_space_level, 100,
             "Min suspect level for rejecting spaces");
-  INT_VAR_H(suspect_short_words, 2,
+  INT_VAR_H(suspect_short_words, 2, "Don't Suspect dict wds longer than this");
-            "Don't Suspect dict wds longer than this");
   BOOL_VAR_H(suspect_constrain_1Il, false, "UNLV keep 1Il chars rejected");
   double_VAR_H(suspect_rating_per_ch, 999.9, "Don't touch bad rating limit");
   double_VAR_H(suspect_accept_rating, -999.9, "Accept good rating limit");
@@ -1045,13 +1060,11 @@ class Tesseract : public Wordrec {
   INT_VAR_H(min_sane_x_ht_pixels, 8, "Reject any x-ht lt or eq than this");
   BOOL_VAR_H(tessedit_create_boxfile, false, "Output text with boxes");
   INT_VAR_H(tessedit_page_number, -1,
-            "-1 -> All pages, else specifc page to process");
+            "-1 -> All pages, else specific page to process");
   BOOL_VAR_H(tessedit_write_images, false, "Capture the image from the IPE");
   BOOL_VAR_H(interactive_display_mode, false, "Run interactively?");
   STRING_VAR_H(file_type, ".tif", "Filename extension");
   BOOL_VAR_H(tessedit_override_permuter, true, "According to dict_word");
-  INT_VAR_H(tessdata_manager_debug_level, 0,
-            "Debug level for TessdataManager functions.");
   STRING_VAR_H(tessedit_load_sublangs, "",
                "List of languages to load with this one");
   BOOL_VAR_H(tessedit_use_primary_params_model, false,
@@ -1157,10 +1170,6 @@ class Tesseract : public Wordrec {
                                   PAGE_RES_IT* pr_it,
                                   FILE *output_file);
 
-#ifndef NO_CUBE_BUILD
-  inline CubeRecoContext *GetCubeRecoContext() { return cube_cntxt_; }
-#endif
-
  private:
   // The filename of a backup config file. If not null, then we currently
   // have a temporary debug config file loaded, and backup_config_file_
@@ -1171,12 +1180,14 @@ class Tesseract : public Wordrec {
   // Image used for input to layout analysis and tesseract recognition.
   // May be modified by the ShiroRekhaSplitter to eliminate the top-line.
   Pix* pix_binary_;
-  // Unmodified image used for input to cube. Always valid.
-  Pix* cube_binary_;
   // Grey-level input image if the input was not binary, otherwise NULL.
   Pix* pix_grey_;
+  // Original input image. Color if the input was color.
+  Pix* pix_original_;
   // Thresholds that were used to generate the thresholded image from grey.
   Pix* pix_thresholds_;
+  // Debug images. If non-empty, will be written on destruction.
+  DebugPixa pixa_debug_;
   // Input image resolution after any scaling. The resolution is not well
   // transmitted by operations on Pix, so we keep an independent record here.
   int source_resolution_;
@@ -1199,16 +1210,14 @@ class Tesseract : public Wordrec {
   Tesseract* most_recently_used_;
   // The size of the font table, ie max possible font id + 1.
   int font_table_size_;
-#ifndef NO_CUBE_BUILD
-  // Cube objects.
-  CubeRecoContext* cube_cntxt_;
-  TesseractCubeCombiner *tess_cube_combiner_;
-#endif
   // Equation detector. Note: this pointer is NOT owned by the class.
   EquationDetect* equ_detect_;
+  // LSTM recognizer, if available.
+  LSTMRecognizer* lstm_recognizer_;
+  // Output "page" number (actually line number) using TrainLineRecognizer.
+  int train_line_page_num_;
 };
 
 }  // namespace tesseract
 
-
+#endif  // TESSERACT_CCMAIN_TESSERACTCLASS_H_
-#endif  // TESSERACT_CCMAIN_TESSERACTCLASS_H__

ccmain/thresholder.cpp

@@ -152,19 +152,27 @@ void ImageThresholder::SetImage(const Pix* pix) {
   int depth;
   pixGetDimensions(src, &image_width_, &image_height_, &depth);
   // Convert the image as necessary so it is one of binary, plain RGB, or
-  // 8 bit with no colormap.
+  // 8 bit with no colormap. Guarantee that we always end up with our own copy,
+  // not just a clone of the input.
+  if (pixGetColormap(src)) {
+    Pix* tmp = pixRemoveColormap(src, REMOVE_CMAP_BASED_ON_SRC);
+    depth = pixGetDepth(tmp);
     if (depth > 1 && depth < 8) {
-    pix_ = pixConvertTo8(src, false);
+      pix_ = pixConvertTo8(tmp, false);
-  } else if (pixGetColormap(src)) {
+      pixDestroy(&tmp);
-    pix_ = pixRemoveColormap(src, REMOVE_CMAP_BASED_ON_SRC);
     } else {
-    pix_ = pixClone(src);
+      pix_ = tmp;
+    }
+  } else if (depth > 1 && depth < 8) {
+    pix_ = pixConvertTo8(src, false);
+  } else {
+    pix_ = pixCopy(NULL, src);
   }
   depth = pixGetDepth(pix_);
   pix_channels_ = depth / 8;
   pix_wpl_ = pixGetWpl(pix_);
   scale_ = 1;
-  estimated_res_ = yres_ = pixGetYRes(src);
+  estimated_res_ = yres_ = pixGetYRes(pix_);
   Init();
 }
 
@@ -173,8 +181,11 @@ void ImageThresholder::SetImage(const Pix* pix) {
 // Caller must use pixDestroy to free the created Pix.
 void ImageThresholder::ThresholdToPix(PageSegMode pageseg_mode, Pix** pix) {
   if (pix_channels_ == 0) {
-    // We have a binary image, so it just has to be cloned.
+    // We have a binary image, but it still has to be copied, as this API
-    *pix = GetPixRect();
+    // allows the caller to modify the output.
+    Pix* original = GetPixRect();
+    *pix = pixCopy(nullptr, original);
+    pixDestroy(&original);
   } else {
     OtsuThresholdRectToPix(pix_, pix);
   }
@@ -257,10 +268,10 @@ void ImageThresholder::OtsuThresholdRectToPix(Pix* src_pix,
   OpenclDevice od;
   if ((num_channels == 4 || num_channels == 1) &&
       od.selectedDeviceIsOpenCL() && rect_top_ == 0 && rect_left_ == 0 ) {
-    od.ThresholdRectToPixOCL((const unsigned char*)pixGetData(src_pix),
+    od.ThresholdRectToPixOCL((unsigned char*)pixGetData(src_pix), num_channels,
-                             num_channels, pixGetWpl(src_pix) * 4,
+                             pixGetWpl(src_pix) * 4, thresholds, hi_values,
-                             thresholds, hi_values, out_pix /*pix_OCL*/,
+                             out_pix /*pix_OCL*/, rect_height_, rect_width_,
-                             rect_height_, rect_width_, rect_top_, rect_left_);
+                             rect_top_, rect_left_);
   } else {
 #endif
     ThresholdRectToPix(src_pix, num_channels, thresholds, hi_values, out_pix);

ccmain/thresholder.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_THRESHOLDER_H__
+#ifndef TESSERACT_CCMAIN_THRESHOLDER_H_
-#define TESSERACT_CCMAIN_THRESHOLDER_H__
+#define TESSERACT_CCMAIN_THRESHOLDER_H_
 
 #include "platform.h"
 #include "publictypes.h"
@@ -186,4 +186,4 @@ class TESS_API ImageThresholder {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_THRESHOLDER_H__
+#endif  // TESSERACT_CCMAIN_THRESHOLDER_H_

ccstruct/Makefile.am

@@ -12,7 +12,7 @@ endif
 include_HEADERS = publictypes.h
 noinst_HEADERS = \
     blamer.h blckerr.h blobbox.h blobs.h blread.h boxread.h boxword.h ccstruct.h coutln.h crakedge.h \
-    detlinefit.h dppoint.h fontinfo.h genblob.h hpdsizes.h \
+    debugpixa.h detlinefit.h dppoint.h fontinfo.h genblob.h hpdsizes.h \
     imagedata.h \
     ipoints.h \
     linlsq.h matrix.h mod128.h normalis.h \

ccstruct/blamer.cpp

@@ -317,7 +317,7 @@ void BlamerBundle::SetChopperBlame(const WERD_RES* word, bool debug) {
   int num_blobs = word->chopped_word->blobs.size();
   int box_index = 0;
   int blob_index = 0;
-  inT16 truth_x;
+  inT16 truth_x = -1;
   while (box_index < truth_word_.length() && blob_index < num_blobs) {
     truth_x = norm_truth_word_.BlobBox(box_index).right();
     TBLOB * curr_blob = word->chopped_word->blobs[blob_index];

ccstruct/blobbox.cpp

@@ -31,7 +31,9 @@
 #define PROJECTION_MARGIN 10     //arbitrary
 #define EXTERN
 
-ELISTIZE (BLOBNBOX) ELIST2IZE (TO_ROW) ELISTIZE (TO_BLOCK)
+ELISTIZE(BLOBNBOX)
+ELIST2IZE(TO_ROW)
+ELISTIZE(TO_BLOCK)
 
 // Up to 30 degrees is allowed for rotations of diacritic blobs.
 const double kCosSmallAngle = 0.866;
@@ -176,7 +178,7 @@ void BLOBNBOX::NeighbourGaps(int gaps[BND_COUNT]) const {
     gaps[dir] = MAX_INT16;
     BLOBNBOX* neighbour = neighbours_[dir];
     if (neighbour != NULL) {
-      TBOX n_box = neighbour->bounding_box();
+      const TBOX& n_box = neighbour->bounding_box();
       if (dir == BND_LEFT || dir == BND_RIGHT) {
         gaps[dir] = box.x_gap(n_box);
       } else {

ccstruct/blobs.cpp

@@ -815,12 +815,10 @@ void TWERD::BLNormalize(const BLOCK* block, const ROW* row, Pix* pix,
   float input_y_offset = 0.0f;
   float final_y_offset = static_cast<float>(kBlnBaselineOffset);
   float scale = kBlnXHeight / x_height;
-  if (hint == tesseract::OEM_CUBE_ONLY || row == NULL) {
+  if (row == NULL) {
     word_middle = word_box.left();
     input_y_offset = word_box.bottom();
     final_y_offset = 0.0f;
-    if (hint == tesseract::OEM_CUBE_ONLY)
-      scale = 1.0f;
   } else {
     input_y_offset = row->base_line(word_middle) + baseline_shift;
   }
@@ -834,7 +832,7 @@ void TWERD::BLNormalize(const BLOCK* block, const ROW* row, Pix* pix,
       baseline = blob_box.bottom();
       blob_scale = ClipToRange(kBlnXHeight * 4.0f / (3 * blob_box.height()),
                                scale, scale * 1.5f);
-    } else if (row != NULL && hint != tesseract::OEM_CUBE_ONLY) {
+    } else if (row != NULL) {
       baseline = row->base_line(mid_x) + baseline_shift;
     }
     // The image will be 8-bit grey if the input was grey or color. Note that in

ccstruct/boxread.cpp

@@ -34,8 +34,7 @@ FILE* OpenBoxFile(const STRING& fname) {
   STRING filename = BoxFileName(fname);
   FILE* box_file = NULL;
   if (!(box_file = fopen(filename.string(), "rb"))) {
-    CANTOPENFILE.error("read_next_box", TESSEXIT,
+    CANTOPENFILE.error("read_next_box", TESSEXIT, "Can't open box file %s",
-                       "Can't open box file %s",
                        filename.string());
   }
   return box_file;
@@ -56,6 +55,8 @@ bool ReadAllBoxes(int target_page, bool skip_blanks, const STRING& filename,
   GenericVector<char> box_data;
   if (!tesseract::LoadDataFromFile(BoxFileName(filename), &box_data))
     return false;
+  // Convert the array of bytes to a string, so it can be used by the parser.
+  box_data.push_back('\0');
   return ReadMemBoxes(target_page, skip_blanks, &box_data[0], boxes, texts,
                       box_texts, pages);
 }

ccstruct/boxread.h

@@ -17,8 +17,8 @@
  *
  **********************************************************************/
 
-#ifndef TESSERACT_CCUTIL_BOXREAD_H__
+#ifndef TESSERACT_CCUTIL_BOXREAD_H_
-#define TESSERACT_CCUTIL_BOXREAD_H__
+#define TESSERACT_CCUTIL_BOXREAD_H_
 
 #include <stdio.h>
 #include "genericvector.h"
@@ -82,4 +82,4 @@ bool ParseBoxFileStr(const char* boxfile_str, int* page_number,
 void MakeBoxFileStr(const char* unichar_str, const TBOX& box, int page_num,
                     STRING* box_str);
 
-#endif  // TESSERACT_CCUTIL_BOXREAD_H__
+#endif  // TESSERACT_CCUTIL_BOXREAD_H_

ccstruct/boxword.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CSTRUCT_BOXWORD_H__
+#ifndef TESSERACT_CSTRUCT_BOXWORD_H_
-#define TESSERACT_CSTRUCT_BOXWORD_H__
+#define TESSERACT_CSTRUCT_BOXWORD_H_
 
 #include "genericvector.h"
 #include "rect.h"
@@ -82,9 +82,7 @@ class BoxWord {
   const TBOX& bounding_box() const {
     return bbox_;
   }
-  int length() const {
+  int length() const { return length_; }
-    return length_;
-  }
   const TBOX& BlobBox(int index) const {
     return boxes_[index];
   }
@@ -99,5 +97,4 @@ class BoxWord {
 
 }  // namespace tesseract.
 
-
+#endif  // TESSERACT_CSTRUCT_BOXWORD_H_
-#endif  // TESSERACT_CSTRUCT_BOXWORD_H__

ccstruct/ccstruct.h

@@ -16,8 +16,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCSTRUCT_CCSTRUCT_H__
+#ifndef TESSERACT_CCSTRUCT_CCSTRUCT_H_
-#define TESSERACT_CCSTRUCT_CCSTRUCT_H__
+#define TESSERACT_CCSTRUCT_CCSTRUCT_H_
 
 #include "cutil.h"
 
@@ -40,5 +40,4 @@ class CCStruct : public CUtil {
 class Tesseract;
 }  // namespace tesseract
 
-
+#endif  // TESSERACT_CCSTRUCT_CCSTRUCT_H_
-#endif  // TESSERACT_CCSTRUCT_CCSTRUCT_H__

ccstruct/coutln.cpp

@@ -48,9 +48,9 @@ ICOORD C_OUTLINE::step_coords[4] = {
  * @param length length of loop
  */
 
-C_OUTLINE::C_OUTLINE (CRACKEDGE * startpt, ICOORD bot_left, 
+C_OUTLINE::C_OUTLINE(CRACKEDGE* startpt, ICOORD bot_left, ICOORD top_right,
-                      ICOORD top_right, inT16 length)
+                     inT16 length)
-    : box (bot_left, top_right), start (startpt->pos), offsets(NULL) {
+    : box(bot_left, top_right), start(startpt->pos), offsets(NULL) {
   inT16 stepindex;               //index to step
   CRACKEDGE *edgept;             //current point
 
@@ -71,7 +71,6 @@ C_OUTLINE::C_OUTLINE (CRACKEDGE * startpt, ICOORD bot_left,
   }
 }
 
-
 /**
  * @name C_OUTLINE::C_OUTLINE
  *
@@ -139,7 +138,7 @@ inT16 length                     //length of loop
  * @param rotation rotate to coord
  */
 
-C_OUTLINE::C_OUTLINE(C_OUTLINE *srcline, FCOORD rotation) : offsets(NULL) {
+C_OUTLINE::C_OUTLINE(C_OUTLINE* srcline, FCOORD rotation) : offsets(NULL) {
   TBOX new_box;                   //easy bounding
   inT16 stepindex;               //index to step
   inT16 dirdiff;                 //direction change
@@ -300,7 +299,6 @@ inT32 C_OUTLINE::perimeter() const {
   return total_steps;
 }
 
-
 /**
  * @name C_OUTLINE::outer_area
  *
@@ -332,7 +330,6 @@ inT32 C_OUTLINE::outer_area() const {
   return total;
 }
 
-
 /**
  * @name C_OUTLINE::count_transitions
  *
@@ -459,7 +456,6 @@ inT32 C_OUTLINE::count_transitions(inT32 threshold) {
   return total;
 }
 
-
 /**
  * @name C_OUTLINE::operator<
  *
@@ -468,8 +464,7 @@ inT32 C_OUTLINE::count_transitions(inT32 threshold) {
  */
 
 BOOL8
-C_OUTLINE::operator< (const C_OUTLINE & other) const
+C_OUTLINE::operator<(const C_OUTLINE& other) const {
-{
   inT16 count = 0;               //winding count
   ICOORD pos;                    //position of point
   inT32 stepindex;               //index to cstep
@@ -495,7 +490,6 @@ C_OUTLINE::operator< (const C_OUTLINE & other) const
   return count != 0;
 }
 
-
 /**
  * @name C_OUTLINE::winding_number
  *
@@ -534,7 +528,6 @@ inT16 C_OUTLINE::winding_number(ICOORD point) const {
   return count;                  //winding number
 }
 
-
 /**
  * C_OUTLINE::turn_direction
  *
@@ -563,7 +556,6 @@ inT16 C_OUTLINE::turn_direction() const {  //winding number
   return count;                  //winding number
 }
 
-
 /**
  * @name C_OUTLINE::reverse
  *
@@ -586,7 +578,6 @@ void C_OUTLINE::reverse() {  //reverse drection
   }
 }
 
-
 /**
  * @name C_OUTLINE::move
  *
@@ -661,14 +652,27 @@ static void ComputeGradient(const l_uint32* data, int wpl,
                             int x, int y, int width, int height,
                             ICOORD* gradient) {
   const l_uint32* line = data + y * wpl;
-  int pix_x_y = x < width && y < height ?
+  int pix_x_y =
-      GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line)), x) : 255;
+      x < width && y < height
-  int pix_x_prevy = x < width && y > 0 ?
+          ? GET_DATA_BYTE(
-      GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line - wpl)), x) : 255;
+                const_cast<void*>(reinterpret_cast<const void*>(line)), x)
-  int pix_prevx_prevy = x > 0 && y > 0 ?
+          : 255;
-      GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<void const*>(line - wpl)), x - 1) : 255;
+  int pix_x_prevy =
-  int pix_prevx_y = x > 0 && y < height ?
+      x < width && y > 0
-      GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line)), x - 1) : 255;
+          ? GET_DATA_BYTE(
+                const_cast<void*>(reinterpret_cast<const void*>(line - wpl)), x)
+          : 255;
+  int pix_prevx_prevy =
+      x > 0 && y > 0
+          ? GET_DATA_BYTE(
+                const_cast<void*>(reinterpret_cast<void const*>(line - wpl)),
+                x - 1)
+          : 255;
+  int pix_prevx_y =
+      x > 0 && y < height
+          ? GET_DATA_BYTE(
+                const_cast<void*>(reinterpret_cast<const void*>(line)), x - 1)
+          : 255;
   gradient->set_x(pix_x_y + pix_x_prevy - (pix_prevx_y + pix_prevx_prevy));
   gradient->set_y(pix_x_prevy + pix_prevx_prevy - (pix_x_y + pix_prevx_y));
 }
@@ -684,8 +688,10 @@ static bool EvaluateVerticalDiff(const l_uint32* data, int wpl, int diff_sign,
   if (y <= 0 || y >= height)
     return false;
   const l_uint32* line = data + y * wpl;
-  int pixel1 = GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line - wpl)), x);
+  int pixel1 = GET_DATA_BYTE(
-  int pixel2 = GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line)), x);
+      const_cast<void*>(reinterpret_cast<const void*>(line - wpl)), x);
+  int pixel2 =
+      GET_DATA_BYTE(const_cast<void*>(reinterpret_cast<const void*>(line)), x);
   int diff = (pixel2 - pixel1) * diff_sign;
   if (diff > *best_diff) {
     *best_diff = diff;
@@ -705,8 +711,10 @@ static bool EvaluateHorizontalDiff(const l_uint32* line, int diff_sign,
                                    int* best_diff, int* best_sum, int* best_x) {
   if (x <= 0 || x >= width)
     return false;
-  int pixel1 = GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line)), x - 1);
+  int pixel1 = GET_DATA_BYTE(
-  int pixel2 = GET_DATA_BYTE(const_cast<void*> (reinterpret_cast<const void *>(line)), x);
+      const_cast<void*>(reinterpret_cast<const void*>(line)), x - 1);
+  int pixel2 =
+      GET_DATA_BYTE(const_cast<void*>(reinterpret_cast<const void*>(line)), x);
   int diff = (pixel2 - pixel1) * diff_sign;
   if (diff > *best_diff) {
     *best_diff = diff;
@@ -954,8 +962,7 @@ void C_OUTLINE::render_outline(int left, int top, Pix* pix) const {
  */
 
 #ifndef GRAPHICS_DISABLED
-void C_OUTLINE::plot(ScrollView* window,
+void C_OUTLINE::plot(ScrollView* window, ScrollView::Color colour) const {
-                     ScrollView::Color colour) const {
   inT16 stepindex;               // index to cstep
   ICOORD pos;                    // current position
   DIR128 stepdir;                // direction of step
@@ -1016,7 +1023,6 @@ void C_OUTLINE::plot_normed(const DENORM& denorm, ScrollView::Color colour,
 }
 #endif
 
-
 /**
  * @name C_OUTLINE::operator=
  *
@@ -1024,7 +1030,7 @@ void C_OUTLINE::plot_normed(const DENORM& denorm, ScrollView::Color colour,
  * @param source assign from this
  */
 
-C_OUTLINE & C_OUTLINE::operator= (const C_OUTLINE & source) {
+C_OUTLINE& C_OUTLINE::operator=(const C_OUTLINE& source) {
   box = source.box;
   start = source.start;
   if (steps != NULL)

ccstruct/debugpixa.h

@@ -0,0 +1,52 @@
+#ifndef TESSERACT_CCSTRUCT_DEBUGPIXA_H_
+#define TESSERACT_CCSTRUCT_DEBUGPIXA_H_
+
+#include "allheaders.h"
+
+namespace tesseract {
+
+// Class to hold a Pixa collection of debug images with captions and save them
+// to a PDF file.
+class DebugPixa {
+ public:
+  // TODO(rays) add another constructor with size control.
+  DebugPixa() {
+    pixa_ = pixaCreate(0);
+    fonts_ = bmfCreate(nullptr, 14);
+  }
+  // If the filename_ has been set and there are any debug images, they are
+  // written to the set filename_.
+  ~DebugPixa() {
+    pixaDestroy(&pixa_);
+    bmfDestroy(&fonts_);
+  }
+
+  // Adds the given pix to the set of pages in the PDF file, with the given
+  // caption added to the top.
+  void AddPix(const Pix* pix, const char* caption) {
+    int depth = pixGetDepth(const_cast<Pix*>(pix));
+    int color = depth < 8 ? 1 : (depth > 8 ? 0x00ff0000 : 0x80);
+    Pix* pix_debug = pixAddSingleTextblock(
+        const_cast<Pix*>(pix), fonts_, caption, color, L_ADD_BELOW, nullptr);
+    pixaAddPix(pixa_, pix_debug, L_INSERT);
+  }
+
+  // Sets the destination filename and enables images to be written to a PDF
+  // on destruction.
+  void WritePDF(const char* filename) {
+    if (pixaGetCount(pixa_) > 0) {
+      pixaConvertToPdf(pixa_, 300, 1.0f, 0, 0, "AllDebugImages", filename);
+      pixaClear(pixa_);
+    }
+  }
+
+ private:
+  // The collection of images to put in the PDF.
+  Pixa* pixa_;
+  // The fonts used to draw text captions.
+  L_Bmf* fonts_;
+};
+
+}  // namespace tesseract
+
+#endif  // TESSERACT_CCSTRUCT_DEBUGPIXA_H_

ccstruct/dppoint.h

@@ -17,8 +17,8 @@
  *
  **********************************************************************/
 
-#ifndef TESSERACT_CCSTRUCT_DPPOINT_H__
+#ifndef TESSERACT_CCSTRUCT_DPPOINT_H_
-#define TESSERACT_CCSTRUCT_DPPOINT_H__
+#define TESSERACT_CCSTRUCT_DPPOINT_H_
 
 #include "host.h"
 
@@ -98,5 +98,4 @@ class DPPoint {
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCSTRUCT_DPPOINT_H__
+#endif  // TESSERACT_CCSTRUCT_DPPOINT_H_
-

ccstruct/fontinfo.cpp

@@ -31,7 +31,7 @@ bool FontInfo::Serialize(FILE* fp) const {
 }
 // Reads from the given file. Returns false in case of error.
 // If swap is true, assumes a big/little-endian swap is needed.
-bool FontInfo::DeSerialize(bool swap, FILE* fp) {
+bool FontInfo::DeSerialize(bool swap, TFile* fp) {
   if (!read_info(fp, this, swap)) return false;
   if (!read_spacing_info(fp, this, swap)) return false;
   return true;
@@ -51,7 +51,7 @@ bool FontInfoTable::Serialize(FILE* fp) const {
 }
 // Reads from the given file. Returns false in case of error.
 // If swap is true, assumes a big/little-endian swap is needed.
-bool FontInfoTable::DeSerialize(bool swap, FILE* fp) {
+bool FontInfoTable::DeSerialize(bool swap, TFile* fp) {
   truncate(0);
   return this->DeSerializeClasses(swap, fp);
 }
@@ -149,19 +149,15 @@ void FontSetDeleteCallback(FontSet fs) {
 
 /*---------------------------------------------------------------------------*/
 // Callbacks used by UnicityTable to read/write FontInfo/FontSet structures.
-bool read_info(FILE* f, FontInfo* fi, bool swap) {
+bool read_info(TFile* f, FontInfo* fi, bool swap) {
   inT32 size;
-  if (fread(&size, sizeof(size), 1, f) != 1) return false;
+  if (f->FReadEndian(&size, sizeof(size), 1, swap) != 1) return false;
-  if (swap)
-    Reverse32(&size);
   char* font_name = new char[size + 1];
   fi->name = font_name;
-  if (static_cast<int>(fread(font_name, sizeof(*font_name), size, f)) != size)
+  if (f->FRead(font_name, sizeof(*font_name), size) != size) return false;
-    return false;
   font_name[size] = '\0';
-  if (fread(&fi->properties, sizeof(fi->properties), 1, f) != 1) return false;
+  if (f->FReadEndian(&fi->properties, sizeof(fi->properties), 1, swap) != 1)
-  if (swap)
+    return false;
-    Reverse32(&fi->properties);
   return true;
 }
 
@@ -174,26 +170,22 @@ bool write_info(FILE* f, const FontInfo& fi) {
   return true;
 }
 
-bool read_spacing_info(FILE *f, FontInfo* fi, bool swap) {
+bool read_spacing_info(TFile* f, FontInfo* fi, bool swap) {
   inT32 vec_size, kern_size;
-  if (fread(&vec_size, sizeof(vec_size), 1, f) != 1) return false;
+  if (f->FReadEndian(&vec_size, sizeof(vec_size), 1, swap) != 1) return false;
-  if (swap) Reverse32(&vec_size);
   ASSERT_HOST(vec_size >= 0);
   if (vec_size == 0) return true;
   fi->init_spacing(vec_size);
   for (int i = 0; i < vec_size; ++i) {
     FontSpacingInfo *fs = new FontSpacingInfo();
-    if (fread(&fs->x_gap_before, sizeof(fs->x_gap_before), 1, f) != 1 ||
+    if (f->FReadEndian(&fs->x_gap_before, sizeof(fs->x_gap_before), 1, swap) !=
-        fread(&fs->x_gap_after, sizeof(fs->x_gap_after), 1, f) != 1 ||
+            1 ||
-        fread(&kern_size, sizeof(kern_size), 1, f) != 1) {
+        f->FReadEndian(&fs->x_gap_after, sizeof(fs->x_gap_after), 1, swap) !=
+            1 ||
+        f->FReadEndian(&kern_size, sizeof(kern_size), 1, swap) != 1) {
       delete fs;
       return false;
     }
-    if (swap) {
-      ReverseN(&(fs->x_gap_before), sizeof(fs->x_gap_before));
-      ReverseN(&(fs->x_gap_after), sizeof(fs->x_gap_after));
-      Reverse32(&kern_size);
-    }
     if (kern_size < 0) {  // indication of a NULL entry in fi->spacing_vec
       delete fs;
       continue;
@@ -237,16 +229,12 @@ bool write_spacing_info(FILE* f, const FontInfo& fi) {
   return true;
 }
 
-bool read_set(FILE* f, FontSet* fs, bool swap) {
+bool read_set(TFile* f, FontSet* fs, bool swap) {
-  if (fread(&fs->size, sizeof(fs->size), 1, f) != 1) return false;
+  if (f->FReadEndian(&fs->size, sizeof(fs->size), 1, swap) != 1) return false;
-  if (swap)
-    Reverse32(&fs->size);
   fs->configs = new int[fs->size];
-  for (int i = 0; i < fs->size; ++i) {
+  if (f->FReadEndian(fs->configs, sizeof(fs->configs[0]), fs->size, swap) !=
-    if (fread(&fs->configs[i], sizeof(fs->configs[i]), 1, f) != 1) return false;
+      fs->size)
-    if (swap)
+    return false;
-      Reverse32(&fs->configs[i]);
-  }
   return true;
 }
 

ccstruct/fontinfo.h

@@ -67,7 +67,7 @@ struct FontInfo {
   bool Serialize(FILE* fp) const;
   // Reads from the given file. Returns false in case of error.
   // If swap is true, assumes a big/little-endian swap is needed.
-  bool DeSerialize(bool swap, FILE* fp);
+  bool DeSerialize(bool swap, TFile* fp);
 
   // Reserves unicharset_size spots in spacing_vec.
   void init_spacing(int unicharset_size) {
@@ -152,7 +152,7 @@ class FontInfoTable : public GenericVector<FontInfo> {
   bool Serialize(FILE* fp) const;
   // Reads from the given file. Returns false in case of error.
   // If swap is true, assumes a big/little-endian swap is needed.
-  bool DeSerialize(bool swap, FILE* fp);
+  bool DeSerialize(bool swap, TFile* fp);
 
   // Returns true if the given set of fonts includes one with the same
   // properties as font_id.
@@ -177,11 +177,11 @@ void FontInfoDeleteCallback(FontInfo f);
 void FontSetDeleteCallback(FontSet fs);
 
 // Callbacks used by UnicityTable to read/write FontInfo/FontSet structures.
-bool read_info(FILE* f, FontInfo* fi, bool swap);
+bool read_info(TFile* f, FontInfo* fi, bool swap);
 bool write_info(FILE* f, const FontInfo& fi);
-bool read_spacing_info(FILE *f, FontInfo* fi, bool swap);
+bool read_spacing_info(TFile* f, FontInfo* fi, bool swap);
 bool write_spacing_info(FILE* f, const FontInfo& fi);
-bool read_set(FILE* f, FontSet* fs, bool swap);
+bool read_set(TFile* f, FontSet* fs, bool swap);
 bool write_set(FILE* f, const FontSet& fs);
 
 }  // namespace tesseract.

ccstruct/hpdsizes.h

@@ -1,3 +1,12 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+// http://www.apache.org/licenses/LICENSE-2.0
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
 #ifndef           HPDSIZES_H
 #define           HPDSIZES_H
 

ccstruct/imagedata.cpp

@@ -24,12 +24,22 @@
 
 #include "imagedata.h"
 
+#if defined(__MINGW32__)
+#include <unistd.h>
+#else
+#include <thread>
+#endif
+
 #include "allheaders.h"
 #include "boxread.h"
 #include "callcpp.h"
 #include "helpers.h"
 #include "tprintf.h"
 
+// Number of documents to read ahead while training. Doesn't need to be very
+// large.
+const int kMaxReadAhead = 8;
+
 namespace tesseract {
 
 WordFeature::WordFeature() : x_(0), y_(0), dir_(0) {
@@ -182,6 +192,19 @@ bool ImageData::DeSerialize(bool swap, TFile* fp) {
   return true;
 }
 
+// As DeSerialize, but only seeks past the data - hence a static method.
+bool ImageData::SkipDeSerialize(bool swap, TFile* fp) {
+  if (!STRING::SkipDeSerialize(swap, fp)) return false;
+  inT32 page_number;
+  if (fp->FRead(&page_number, sizeof(page_number), 1) != 1) return false;
+  if (!GenericVector<char>::SkipDeSerialize(swap, fp)) return false;
+  if (!STRING::SkipDeSerialize(swap, fp)) return false;
+  if (!GenericVector<TBOX>::SkipDeSerialize(swap, fp)) return false;
+  if (!GenericVector<STRING>::SkipDeSerializeClasses(swap, fp)) return false;
+  inT8 vertical = 0;
+  return fp->FRead(&vertical, sizeof(vertical), 1) == 1;
+}
+
 // Saves the given Pix as a PNG-encoded string and destroys it.
 void ImageData::SetPix(Pix* pix) {
   SetPixInternal(pix, &image_data_);
@@ -195,9 +218,10 @@ Pix* ImageData::GetPix() const {
 // Gets anything and everything with a non-NULL pointer, prescaled to a
 // given target_height (if 0, then the original image height), and aligned.
 // Also returns (if not NULL) the width and height of the scaled image.
-// The return value is the scale factor that was applied to the image to
+// The return value is the scaled Pix, which must be pixDestroyed after use,
-// achieve the target_height.
+// and scale_factor (if not NULL) is set to the scale factor that was applied
-float ImageData::PreScale(int target_height, Pix** pix,
+// to the image to achieve the target_height.
+Pix* ImageData::PreScale(int target_height, int max_height, float* scale_factor,
                          int* scaled_width, int* scaled_height,
                          GenericVector<TBOX>* boxes) const {
   int input_width = 0;
@@ -206,26 +230,22 @@ float ImageData::PreScale(int target_height, Pix** pix,
   ASSERT_HOST(src_pix != NULL);
   input_width = pixGetWidth(src_pix);
   input_height = pixGetHeight(src_pix);
-  if (target_height == 0)
+  if (target_height == 0) {
-    target_height = input_height;
+    target_height = MIN(input_height, max_height);
+  }
   float im_factor = static_cast<float>(target_height) / input_height;
   if (scaled_width != NULL)
     *scaled_width = IntCastRounded(im_factor * input_width);
   if (scaled_height != NULL)
     *scaled_height = target_height;
-  if (pix != NULL) {
   // Get the scaled image.
-    pixDestroy(pix);
+  Pix* pix = pixScale(src_pix, im_factor, im_factor);
-    *pix = pixScale(src_pix, im_factor, im_factor);
+  if (pix == NULL) {
-    if (*pix == NULL) {
     tprintf("Scaling pix of size %d, %d by factor %g made null pix!!\n",
             input_width, input_height, im_factor);
   }
-    if (scaled_width != NULL)
+  if (scaled_width != NULL) *scaled_width = pixGetWidth(pix);
-      *scaled_width = pixGetWidth(*pix);
+  if (scaled_height != NULL) *scaled_height = pixGetHeight(pix);
-    if (scaled_height != NULL)
-      *scaled_height = pixGetHeight(*pix);
-  }
   pixDestroy(&src_pix);
   if (boxes != NULL) {
     // Get the boxes.
@@ -241,7 +261,8 @@ float ImageData::PreScale(int target_height, Pix** pix,
       boxes->push_back(box);
     }
   }
-  return im_factor;
+  if (scale_factor != NULL) *scale_factor = im_factor;
+  return pix;
 }
 
 int ImageData::MemoryUsed() const {
@@ -266,19 +287,20 @@ void ImageData::Display() const {
   // Draw the boxes.
   win->Pen(ScrollView::RED);
   win->Brush(ScrollView::NONE);
-  win->TextAttributes("Arial", kTextSize, false, false, false);
+  int text_size = kTextSize;
+  if (!boxes_.empty() && boxes_[0].height() * 2 < text_size)
+    text_size = boxes_[0].height() * 2;
+  win->TextAttributes("Arial", text_size, false, false, false);
+  if (!boxes_.empty()) {
     for (int b = 0; b < boxes_.size(); ++b) {
       boxes_[b].plot(win);
       win->Text(boxes_[b].left(), height + kTextSize, box_texts_[b].string());
-    TBOX scaled(boxes_[b]);
-    scaled.scale(256.0 / height);
-    scaled.plot(win);
     }
+  } else {
     // The full transcription.
     win->Pen(ScrollView::CYAN);
     win->Text(0, height + kTextSize * 2, transcription_.string());
-  // Add the features.
+  }
-  win->Pen(ScrollView::GREEN);
   win->Update();
   window_wait(win);
 #endif
@@ -340,27 +362,51 @@ bool ImageData::AddBoxes(const char* box_text) {
   return false;
 }
 
-DocumentData::DocumentData(const STRING& name)
+// Thread function to call ReCachePages.
-  : document_name_(name), pages_offset_(0), total_pages_(0),
+void* ReCachePagesFunc(void* data) {
-    memory_used_(0), max_memory_(0), reader_(NULL) {}
+  DocumentData* document_data = reinterpret_cast<DocumentData*>(data);
+  document_data->ReCachePages();
+  return NULL;
+}
 
-DocumentData::~DocumentData() {}
+DocumentData::DocumentData(const STRING& name)
+    : document_name_(name),
+      pages_offset_(-1),
+      total_pages_(-1),
+      memory_used_(0),
+      max_memory_(0),
+      reader_(NULL) {}
+
+DocumentData::~DocumentData() {
+  SVAutoLock lock_p(&pages_mutex_);
+  SVAutoLock lock_g(&general_mutex_);
+}
 
 // Reads all the pages in the given lstmf filename to the cache. The reader
 // is used to read the file.
 bool DocumentData::LoadDocument(const char* filename, const char* lang,
                                 int start_page, inT64 max_memory,
                                 FileReader reader) {
+  SetDocument(filename, lang, max_memory, reader);
+  pages_offset_ = start_page;
+  return ReCachePages();
+}
+
+// Sets up the document, without actually loading it.
+void DocumentData::SetDocument(const char* filename, const char* lang,
+                               inT64 max_memory, FileReader reader) {
+  SVAutoLock lock_p(&pages_mutex_);
+  SVAutoLock lock(&general_mutex_);
   document_name_ = filename;
   lang_ = lang;
-  pages_offset_ = start_page;
+  pages_offset_ = -1;
   max_memory_ = max_memory;
   reader_ = reader;
-  return ReCachePages();
 }
 
 // Writes all the pages to the given filename. Returns false on error.
 bool DocumentData::SaveDocument(const char* filename, FileWriter writer) {
+  SVAutoLock lock(&pages_mutex_);
   TFile fp;
   fp.OpenWrite(NULL);
   if (!pages_.Serialize(&fp) || !fp.CloseWrite(filename, writer)) {
@@ -370,112 +416,184 @@ bool DocumentData::SaveDocument(const char* filename, FileWriter writer) {
   return true;
 }
 bool DocumentData::SaveToBuffer(GenericVector<char>* buffer) {
+  SVAutoLock lock(&pages_mutex_);
   TFile fp;
   fp.OpenWrite(buffer);
   return pages_.Serialize(&fp);
 }
 
-// Returns a pointer to the page with the given index, modulo the total
+// Adds the given page data to this document, counting up memory.
-// number of pages, recaching if needed.
+void DocumentData::AddPageToDocument(ImageData* page) {
-const ImageData* DocumentData::GetPage(int index) {
+  SVAutoLock lock(&pages_mutex_);
-  index = Modulo(index, total_pages_);
+  pages_.push_back(page);
-  if (index < pages_offset_ || index >= pages_offset_ + pages_.size()) {
+  set_memory_used(memory_used() + page->MemoryUsed());
-    pages_offset_ = index;
-    if (!ReCachePages()) return NULL;
-  }
-  return pages_[index - pages_offset_];
 }
 
-// Loads as many pages can fit in max_memory_ starting at index pages_offset_.
+// If the given index is not currently loaded, loads it using a separate
+// thread.
+void DocumentData::LoadPageInBackground(int index) {
+  ImageData* page = NULL;
+  if (IsPageAvailable(index, &page)) return;
+  SVAutoLock lock(&pages_mutex_);
+  if (pages_offset_ == index) return;
+  pages_offset_ = index;
+  pages_.clear();
+  SVSync::StartThread(ReCachePagesFunc, this);
+}
+
+// Returns a pointer to the page with the given index, modulo the total
+// number of pages. Blocks until the background load is completed.
+const ImageData* DocumentData::GetPage(int index) {
+  ImageData* page = NULL;
+  while (!IsPageAvailable(index, &page)) {
+    // If there is no background load scheduled, schedule one now.
+    pages_mutex_.Lock();
+    bool needs_loading = pages_offset_ != index;
+    pages_mutex_.Unlock();
+    if (needs_loading) LoadPageInBackground(index);
+    // We can't directly load the page, or the background load will delete it
+    // while the caller is using it, so give it a chance to work.
+#if defined(__MINGW32__)
+    sleep(1);
+#else
+    std::this_thread::sleep_for(std::chrono::seconds(1));
+#endif
+  }
+  return page;
+}
+
+// Returns true if the requested page is available, and provides a pointer,
+// which may be NULL if the document is empty. May block, even though it
+// doesn't guarantee to return true.
+bool DocumentData::IsPageAvailable(int index, ImageData** page) {
+  SVAutoLock lock(&pages_mutex_);
+  int num_pages = NumPages();
+  if (num_pages == 0 || index < 0) {
+    *page = NULL;  // Empty Document.
+    return true;
+  }
+  if (num_pages > 0) {
+    index = Modulo(index, num_pages);
+    if (pages_offset_ <= index && index < pages_offset_ + pages_.size()) {
+      *page = pages_[index - pages_offset_];  // Page is available already.
+      return true;
+    }
+  }
+  return false;
+}
+
+// Removes all pages from memory and frees the memory, but does not forget
+// the document metadata.
+inT64 DocumentData::UnCache() {
+  SVAutoLock lock(&pages_mutex_);
+  inT64 memory_saved = memory_used();
+  pages_.clear();
+  pages_offset_ = -1;
+  set_total_pages(-1);
+  set_memory_used(0);
+  tprintf("Unloaded document %s, saving %d memory\n", document_name_.string(),
+          memory_saved);
+  return memory_saved;
+}
+
+// Shuffles all the pages in the document.
+void DocumentData::Shuffle() {
+  TRand random;
+  // Different documents get shuffled differently, but the same for the same
+  // name.
+  random.set_seed(document_name_.string());
+  int num_pages = pages_.size();
+  // Execute one random swap for each page in the document.
+  for (int i = 0; i < num_pages; ++i) {
+    int src = random.IntRand() % num_pages;
+    int dest = random.IntRand() % num_pages;
+    std::swap(pages_[src], pages_[dest]);
+  }
+}
+
+// Locks the pages_mutex_ and Loads as many pages can fit in max_memory_
+// starting at index pages_offset_.
 bool DocumentData::ReCachePages() {
+  SVAutoLock lock(&pages_mutex_);
   // Read the file.
-  TFile fp;
+  set_total_pages(0);
-  if (!fp.Open(document_name_, reader_)) return false;
+  set_memory_used(0);
-  memory_used_ = 0;
+  int loaded_pages = 0;
-  if (!pages_.DeSerialize(false, &fp)) {
-    tprintf("Deserialize failed: %s\n", document_name_.string());
   pages_.truncate(0);
+  TFile fp;
+  if (!fp.Open(document_name_, reader_) ||
+      !PointerVector<ImageData>::DeSerializeSize(false, &fp, &loaded_pages) ||
+      loaded_pages <= 0) {
+    tprintf("Deserialize header failed: %s\n", document_name_.string());
     return false;
   }
-  total_pages_ = pages_.size();
+  pages_offset_ %= loaded_pages;
-  pages_offset_ %= total_pages_;
+  // Skip pages before the first one we want, and load the rest until max
-  // Delete pages before the first one we want, and relocate the rest.
+  // memory and skip the rest after that.
   int page;
-  for (page = 0; page < pages_.size(); ++page) {
+  for (page = 0; page < loaded_pages; ++page) {
-    if (page < pages_offset_) {
+    if (page < pages_offset_ ||
-      delete pages_[page];
+        (max_memory_ > 0 && memory_used() > max_memory_)) {
-      pages_[page] = NULL;
+      if (!PointerVector<ImageData>::DeSerializeSkip(false, &fp)) break;
     } else {
-      ImageData* image_data = pages_[page];
+      if (!pages_.DeSerializeElement(false, &fp)) break;
-      if (max_memory_ > 0 && page > pages_offset_ &&
+      ImageData* image_data = pages_.back();
-          memory_used_ + image_data->MemoryUsed() > max_memory_)
-        break;  // Don't go over memory quota unless the first image.
       if (image_data->imagefilename().length() == 0) {
         image_data->set_imagefilename(document_name_);
         image_data->set_page_number(page);
       }
       image_data->set_language(lang_);
-      memory_used_ += image_data->MemoryUsed();
+      set_memory_used(memory_used() + image_data->MemoryUsed());
-      if (pages_offset_ != 0) {
-        pages_[page - pages_offset_] = image_data;
-        pages_[page] = NULL;
     }
   }
+  if (page < loaded_pages) {
+    tprintf("Deserialize failed: %s read %d/%d pages\n",
+            document_name_.string(), page, loaded_pages);
+    pages_.truncate(0);
+  } else {
+    tprintf("Loaded %d/%d pages (%d-%d) of document %s\n", pages_.size(),
+            loaded_pages, pages_offset_ + 1, pages_offset_ + pages_.size(),
+            document_name_.string());
   }
-  pages_.truncate(page - pages_offset_);
+  set_total_pages(loaded_pages);
-  tprintf("Loaded %d/%d pages (%d-%d) of document %s\n",
-          pages_.size(), total_pages_, pages_offset_,
-          pages_offset_ + pages_.size(), document_name_.string());
   return !pages_.empty();
 }
 
-// Adds the given page data to this document, counting up memory.
-void DocumentData::AddPageToDocument(ImageData* page) {
-  pages_.push_back(page);
-  memory_used_ += page->MemoryUsed();
-}
-
 // A collection of DocumentData that knows roughly how much memory it is using.
 DocumentCache::DocumentCache(inT64 max_memory)
-  : total_pages_(0), memory_used_(0), max_memory_(max_memory) {}
+    : num_pages_per_doc_(0), max_memory_(max_memory) {}
 DocumentCache::~DocumentCache() {}
 
 // Adds all the documents in the list of filenames, counting memory.
 // The reader is used to read the files.
 bool DocumentCache::LoadDocuments(const GenericVector<STRING>& filenames,
-                                  const char* lang, FileReader reader) {
+                                  const char* lang,
-  inT64 fair_share_memory = max_memory_ / filenames.size();
+                                  CachingStrategy cache_strategy,
+                                  FileReader reader) {
+  cache_strategy_ = cache_strategy;
+  inT64 fair_share_memory = 0;
+  // In the round-robin case, each DocumentData handles restricting its content
+  // to its fair share of memory. In the sequential case, DocumentCache
+  // determines which DocumentDatas are held entirely in memory.
+  if (cache_strategy_ == CS_ROUND_ROBIN)
+    fair_share_memory = max_memory_ / filenames.size();
   for (int arg = 0; arg < filenames.size(); ++arg) {
     STRING filename = filenames[arg];
     DocumentData* document = new DocumentData(filename);
-    if (document->LoadDocument(filename.string(), lang, 0,
+    document->SetDocument(filename.string(), lang, fair_share_memory, reader);
-                               fair_share_memory, reader)) {
     AddToCache(document);
-    } else {
-      tprintf("Failed to load image %s!\n", filename.string());
-      delete document;
   }
+  if (!documents_.empty()) {
+    // Try to get the first page now to verify the list of filenames.
+    if (GetPageBySerial(0) != NULL) return true;
+    tprintf("Load of page 0 failed!\n");
   }
-  tprintf("Loaded %d pages, total %gMB\n",
+  return false;
-          total_pages_, memory_used_ / 1048576.0);
-  return total_pages_ > 0;
 }
 
-// Adds document to the cache, throwing out other documents if needed.
+// Adds document to the cache.
 bool DocumentCache::AddToCache(DocumentData* data) {
-  inT64 new_memory = data->memory_used();
-  memory_used_ += new_memory;
   documents_.push_back(data);
-  total_pages_ += data->NumPages();
-  // Delete the first item in the array, and other pages of the same name
-  // while memory is full.
-  while (memory_used_ >= max_memory_ && max_memory_ > 0) {
-    tprintf("Memory used=%lld vs max=%lld, discarding doc of size %lld\n",
-            memory_used_ , max_memory_, documents_[0]->memory_used());
-    memory_used_ -= documents_[0]->memory_used();
-    total_pages_ -= documents_[0]->NumPages();
-    documents_.remove(0);
-  }
   return true;
 }
 
@@ -488,11 +606,104 @@ DocumentData* DocumentCache::FindDocument(const STRING& document_name) const {
   return NULL;
 }
 
+// Returns the total number of pages in an epoch. For CS_ROUND_ROBIN cache
+// strategy, could take a long time.
+int DocumentCache::TotalPages() {
+  if (cache_strategy_ == CS_SEQUENTIAL) {
+    // In sequential mode, we assume each doc has the same number of pages
+    // whether it is true or not.
+    if (num_pages_per_doc_ == 0) GetPageSequential(0);
+    return num_pages_per_doc_ * documents_.size();
+  }
+  int total_pages = 0;
+  int num_docs = documents_.size();
+  for (int d = 0; d < num_docs; ++d) {
+    // We have to load a page to make NumPages() valid.
+    documents_[d]->GetPage(0);
+    total_pages += documents_[d]->NumPages();
+  }
+  return total_pages;
+}
+
 // Returns a page by serial number, selecting them in a round-robin fashion
-// from all the documents.
+// from all the documents. Highly disk-intensive, but doesn't need samples
-const ImageData* DocumentCache::GetPageBySerial(int serial) {
+// to be shuffled between files to begin with.
-  int document_index = serial % documents_.size();
+const ImageData* DocumentCache::GetPageRoundRobin(int serial) {
-  return documents_[document_index]->GetPage(serial / documents_.size());
+  int num_docs = documents_.size();
+  int doc_index = serial % num_docs;
+  const ImageData* doc = documents_[doc_index]->GetPage(serial / num_docs);
+  for (int offset = 1; offset <= kMaxReadAhead && offset < num_docs; ++offset) {
+    doc_index = (serial + offset) % num_docs;
+    int page = (serial + offset) / num_docs;
+    documents_[doc_index]->LoadPageInBackground(page);
+  }
+  return doc;
+}
+
+// Returns a page by serial number, selecting them in sequence from each file.
+// Requires the samples to be shuffled between the files to give a random or
+// uniform distribution of data. Less disk-intensive than GetPageRoundRobin.
+const ImageData* DocumentCache::GetPageSequential(int serial) {
+  int num_docs = documents_.size();
+  ASSERT_HOST(num_docs > 0);
+  if (num_pages_per_doc_ == 0) {
+    // Use the pages in the first doc as the number of pages in each doc.
+    documents_[0]->GetPage(0);
+    num_pages_per_doc_ = documents_[0]->NumPages();
+    if (num_pages_per_doc_ == 0) {
+      tprintf("First document cannot be empty!!\n");
+      ASSERT_HOST(num_pages_per_doc_ > 0);
+    }
+    // Get rid of zero now if we don't need it.
+    if (serial / num_pages_per_doc_ % num_docs > 0) documents_[0]->UnCache();
+  }
+  int doc_index = serial / num_pages_per_doc_ % num_docs;
+  const ImageData* doc =
+      documents_[doc_index]->GetPage(serial % num_pages_per_doc_);
+  // Count up total memory. Background loading makes it more complicated to
+  // keep a running count.
+  inT64 total_memory = 0;
+  for (int d = 0; d < num_docs; ++d) {
+    total_memory += documents_[d]->memory_used();
+  }
+  if (total_memory >= max_memory_) {
+    // Find something to un-cache.
+    // If there are more than 3 in front, then serial is from the back reader
+    // of a pair of readers. If we un-cache from in-front-2 to 2-ahead, then
+    // we create a hole between them and then un-caching the backmost occupied
+    // will work for both.
+    int num_in_front = CountNeighbourDocs(doc_index, 1);
+    for (int offset = num_in_front - 2;
+         offset > 1 && total_memory >= max_memory_; --offset) {
+      int next_index = (doc_index + offset) % num_docs;
+      total_memory -= documents_[next_index]->UnCache();
+    }
+    // If that didn't work, the best solution is to un-cache from the back. If
+    // we take away the document that a 2nd reader is using, it will put it
+    // back and make a hole between.
+    int num_behind = CountNeighbourDocs(doc_index, -1);
+    for (int offset = num_behind; offset < 0 && total_memory >= max_memory_;
+         ++offset) {
+      int next_index = (doc_index + offset + num_docs) % num_docs;
+      total_memory -= documents_[next_index]->UnCache();
+    }
+  }
+  int next_index = (doc_index + 1) % num_docs;
+  if (!documents_[next_index]->IsCached() && total_memory < max_memory_) {
+    documents_[next_index]->LoadPageInBackground(0);
+  }
+  return doc;
+}
+
+// Helper counts the number of adjacent cached neighbours of index looking in
+// direction dir, ie index+dir, index+2*dir etc.
+int DocumentCache::CountNeighbourDocs(int index, int dir) {
+  int num_docs = documents_.size();
+  for (int offset = dir; abs(offset) < num_docs; offset += dir) {
+    int offset_index = (index + offset + num_docs) % num_docs;
+    if (!documents_[offset_index]->IsCached()) return offset - dir;
+  }
+  return num_docs;
 }
 
 }  // namespace tesseract.

ccstruct/imagedata.h

@@ -25,6 +25,7 @@
 #include "normalis.h"
 #include "rect.h"
 #include "strngs.h"
+#include "svutil.h"
 
 struct Pix;
 
@@ -34,8 +35,22 @@ namespace tesseract {
 const int kFeaturePadding = 2;
 // Number of pixels to pad around text boxes.
 const int kImagePadding = 4;
-// Number of training images to combine into a mini-batch for training.
+
-const int kNumPagesPerMiniBatch = 100;
+// Enum to determine the caching and data sequencing strategy.
+enum CachingStrategy {
+  // Reads all of one file before moving on to the next. Requires samples to be
+  // shuffled across files. Uses the count of samples in the first file as
+  // the count in all the files to achieve high-speed random access. As a
+  // consequence, if subsequent files are smaller, they get entries used more
+  // than once, and if subsequent files are larger, some entries are not used.
+  // Best for larger data sets that don't fit in memory.
+  CS_SEQUENTIAL,
+  // Reads one sample from each file in rotation. Does not require shuffled
+  // samples, but is extremely disk-intensive. Samples in smaller files also
+  // get used more often than samples in larger files.
+  // Best for smaller data sets that mostly fit in memory.
+  CS_ROUND_ROBIN,
+};
 
 class WordFeature {
  public:
@@ -103,6 +118,8 @@ class ImageData {
   // Reads from the given file. Returns false in case of error.
   // If swap is true, assumes a big/little-endian swap is needed.
   bool DeSerialize(bool swap, TFile* fp);
+  // As DeSerialize, but only seeks past the data - hence a static method.
+  static bool SkipDeSerialize(bool swap, tesseract::TFile* fp);
 
   // Other accessors.
   const STRING& imagefilename() const {
@@ -145,9 +162,10 @@ class ImageData {
   // Gets anything and everything with a non-NULL pointer, prescaled to a
   // given target_height (if 0, then the original image height), and aligned.
   // Also returns (if not NULL) the width and height of the scaled image.
-  // The return value is the scale factor that was applied to the image to
+  // The return value is the scaled Pix, which must be pixDestroyed after use,
-  // achieve the target_height.
+  // and scale_factor (if not NULL) is set to the scale factor that was applied
-  float PreScale(int target_height, Pix** pix,
+  // to the image to achieve the target_height.
+  Pix* PreScale(int target_height, int max_height, float* scale_factor,
                 int* scaled_width, int* scaled_height,
                 GenericVector<TBOX>* boxes) const;
 
@@ -184,6 +202,8 @@ class ImageData {
 
 // A collection of ImageData that knows roughly how much memory it is using.
 class DocumentData {
+  friend void* ReCachePagesFunc(void* data);
+
  public:
   explicit DocumentData(const STRING& name);
   ~DocumentData();
@@ -192,6 +212,9 @@ class DocumentData {
   // is used to read the file.
   bool LoadDocument(const char* filename, const char* lang, int start_page,
                     inT64 max_memory, FileReader reader);
+  // Sets up the document, without actually loading it.
+  void SetDocument(const char* filename, const char* lang, inT64 max_memory,
+                   FileReader reader);
   // Writes all the pages to the given filename. Returns false on error.
   bool SaveDocument(const char* filename, FileWriter writer);
   bool SaveToBuffer(GenericVector<char>* buffer);
@@ -200,26 +223,64 @@ class DocumentData {
   void AddPageToDocument(ImageData* page);
 
   const STRING& document_name() const {
+    SVAutoLock lock(&general_mutex_);
     return document_name_;
   }
   int NumPages() const {
+    SVAutoLock lock(&general_mutex_);
     return total_pages_;
   }
   inT64 memory_used() const {
+    SVAutoLock lock(&general_mutex_);
     return memory_used_;
   }
+  // If the given index is not currently loaded, loads it using a separate
+  // thread. Note: there are 4 cases:
+  // Document uncached: IsCached() returns false, total_pages_ < 0.
+  // Required page is available: IsPageAvailable returns true. In this case,
+  // total_pages_ > 0 and
+  // pages_offset_ <= index%total_pages_ <= pages_offset_+pages_.size()
+  // Pages are loaded, but the required one is not.
+  // The requested page is being loaded by LoadPageInBackground. In this case,
+  // index == pages_offset_. Once the loading starts, the pages lock is held
+  // until it completes, at which point IsPageAvailable will unblock and return
+  // true.
+  void LoadPageInBackground(int index);
   // Returns a pointer to the page with the given index, modulo the total
-  // number of pages, recaching if needed.
+  // number of pages. Blocks until the background load is completed.
   const ImageData* GetPage(int index);
+  // Returns true if the requested page is available, and provides a pointer,
+  // which may be NULL if the document is empty. May block, even though it
+  // doesn't guarantee to return true.
+  bool IsPageAvailable(int index, ImageData** page);
   // Takes ownership of the given page index. The page is made NULL in *this.
   ImageData* TakePage(int index) {
+    SVAutoLock lock(&pages_mutex_);
     ImageData* page = pages_[index];
     pages_[index] = NULL;
     return page;
   }
+  // Returns true if the document is currently loaded or in the process of
+  // loading.
+  bool IsCached() const { return NumPages() >= 0; }
+  // Removes all pages from memory and frees the memory, but does not forget
+  // the document metadata. Returns the memory saved.
+  inT64 UnCache();
+  // Shuffles all the pages in the document.
+  void Shuffle();
 
  private:
-  // Loads as many pages can fit in max_memory_ starting at index pages_offset_.
+  // Sets the value of total_pages_ behind a mutex.
+  void set_total_pages(int total) {
+    SVAutoLock lock(&general_mutex_);
+    total_pages_ = total;
+  }
+  void set_memory_used(inT64 memory_used) {
+    SVAutoLock lock(&general_mutex_);
+    memory_used_ = memory_used;
+  }
+  // Locks the pages_mutex_ and Loads as many pages can fit in max_memory_
+  // starting at index pages_offset_.
   bool ReCachePages();
 
  private:
@@ -239,43 +300,77 @@ class DocumentData {
   inT64 max_memory_;
   // Saved reader from LoadDocument to allow re-caching.
   FileReader reader_;
+  // Mutex that protects pages_ and pages_offset_ against multiple parallel
+  // loads, and provides a wait for page.
+  SVMutex pages_mutex_;
+  // Mutex that protects other data members that callers want to access without
+  // waiting for a load operation.
+  mutable SVMutex general_mutex_;
 };
 
 // A collection of DocumentData that knows roughly how much memory it is using.
+// Note that while it supports background read-ahead, it assumes that a single
+// thread is accessing documents, ie it is not safe for multiple threads to
+// access different documents in parallel, as one may de-cache the other's
+// content.
 class DocumentCache {
  public:
   explicit DocumentCache(inT64 max_memory);
   ~DocumentCache();
 
+  // Deletes all existing documents from the cache.
+  void Clear() {
+    documents_.clear();
+    num_pages_per_doc_ = 0;
+  }
   // Adds all the documents in the list of filenames, counting memory.
   // The reader is used to read the files.
   bool LoadDocuments(const GenericVector<STRING>& filenames, const char* lang,
-                     FileReader reader);
+                     CachingStrategy cache_strategy, FileReader reader);
 
-  // Adds document to the cache, throwing out other documents if needed.
+  // Adds document to the cache.
   bool AddToCache(DocumentData* data);
 
   // Finds and returns a document by name.
   DocumentData* FindDocument(const STRING& document_name) const;
 
-  // Returns a page by serial number, selecting them in a round-robin fashion
+  // Returns a page by serial number using the current cache_strategy_ to
-  // from all the documents.
+  // determine the mapping from serial number to page.
-  const ImageData* GetPageBySerial(int serial);
+  const ImageData* GetPageBySerial(int serial) {
+    if (cache_strategy_ == CS_SEQUENTIAL)
+      return GetPageSequential(serial);
+    else
+      return GetPageRoundRobin(serial);
+  }
 
   const PointerVector<DocumentData>& documents() const {
     return documents_;
   }
-  int total_pages() const {
+  // Returns the total number of pages in an epoch. For CS_ROUND_ROBIN cache
-    return total_pages_;
+  // strategy, could take a long time.
-  }
+  int TotalPages();
 
  private:
+  // Returns a page by serial number, selecting them in a round-robin fashion
+  // from all the documents. Highly disk-intensive, but doesn't need samples
+  // to be shuffled between files to begin with.
+  const ImageData* GetPageRoundRobin(int serial);
+  // Returns a page by serial number, selecting them in sequence from each file.
+  // Requires the samples to be shuffled between the files to give a random or
+  // uniform distribution of data. Less disk-intensive than GetPageRoundRobin.
+  const ImageData* GetPageSequential(int serial);
+
+  // Helper counts the number of adjacent cached neighbour documents_ of index
+  // looking in direction dir, ie index+dir, index+2*dir etc.
+  int CountNeighbourDocs(int index, int dir);
+
   // A group of pages that corresponds in some loose way to a document.
   PointerVector<DocumentData> documents_;
-  // Total of all pages.
+  // Strategy to use for caching and serializing data samples.
-  int total_pages_;
+  CachingStrategy cache_strategy_;
-  // Total of all memory used by the cache.
+  // Number of pages in the first document, used as a divisor in
-  inT64 memory_used_;
+  // GetPageSequential to determine the document index.
+  int num_pages_per_doc_;
   // Max memory allowed in this cache.
   inT64 max_memory_;
 };

ccstruct/matrix.h

@@ -1,8 +1,12 @@
 /* -*-C-*-
  ******************************************************************************
- *
  * File:         matrix.h  (Formerly matrix.h)
- * Description:  Ratings matrix code. (Used by associator)
+ * Description:  Generic 2-d array/matrix and banded triangular matrix class.
+ * Author:       Ray Smith
+ * TODO(rays) Separate from ratings matrix, which it also contains:
+ *
+ * Descrition:   Ratings matrix class (specialization of banded matrix).
+ *               Segmentation search matrix of lists of BLOB_CHOICE.
  * Author:       Mark Seaman, OCR Technology
  * Created:      Wed May 16 13:22:06 1990
  * Modified:     Tue Mar 19 16:00:20 1991 (Mark Seaman) marks@hpgrlt
@@ -22,12 +26,16 @@
  ** limitations under the License.
  *
  *********************************************************************************/
-#ifndef TESSERACT_CCSTRUCT_MATRIX_H__
+#ifndef TESSERACT_CCSTRUCT_MATRIX_H_
-#define TESSERACT_CCSTRUCT_MATRIX_H__
+#define TESSERACT_CCSTRUCT_MATRIX_H_
 
+#include <math.h>
 #include "kdpair.h"
+#include "points.h"
+#include "serialis.h"
 #include "unicharset.h"
 
+class BLOB_CHOICE;
 class BLOB_CHOICE_LIST;
 
 #define NOT_CLASSIFIED reinterpret_cast<BLOB_CHOICE_LIST*>(0)
@@ -44,34 +52,60 @@ class GENERIC_2D_ARRAY {
   // either pass the memory in, or allocate after by calling Resize().
   GENERIC_2D_ARRAY(int dim1, int dim2, const T& empty, T* array)
     : empty_(empty), dim1_(dim1), dim2_(dim2), array_(array)  {
+    size_allocated_ = dim1 * dim2;
   }
   // Original constructor for a full rectangular matrix DOES allocate memory
   // and initialize it to empty.
   GENERIC_2D_ARRAY(int dim1, int dim2, const T& empty)
     : empty_(empty), dim1_(dim1), dim2_(dim2)  {
-    array_ = new T[dim1_ * dim2_];
+    int new_size = dim1 * dim2;
-    for (int x = 0; x < dim1_; x++)
+    array_ = new T[new_size];
-      for (int y = 0; y < dim2_; y++)
+    size_allocated_ = new_size;
-        this->put(x, y, empty_);
+    for (int i = 0; i < size_allocated_; ++i)
+      array_[i] = empty_;
+  }
+  // Default constructor for array allocation. Use Resize to set the size.
+  GENERIC_2D_ARRAY()
+    : array_(NULL), empty_(static_cast<T>(0)), dim1_(0), dim2_(0),
+      size_allocated_(0) {
+  }
+  GENERIC_2D_ARRAY(const GENERIC_2D_ARRAY<T>& src)
+    : array_(NULL), empty_(static_cast<T>(0)), dim1_(0), dim2_(0),
+      size_allocated_(0) {
+    *this = src;
   }
   virtual ~GENERIC_2D_ARRAY() { delete[] array_; }
 
+  void operator=(const GENERIC_2D_ARRAY<T>& src) {
+    ResizeNoInit(src.dim1(), src.dim2());
+    memcpy(array_, src.array_, num_elements() * sizeof(array_[0]));
+  }
+
+  // Reallocate the array to the given size. Does not keep old data, but does
+  // not initialize the array either.
+  void ResizeNoInit(int size1, int size2) {
+    int new_size = size1 * size2;
+    if (new_size > size_allocated_) {
+      delete [] array_;
+      array_ = new T[new_size];
+      size_allocated_ = new_size;
+    }
+    dim1_ = size1;
+    dim2_ = size2;
+  }
+
   // Reallocate the array to the given size. Does not keep old data.
   void Resize(int size1, int size2, const T& empty) {
     empty_ = empty;
-    if (size1 != dim1_ || size2 != dim2_) {
+    ResizeNoInit(size1, size2);
-      dim1_ = size1;
-      dim2_ = size2;
-      delete [] array_;
-      array_ = new T[dim1_ * dim2_];
-    }
     Clear();
   }
 
   // Reallocate the array to the given size, keeping old data.
   void ResizeWithCopy(int size1, int size2) {
     if (size1 != dim1_ || size2 != dim2_) {
-      T* new_array = new T[size1 * size2];
+      int new_size = size1 * size2;
+      T* new_array = new T[new_size];
       for (int col = 0; col < size1; ++col) {
         for (int row = 0; row < size2; ++row) {
           int old_index = col * dim2() + row;
@@ -87,6 +121,7 @@ class GENERIC_2D_ARRAY {
       array_ = new_array;
       dim1_ = size1;
       dim2_ = size2;
+      size_allocated_ = new_size;
     }
   }
 
@@ -106,9 +141,16 @@ class GENERIC_2D_ARRAY {
     if (fwrite(array_, sizeof(*array_), size, fp) != size) return false;
     return true;
   }
+  bool Serialize(tesseract::TFile* fp) const {
+    if (!SerializeSize(fp)) return false;
+    if (fp->FWrite(&empty_, sizeof(empty_), 1) != 1) return false;
+    int size = num_elements();
+    if (fp->FWrite(array_, sizeof(*array_), size) != size) return false;
+    return true;
+  }
 
   // Reads from the given file. Returns false in case of error.
-  // Only works with bitwise-serializeable typ
+  // Only works with bitwise-serializeable types!
   // If swap is true, assumes a big/little-endian swap is needed.
   bool DeSerialize(bool swap, FILE* fp) {
     if (!DeSerializeSize(swap, fp)) return false;
@@ -122,6 +164,18 @@ class GENERIC_2D_ARRAY {
     }
     return true;
   }
+  bool DeSerialize(bool swap, tesseract::TFile* fp) {
+    if (!DeSerializeSize(swap, fp)) return false;
+    if (fp->FRead(&empty_, sizeof(empty_), 1) != 1) return false;
+    if (swap) ReverseN(&empty_, sizeof(empty_));
+    int size = num_elements();
+    if (fp->FRead(array_, sizeof(*array_), size) != size) return false;
+    if (swap) {
+      for (int i = 0; i < size; ++i)
+        ReverseN(&array_[i], sizeof(array_[i]));
+    }
+    return true;
+  }
 
   // Writes to the given file. Returns false in case of error.
   // Assumes a T::Serialize(FILE*) const function.
@@ -163,11 +217,17 @@ class GENERIC_2D_ARRAY {
   }
 
   // Put a list element into the matrix at a specific location.
+  void put(ICOORD pos, const T& thing) {
+    array_[this->index(pos.x(), pos.y())] = thing;
+  }
   void put(int column, int row, const T& thing) {
     array_[this->index(column, row)] = thing;
   }
 
   // Get the item at a specified location from the matrix.
+  T get(ICOORD pos) const {
+    return array_[this->index(pos.x(), pos.y())];
+  }
   T get(int column, int row) const {
     return array_[this->index(column, row)];
   }
@@ -187,6 +247,207 @@ class GENERIC_2D_ARRAY {
     return &array_[this->index(column, 0)];
   }
 
+  // Adds addend to *this, element-by-element.
+  void operator+=(const GENERIC_2D_ARRAY<T>& addend) {
+    if (dim2_ == addend.dim2_) {
+      // Faster if equal size in the major dimension.
+      int size = MIN(num_elements(), addend.num_elements());
+      for (int i = 0; i < size; ++i) {
+        array_[i] += addend.array_[i];
+      }
+    } else {
+      for (int x = 0; x < dim1_; x++) {
+        for (int y = 0; y < dim2_; y++) {
+          (*this)(x, y) += addend(x, y);
+        }
+      }
+    }
+  }
+  // Subtracts minuend from *this, element-by-element.
+  void operator-=(const GENERIC_2D_ARRAY<T>& minuend) {
+    if (dim2_ == minuend.dim2_) {
+      // Faster if equal size in the major dimension.
+      int size = MIN(num_elements(), minuend.num_elements());
+      for (int i = 0; i < size; ++i) {
+        array_[i] -= minuend.array_[i];
+      }
+    } else {
+      for (int x = 0; x < dim1_; x++) {
+        for (int y = 0; y < dim2_; y++) {
+          (*this)(x, y) -= minuend(x, y);
+        }
+      }
+    }
+  }
+  // Adds addend to all elements.
+  void operator+=(const T& addend) {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      array_[i] += addend;
+    }
+  }
+  // Multiplies *this by factor, element-by-element.
+  void operator*=(const T& factor) {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      array_[i] *= factor;
+    }
+  }
+  // Clips *this to the given range.
+  void Clip(const T& rangemin, const T& rangemax) {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      array_[i] = ClipToRange(array_[i], rangemin, rangemax);
+    }
+  }
+  // Returns true if all elements of *this are within the given range.
+  // Only uses operator<
+  bool WithinBounds(const T& rangemin, const T& rangemax) const {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      const T& value = array_[i];
+      if (value < rangemin || rangemax < value)
+        return false;
+    }
+    return true;
+  }
+  // Normalize the whole array.
+  double Normalize() {
+    int size = num_elements();
+    if (size <= 0) return 0.0;
+    // Compute the mean.
+    double mean = 0.0;
+    for (int i = 0; i < size; ++i) {
+      mean += array_[i];
+    }
+    mean /= size;
+    // Subtract the mean and compute the standard deviation.
+    double sd = 0.0;
+    for (int i = 0; i < size; ++i) {
+      double normed = array_[i] - mean;
+      array_[i] = normed;
+      sd += normed * normed;
+    }
+    sd = sqrt(sd / size);
+    if (sd > 0.0) {
+      // Divide by the sd.
+      for (int i = 0; i < size; ++i) {
+        array_[i] /= sd;
+      }
+    }
+    return sd;
+  }
+
+  // Returns the maximum value of the array.
+  T Max() const {
+    int size = num_elements();
+    if (size <= 0) return empty_;
+    // Compute the max.
+    T max_value = array_[0];
+    for (int i = 1; i < size; ++i) {
+      const T& value = array_[i];
+      if (value > max_value) max_value = value;
+    }
+    return max_value;
+  }
+
+  // Returns the maximum absolute value of the array.
+  T MaxAbs() const {
+    int size = num_elements();
+    if (size <= 0) return empty_;
+    // Compute the max.
+    T max_abs = static_cast<T>(0);
+    for (int i = 0; i < size; ++i) {
+      T value = static_cast<T>(fabs(array_[i]));
+      if (value > max_abs) max_abs = value;
+    }
+    return max_abs;
+  }
+
+  // Accumulates the element-wise sums of squares of src into *this.
+  void SumSquares(const GENERIC_2D_ARRAY<T>& src) {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      array_[i] += src.array_[i] * src.array_[i];
+    }
+  }
+
+  // Scales each element using the ada-grad algorithm, ie array_[i] by
+  // sqrt(num_samples/max(1,sqsum[i])).
+  void AdaGradScaling(const GENERIC_2D_ARRAY<T>& sqsum, int num_samples) {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      array_[i] *= sqrt(num_samples / MAX(1.0, sqsum.array_[i]));
+    }
+  }
+
+  void AssertFinite() const {
+    int size = num_elements();
+    for (int i = 0; i < size; ++i) {
+      ASSERT_HOST(isfinite(array_[i]));
+    }
+  }
+
+  // REGARDLESS OF THE CURRENT DIMENSIONS, treats the data as a
+  // num_dims-dimensional array/tensor with dimensions given by dims, (ordered
+  // from most significant to least significant, the same as standard C arrays)
+  // and moves src_dim to dest_dim, with the initial dest_dim and any dimensions
+  // in between shifted towards the hole left by src_dim. Example:
+  // Current data content: array_=[0, 1, 2, ....119]
+  //   perhaps *this may be of dim[40, 3], with values [[0, 1, 2][3, 4, 5]...
+  //   but the current dimensions are irrelevant.
+  // num_dims = 4, dims=[5, 4, 3, 2]
+  // src_dim=3, dest_dim=1
+  // tensor=[[[[0, 1][2, 3][4, 5]]
+  //          [[6, 7][8, 9][10, 11]]
+  //          [[12, 13][14, 15][16, 17]]
+  //          [[18, 19][20, 21][22, 23]]]
+  //         [[[24, 25]...
+  // output dims =[5, 2, 4, 3]
+  // output tensor=[[[[0, 2, 4][6, 8, 10][12, 14, 16][18, 20, 22]]
+  //                 [[1, 3, 5][7, 9, 11][13, 15, 17][19, 21, 23]]]
+  //                [[[24, 26, 28]...
+  // which is stored in the array_ as:
+  //   [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 1, 3, 5, 7, 9, 11, 13...]
+  // NOTE: the 2 stored matrix dimensions are simply copied from *this. To
+  // change the dimensions after the transpose, use ResizeNoInit.
+  // Higher dimensions above 2 are strictly the responsibility of the caller.
+  void RotatingTranspose(const int* dims, int num_dims, int src_dim,
+                         int dest_dim, GENERIC_2D_ARRAY<T>* result) const {
+    int max_d = MAX(src_dim, dest_dim);
+    int min_d = MIN(src_dim, dest_dim);
+    // In a tensor of shape [d0, d1... min_d, ... max_d, ... dn-2, dn-1], the
+    // ends outside of min_d and max_d are unaffected, with [max_d +1, dn-1]
+    // being contiguous blocks of data that will move together, and
+    // [d0, min_d -1] being replicas of the transpose operation.
+    // num_replicas represents the large dimensions unchanged by the operation.
+    // move_size represents the small dimensions unchanged by the operation.
+    // src_step represents the stride in the src between each adjacent group
+    // in the destination.
+    int num_replicas = 1, move_size = 1, src_step = 1;
+    for (int d = 0; d < min_d; ++d) num_replicas *= dims[d];
+    for (int d = max_d + 1; d < num_dims; ++d) move_size *= dims[d];
+    for (int d = src_dim + 1; d < num_dims; ++d) src_step *= dims[d];
+    if (src_dim > dest_dim) src_step *= dims[src_dim];
+    // wrap_size is the size of a single replica, being the amount that is
+    // handled num_replicas times.
+    int wrap_size = move_size;
+    for (int d = min_d; d <= max_d; ++d) wrap_size *= dims[d];
+    result->ResizeNoInit(dim1_, dim2_);
+    result->empty_ = empty_;
+    const T* src = array_;
+    T* dest = result->array_;
+    for (int replica = 0; replica < num_replicas; ++replica) {
+      for (int start = 0; start < src_step; start += move_size) {
+        for (int pos = start; pos < wrap_size; pos += src_step) {
+          memcpy(dest, src + pos, sizeof(*dest) * move_size);
+          dest += move_size;
+        }
+      }
+      src += wrap_size;
+    }
+  }
+
   // Delete objects pointed to by array_[i].
   void delete_matrix_pointers() {
     int size = num_elements();
@@ -206,6 +467,13 @@ class GENERIC_2D_ARRAY {
     if (fwrite(&size, sizeof(size), 1, fp) != 1) return false;
     return true;
   }
+  bool SerializeSize(tesseract::TFile* fp) const {
+    inT32 size = dim1_;
+    if (fp->FWrite(&size, sizeof(size), 1) != 1) return false;
+    size = dim2_;
+    if (fp->FWrite(&size, sizeof(size), 1) != 1) return false;
+    return true;
+  }
   // Factored helper to deserialize the size.
   // If swap is true, assumes a big/little-endian swap is needed.
   bool DeSerializeSize(bool swap, FILE* fp) {
@@ -219,11 +487,26 @@ class GENERIC_2D_ARRAY {
     Resize(size1, size2, empty_);
     return true;
   }
+  bool DeSerializeSize(bool swap, tesseract::TFile* fp) {
+    inT32 size1, size2;
+    if (fp->FRead(&size1, sizeof(size1), 1) != 1) return false;
+    if (fp->FRead(&size2, sizeof(size2), 1) != 1) return false;
+    if (swap) {
+      ReverseN(&size1, sizeof(size1));
+      ReverseN(&size2, sizeof(size2));
+    }
+    Resize(size1, size2, empty_);
+    return true;
+  }
 
   T* array_;
   T empty_;   // The unused cell.
   int dim1_;  // Size of the 1st dimension in indexing functions.
   int dim2_;  // Size of the 2nd dimension in indexing functions.
+  // The total size to which the array can be expanded before a realloc is
+  // needed. If Resize is used, memory is retained so it can be re-expanded
+  // without a further alloc, and this stores the allocated size.
+  int size_allocated_;
 };
 
 // A generic class to store a banded triangular matrix with entries of type T.
@@ -349,4 +632,4 @@ struct MATRIX_COORD {
 // The MatrixCoordPair contains a MATRIX_COORD and its priority.
 typedef tesseract::KDPairInc<float, MATRIX_COORD> MatrixCoordPair;
 
-#endif  // TESSERACT_CCSTRUCT_MATRIX_H__
+#endif  // TESSERACT_CCSTRUCT_MATRIX_H_

ccstruct/mod128.cpp

@@ -86,16 +86,3 @@ DIR128::DIR128(                 //from fcoord
   while (high - low > 1);
   dir = low;
 }
-
-
-/**********************************************************************
- * dir_to_gradient
- *
- * Convert a direction to a vector.
- **********************************************************************/
-
-#if 0 // code is buggy for negative dir and unused
-ICOORD DIR128::vector() const {  //convert to vector
-  return dirtab[dir];            //easy really
-}
-#endif

ccstruct/mod128.h

@@ -77,7 +77,6 @@ class DLLSYM DIR128
     inT8 get_dir() const {  //access function
       return dir;
     }
-    ICOORD vector() const;  //turn to vector
 
   private:
     inT8 dir;                    //a direction

ccstruct/otsuthr.cpp

@@ -52,22 +52,15 @@ int OtsuThreshold(Pix* src_pix, int left, int top, int width, int height,
   // only use opencl if compiled w/ OpenCL and selected device is opencl
 #ifdef USE_OPENCL
   // all of channel 0 then all of channel 1...
-    int *histogramAllChannels = new int[kHistogramSize * num_channels];
+  int* histogramAllChannels = new int[kHistogramSize * num_channels];
 
   // Calculate Histogram on GPU
   OpenclDevice od;
-    if (od.selectedDeviceIsOpenCL() &&
+  if (od.selectedDeviceIsOpenCL() && (num_channels == 1 || num_channels == 4) &&
-        (num_channels == 1 || num_channels == 4) && top == 0 && left == 0 ) {
+      top == 0 && left == 0) {
-      od.HistogramRectOCL(
+    od.HistogramRectOCL((unsigned char*)pixGetData(src_pix), num_channels,
-          (const unsigned char*)pixGetData(src_pix),
+                        pixGetWpl(src_pix) * 4, left, top, width, height,
-          num_channels,
+                        kHistogramSize, histogramAllChannels);
-          pixGetWpl(src_pix) * 4,
-          left,
-          top,
-          width,
-          height,
-          kHistogramSize,
-          histogramAllChannels);
 
     // Calculate Threshold from Histogram on cpu
     for (int ch = 0; ch < num_channels; ++ch) {
@@ -143,7 +136,6 @@ int OtsuThreshold(Pix* src_pix, int left, int top, int width, int height,
   delete[] histogramAllChannels;
 #endif  // USE_OPENCL
 
-
   if (!any_good_hivalue) {
     // Use the best of the ones that were not good enough.
     (*hi_values)[best_hi_index] = best_hi_value;

ccstruct/otsuthr.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCMAIN_OTSUTHR_H__
+#ifndef TESSERACT_CCMAIN_OTSUTHR_H_
-#define TESSERACT_CCMAIN_OTSUTHR_H__
+#define TESSERACT_CCMAIN_OTSUTHR_H_
 
 struct Pix;
 
@@ -53,4 +53,4 @@ int OtsuStats(const int* histogram, int* H_out, int* omega0_out);
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCMAIN_OTSUTHR_H__
+#endif  // TESSERACT_CCMAIN_OTSUTHR_H_

ccstruct/pageres.cpp

@@ -303,8 +303,9 @@ bool WERD_RES::SetupForRecognition(const UNICHARSET& unicharset_in,
       static_cast<tesseract::OcrEngineMode>(norm_mode);
   tesseract = tess;
   POLY_BLOCK* pb = block != NULL ? block->poly_block() : NULL;
-  if ((norm_mode_hint != tesseract::OEM_CUBE_ONLY &&
+  if ((norm_mode_hint != tesseract::OEM_LSTM_ONLY &&
-       word->cblob_list()->empty()) || (pb != NULL && !pb->IsText())) {
+       word->cblob_list()->empty()) ||
+      (pb != NULL && !pb->IsText())) {
     // Empty words occur when all the blobs have been moved to the rej_blobs
     // list, which seems to occur frequently in junk.
     SetupFake(unicharset_in);
@@ -528,13 +529,12 @@ void WERD_RES::FilterWordChoices(int debug_level) {
       if (choice->unichar_id(i) != best_choice->unichar_id(j) &&
           choice->certainty(i) - best_choice->certainty(j) < threshold) {
         if (debug_level >= 2) {
-          STRING label;
+          choice->print("WorstCertaintyDiffWorseThan");
-          label.add_str_int("\nDiscarding bad choice #", index);
+          tprintf(
-          choice->print(label.string());
+              "i %d j %d Choice->Blob[i].Certainty %.4g"
-          tprintf("i %d j %d Chunk %d Choice->Blob[i].Certainty %.4g"
+              " WorstOtherChoiceCertainty %g Threshold %g\n",
-              " BestChoice->ChunkCertainty[Chunk] %g Threshold %g\n",
+              i, j, choice->certainty(i), best_choice->certainty(j), threshold);
-              i, j, chunk, choice->certainty(i),
+          tprintf("Discarding bad choice #%d\n", index);
-              best_choice->certainty(j), threshold);
         }
         delete it.extract();
         break;
@@ -882,17 +882,18 @@ void WERD_RES::FakeClassifyWord(int blob_count, BLOB_CHOICE** choices) {
     choice_it.add_after_then_move(choices[c]);
     ratings->put(c, c, choice_list);
   }
-  FakeWordFromRatings();
+  FakeWordFromRatings(TOP_CHOICE_PERM);
   reject_map.initialise(blob_count);
+  best_state.init_to_size(blob_count, 1);
   done = true;
 }
 
 // Creates a WERD_CHOICE for the word using the top choices from the leading
 // diagonal of the ratings matrix.
-void WERD_RES::FakeWordFromRatings() {
+void WERD_RES::FakeWordFromRatings(PermuterType permuter) {
   int num_blobs = ratings->dimension();
   WERD_CHOICE* word_choice = new WERD_CHOICE(uch_set, num_blobs);
-  word_choice->set_permuter(TOP_CHOICE_PERM);
+  word_choice->set_permuter(permuter);
   for (int b = 0; b < num_blobs; ++b) {
     UNICHAR_ID unichar_id = UNICHAR_SPACE;
     float rating = MAX_INT32;
@@ -1105,6 +1106,7 @@ void WERD_RES::InitNonPointers() {
   x_height = 0.0;
   caps_height = 0.0;
   baseline_shift = 0.0f;
+  space_certainty = 0.0f;
   guessed_x_ht = TRUE;
   guessed_caps_ht = TRUE;
   combination = FALSE;

ccstruct/pageres.h

@@ -295,6 +295,9 @@ class WERD_RES : public ELIST_LINK {
   float x_height;              // post match estimate
   float caps_height;           // post match estimate
   float baseline_shift;        // post match estimate.
+  // Certainty score for the spaces either side of this word (LSTM mode).
+  // MIN this value with the actual word certainty.
+  float space_certainty;
 
   /*
     To deal with fuzzy spaces we need to be able to combine "words" to form
@@ -327,7 +330,7 @@ class WERD_RES : public ELIST_LINK {
   }
   // Deep copies everything except the ratings MATRIX.
   // To get that use deep_copy below.
-  WERD_RES(const WERD_RES &source) : ELIST_LINK(source) {
+  WERD_RES(const WERD_RES& source) : ELIST_LINK(source) {
     InitPointers();
     *this = source;            // see operator=
   }
@@ -590,7 +593,7 @@ class WERD_RES : public ELIST_LINK {
 
   // Creates a WERD_CHOICE for the word using the top choices from the leading
   // diagonal of the ratings matrix.
-  void FakeWordFromRatings();
+  void FakeWordFromRatings(PermuterType permuter);
 
   // Copies the best_choice strings to the correct_text for adaption/training.
   void BestChoiceToCorrectText();
@@ -630,7 +633,7 @@ class WERD_RES : public ELIST_LINK {
   static WERD_RES* deep_copy(const WERD_RES* src) {
     WERD_RES* result = new WERD_RES(*src);
     // That didn't copy the ratings, but we want a copy if there is one to
-    // begin width.
+    // begin with.
     if (src->ratings != NULL)
       result->ratings = src->ratings->DeepCopy();
     return result;

ccstruct/params_training_featdef.h

@@ -126,7 +126,7 @@ typedef GenericVector<ParamsTrainingHypothesis> ParamsTrainingHypothesisList;
 // explored on PASS1, PASS2, fix xheight pass, etc).
 class ParamsTrainingBundle {
  public:
-  ParamsTrainingBundle() {};
+  ParamsTrainingBundle() {}
   // Starts a new hypothesis list.
   // Should be called at the beginning of a new run of the segmentation search.
   void StartHypothesisList() {

ccstruct/pdblock.h

@@ -29,59 +29,45 @@ struct Pix;
 
 CLISTIZEH (PDBLK)
 ///page block
-class PDBLK
+class PDBLK {
-{
   friend class BLOCK_RECT_IT;    //< block iterator
 
  public:
-    ///empty constructor
+  /// empty constructor
   PDBLK() {
     hand_poly = NULL;
     index_ = 0;
   }
-    ///simple constructor
+  /// simple constructor
   PDBLK(inT16 xmin,  //< bottom left
         inT16 ymin,
         inT16 xmax,  //< top right
         inT16 ymax);
 
-    ///set vertex lists
+  /// set vertex lists
   ///@param left list of left vertices
   ///@param right list of right vertices
-    void set_sides(ICOORDELT_LIST *left,
+  void set_sides(ICOORDELT_LIST *left, ICOORDELT_LIST *right);
-                   ICOORDELT_LIST *right);
 
-    ///destructor
+  /// destructor
-    ~PDBLK () {
+  ~PDBLK() { delete hand_poly; }
-      if (hand_poly) delete hand_poly;
-    }
 
-    POLY_BLOCK *poly_block() const {
+  POLY_BLOCK *poly_block() const { return hand_poly; }
-      return hand_poly;
+  /// set the poly block
-    }
+  void set_poly_block(POLY_BLOCK *blk) { hand_poly = blk; }
-    ///set the poly block
+  /// get box
-    void set_poly_block(POLY_BLOCK *blk) {
+  void bounding_box(ICOORD &bottom_left,        // bottom left
-      hand_poly = blk;
+                    ICOORD &top_right) const {  // topright
-    }
+    bottom_left = box.botleft();
-    ///get box
+    top_right = box.topright();
-    void bounding_box(ICOORD &bottom_left,        //bottom left
-                      ICOORD &top_right) const {  //topright
-      bottom_left = box.botleft ();
-      top_right = box.topright ();
-    }
-    ///get real box
-    const TBOX &bounding_box() const {
-      return box;
   }
+  /// get real box
+  const TBOX &bounding_box() const { return box; }
 
-    int index() const {
+  int index() const { return index_; }
-      return index_;
+  void set_index(int value) { index_ = value; }
-    }
-    void set_index(int value) {
-      index_ = value;
-    }
 
-    ///is pt inside block
+  /// is pt inside block
   BOOL8 contains(ICOORD pt);
 
   /// reposition block
@@ -93,19 +79,17 @@ class PDBLK
   // mask image.
   Pix *render_mask(const FCOORD &rerotation, TBOX *mask_box);
 
-    #ifndef GRAPHICS_DISABLED
+#ifndef GRAPHICS_DISABLED
-    ///draw histogram
+  /// draw histogram
   ///@param window window to draw in
   ///@param serial serial number
   ///@param colour colour to draw in
-    void plot(ScrollView* window,
+  void plot(ScrollView *window, inT32 serial, ScrollView::Color colour);
-              inT32 serial,
+#endif  // GRAPHICS_DISABLED
-              ScrollView::Color colour);
-    #endif  // GRAPHICS_DISABLED
 
-    ///assignment
+  /// assignment
   ///@param source from this
-    PDBLK & operator= (const PDBLK & source);
+  PDBLK &operator=(const PDBLK &source);
 
  protected:
   POLY_BLOCK *hand_poly;     //< weird as well

ccstruct/polyblk.cpp

@@ -294,6 +294,8 @@ void POLY_BLOCK::fill(ScrollView* window, ScrollView::Color colour) {
       }
     }
   }
+
+  delete lines;
 }
 #endif
 

ccstruct/publictypes.h

@@ -17,8 +17,8 @@
 //
 ///////////////////////////////////////////////////////////////////////
 
-#ifndef TESSERACT_CCSTRUCT_PUBLICTYPES_H__
+#ifndef TESSERACT_CCSTRUCT_PUBLICTYPES_H_
-#define TESSERACT_CCSTRUCT_PUBLICTYPES_H__
+#define TESSERACT_CCSTRUCT_PUBLICTYPES_H_
 
 // This file contains types that are used both by the API and internally
 // to Tesseract. In order to decouple the API from Tesseract and prevent cyclic
@@ -213,7 +213,7 @@ enum PageIteratorLevel {
 };
 
 /**
- * JUSTIFICATION_UNKNONW
+ * JUSTIFICATION_UNKNOWN
  *   The alignment is not clearly one of the other options.  This could happen
  *   for example if there are only one or two lines of text or the text looks
  *   like source code or poetry.
@@ -235,7 +235,7 @@ enum PageIteratorLevel {
  *
  * JUSTIFICATION_RIGHT
  *   Each line, except possibly the first, is flush to the same right tab stop.
-*/
+ */
 enum ParagraphJustification {
   JUSTIFICATION_UNKNOWN,
   JUSTIFICATION_LEFT,
@@ -255,17 +255,20 @@ enum ParagraphJustification {
 */
 enum OcrEngineMode {
   OEM_TESSERACT_ONLY,           // Run Tesseract only - fastest
-  OEM_CUBE_ONLY,                // Run Cube only - better accuracy, but slower
+  OEM_LSTM_ONLY,                // Run just the LSTM line recognizer.
-  OEM_TESSERACT_CUBE_COMBINED,  // Run both and combine results - best accuracy
+  OEM_TESSERACT_LSTM_COMBINED,  // Run the LSTM recognizer, but allow fallback
-  OEM_DEFAULT                   // Specify this mode when calling init_*(),
+                                // to Tesseract when things get difficult.
+  OEM_DEFAULT,                  // Specify this mode when calling init_*(),
                                 // to indicate that any of the above modes
                                 // should be automatically inferred from the
                                 // variables in the language-specific config,
                                 // command-line configs, or if not specified
                                 // in any of the above should be set to the
                                 // default OEM_TESSERACT_ONLY.
+  OEM_CUBE_ONLY,                // Run Cube only - better accuracy, but slower
+  OEM_TESSERACT_CUBE_COMBINED,  // Run both and combine results - best accuracy
 };
 
 }  // namespace tesseract.
 
-#endif  // TESSERACT_CCSTRUCT_PUBLICTYPES_H__
+#endif  // TESSERACT_CCSTRUCT_PUBLICTYPES_H_

ccstruct/ratngs.h

@@ -288,7 +288,8 @@ class WERD_CHOICE : public ELIST_LINK {
                src_certainty, src_permuter);
   }
   WERD_CHOICE(const char *src_string, const UNICHARSET &unicharset);
-  WERD_CHOICE(const WERD_CHOICE &word) : ELIST_LINK(word), unicharset_(word.unicharset_) {
+  WERD_CHOICE(const WERD_CHOICE &word)
+      : ELIST_LINK(word), unicharset_(word.unicharset_) {
     this->init(word.length());
     this->operator=(word);
   }
@@ -507,6 +508,20 @@ class WERD_CHOICE : public ELIST_LINK {
     }
     return word_str;
   }
+  // Returns true if any unichar_id in the word is a non-space-delimited char.
+  bool ContainsAnyNonSpaceDelimited() const {
+    for (int i = 0; i < length_; ++i) {
+      if (!unicharset_->IsSpaceDelimited(unichar_ids_[i])) return true;
+    }
+    return false;
+  }
+  // Returns true if the word is all spaces.
+  bool IsAllSpaces() const {
+    for (int i = 0; i < length_; ++i) {
+      if (unichar_ids_[i] != UNICHAR_SPACE) return false;
+    }
+    return true;
+  }
 
   // Call this to override the default (strict left to right graphemes)
   // with the fact that some engine produces a "reading order" set of

ccstruct/rect.cpp

@@ -29,9 +29,9 @@
  *
  **********************************************************************/
 
-TBOX::TBOX(                   //constructor
+TBOX::TBOX(            // constructor
-         const ICOORD pt1,  //one corner
+    const ICOORD pt1,  // one corner
-         const ICOORD pt2   //the other corner
+    const ICOORD pt2   // the other corner
     ) {
   if (pt1.x () <= pt2.x ()) {
     if (pt1.y () <= pt2.y ()) {

ccstruct/rejctmap.cpp

@@ -267,10 +267,10 @@ void REJ::full_print(FILE *fp) {
 
 //The REJMAP class has been hacked to use alloc_struct instead of new [].
 //This is to reduce memory fragmentation only as it is rather kludgy.
-//alloc_struct by-passes the call to the contsructor of REJ on each
+// alloc_struct by-passes the call to the constructor of REJ on each
-//array element. Although the constructor is empty, the BITS16 members
+// array element. Although the constructor is empty, the BITS16 members
-//do have a constructor which sets all the flags to 0. The memset
+// do have a constructor which sets all the flags to 0. The memset
-//replaces this functionality.
+// replaces this functionality.
 
 REJMAP::REJMAP(  //classwise copy
                const REJMAP &source) {

ccstruct/rejctmap.h

@@ -48,8 +48,7 @@ OF THIS IMPLIED TEMPORAL ORDERING OF THE FLAGS!!!!
 #include          "bits16.h"
 #include                   "params.h"
 
-enum REJ_FLAGS
+enum REJ_FLAGS {
-{
   /* Reject modes which are NEVER overridden */
   R_TESS_FAILURE,    // PERM Tess didn't classify
   R_SMALL_XHT,       // PERM Xht too small
@@ -83,11 +82,11 @@ enum REJ_FLAGS
   R_UNLV_REJ,   // TEMP ~ turned to - or ^ turned to space
 
   /* Accept modes which occur between the above rejection groups */
-  R_NN_ACCEPT,                   //NN acceptance
+  R_NN_ACCEPT,          // NN acceptance
-  R_HYPHEN_ACCEPT,               //Hyphen acceptance
+  R_HYPHEN_ACCEPT,      // Hyphen acceptance
-  R_MM_ACCEPT,                   //Matrix match acceptance
+  R_MM_ACCEPT,          // Matrix match acceptance
-  R_QUALITY_ACCEPT,              //Accept word in good quality doc
+  R_QUALITY_ACCEPT,     // Accept word in good quality doc
-  R_MINIMAL_REJ_ACCEPT           //Accept EVERYTHING except tess failures
+  R_MINIMAL_REJ_ACCEPT  // Accept EVERYTHING except tess failures
 };
 
 /* REJECT MAP VALUES */

ccstruct/statistc.cpp

@@ -215,7 +215,6 @@ inT32 STATS::min_bucket() const {  // Find min
   return rangemin_ + min;
 }
 
-
 /**********************************************************************
  * STATS::max_bucket
  *

ccutil/Makefile.am

@@ -14,11 +14,11 @@ endif
 include_HEADERS = \
 	basedir.h errcode.h fileerr.h genericvector.h helpers.h host.h memry.h \
 	ndminx.h params.h ocrclass.h platform.h serialis.h strngs.h \
-	tesscallback.h unichar.h unicharmap.h unicharset.h
+	tesscallback.h unichar.h unicharcompress.h unicharmap.h unicharset.h
 
 noinst_HEADERS = \
     ambigs.h bits16.h bitvector.h ccutil.h clst.h doubleptr.h elst2.h \
-    elst.h genericheap.h globaloc.h hashfn.h indexmapbidi.h kdpair.h lsterr.h \
+    elst.h genericheap.h globaloc.h indexmapbidi.h kdpair.h lsterr.h \
     nwmain.h object_cache.h qrsequence.h sorthelper.h stderr.h \
     scanutils.h tessdatamanager.h tprintf.h unicity_table.h unicodes.h \
     universalambigs.h
@@ -38,7 +38,7 @@ libtesseract_ccutil_la_SOURCES = \
     mainblk.cpp memry.cpp \
     serialis.cpp strngs.cpp scanutils.cpp \
     tessdatamanager.cpp tprintf.cpp \
-    unichar.cpp unicharmap.cpp unicharset.cpp unicodes.cpp \
+    unichar.cpp unicharcompress.cpp unicharmap.cpp unicharset.cpp unicodes.cpp \
     params.cpp universalambigs.cpp
 
 if T_WIN

ccutil/ambigs.h

@@ -59,17 +59,18 @@ class UnicharIdArrayUtils {
   // less than length of array2, if any array1[i] is less than array2[i].
   // Returns 0 if the arrays are equal, 1 otherwise.
   // The function assumes that the arrays are terminated by INVALID_UNICHAR_ID.
-  static inline int compare(const UNICHAR_ID array1[],
+  static inline int compare(const UNICHAR_ID *ptr1, const UNICHAR_ID *ptr2) {
-                            const UNICHAR_ID array2[]) {
+    for (;;) {
-    const UNICHAR_ID *ptr1 = array1;
+      const UNICHAR_ID val1 = *ptr1++;
-    const UNICHAR_ID *ptr2 = array2;
+      const UNICHAR_ID val2 = *ptr2++;
-    while (*ptr1 != INVALID_UNICHAR_ID && *ptr2 != INVALID_UNICHAR_ID) {
+      if (val1 != val2) {
-      if (*ptr1 != *ptr2) return *ptr1 < *ptr2 ? -1 : 1;
+        if (val1 == INVALID_UNICHAR_ID) return -1;
-      ++ptr1;
+        if (val2 == INVALID_UNICHAR_ID) return 1;
-      ++ptr2;
+        if (val1 < val2) return -1;
+        return 1;
+      }
+      if (val1 == INVALID_UNICHAR_ID) return 0;
     }
-    if (*ptr1 == INVALID_UNICHAR_ID && *ptr2 == INVALID_UNICHAR_ID) return 0;
-    return *ptr1 == INVALID_UNICHAR_ID ? -1 : 1;
   }
 
   // Look uid in the vector of uids.  If found, the index of the matched