tesseract/ReleaseNotes

Tesseract release notes Feb 2, 2007 - V1.03.
Added mftraining and cntraining. Using an image with a box file, tesseract
generates .tr output files. cntraining runs on the .tr files to make
normproto that lives in tessdata. mftraining runs on the .tr files to
make inttemp and pffmtable in tessdata. These are the main data files
that tesseract uses to recognize characters. At present, the code to make
dictionary files is not yet available, nor are any sample box files or
rebuilt inttemp or documentation to create any of these. Recognition is
still limited to the ASCII set, but when this problem is fixed, documentation
will follow.

Added a new API with adaptive thresholding for grey and color images.
See ccmain/baseapi.h/cpp for details. The main program has been converted
to use the API as an example. See main() in ccmain/tesseractmain.cpp for
details. The API is designed to make it easy to add subclasses with ability
to output the bounding boxes etc from the internal structures. The adaptive
thresholding improves accuracy (most of the time) on non-binary images.

Many memory leaks have been fixed. There are no known leaks left from using
the API correctly.

The adaptive classifier was not operating correctly. This bug, and several
others have been fixed, including poor chopping, an indefinite (if not quite
infinite) loop in the number parser, and a couple of crash bugs. Thanks to
all that have contributed bugs and bug fixes.

It is now possible to build without any of the graphics support to save code
size using #define GRAPHICS_DISABLED. There is also a new EMBEDDED define
for use on operating systems with limited library support.

64-bit and Mac OSX buildability is now included in the mainline source tree.
Thanks to all that have contributed patches and comments to help with that.
1.03 is also endian-independent, apart from the tiff i/o, so if you use
libtiff, the code should run on all platforms, even if you get/create new
data files of a different endinanness.

Some of the bug fixes improve accuracy, and so do some of the changes to
DangAmbigs and user-words.

Tesseract release notes, Oct 4 2006 - V1.02.
Removed dependency on aspirin. *All* code is now licensed under Apache2.0.

Tesseract release notes, Sep 7 2006 - V1.01.

Fixes for this release:
Added mfcpch.cpp and getopt.cpp for VC++.
Fixed problem with greyscale images and no libtiff.
Stopped debug window from being used for the usage output.
Fixed load of inttemp for big-endian architectures.
Fixed some Mac compilation issues.

This version should read uncompressed 8 bit grey and 24 bit color tiffs
without having to have libtiff. It does a dumb threshold though, so don't
expect good results from poor contrast or images of natural scenes etc.

If you just run tesseract with no command line args you should now get a
sensible usage message on linux, with or without X-windows.

If you can get it to compile on a PPC Mac, it may now run correctly,
although not all the build issues are fixed yet.

Building Tesseract:
Windows:
Unpack the tar.gz archive
Open tesseract.dsw in DevStudio (preferably version 6, higher versions will be more difficult)
Set Win32 - Release as the active configuration.
Build.
Copy tesseract.exe from bin.rel up one directory level.
Run tesseract phototest.tif phototest
This will create phototest.txt.

Linux:
Unpack the tar.gz archive
./configure
make
Copy tesseract from ccmain up one directory level (or create a symbolic link)
Run tesseract phototest.tif phototest
This will create phototest.txt.