tesseract/textord/textord.cpp
Ray Smith 0e868ef377 Major change to improve layout analysis for heavily diacritic languages:
Tha, Vie, Kan, Tel etc.
There is a new overlap detector that detects when diacritics
cause a big increase in textline overlap. In such cases, diacritics from
overlap regions are kept separate from layout analysis completely, allowing
textline formation to happen without them. The diacritics are then assigned
to 0, 1 or 2 close words at the end of layout analysis, using and modifying
an old noise detection data path.
The stored diacritics are used or not during recognition according to the
character classifier's liking for them.
2015-05-12 16:47:02 -07:00

391 lines
18 KiB
C++

///////////////////////////////////////////////////////////////////////
// File: textord.cpp
// Description: The top-level text line and word finding functionality.
// Author: Ray Smith
// Created: Fri Mar 13 14:43:01 PDT 2009
//
// (C) Copyright 2009, 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 automatically generated configuration file if running autoconf.
#ifdef HAVE_CONFIG_H
#include "config_auto.h"
#endif
#include "baselinedetect.h"
#include "drawtord.h"
#include "textord.h"
#include "makerow.h"
#include "pageres.h"
#include "tordmain.h"
#include "wordseg.h"
namespace tesseract {
Textord::Textord(CCStruct* ccstruct)
: ccstruct_(ccstruct), use_cjk_fp_model_(false),
// makerow.cpp ///////////////////////////////////////////
BOOL_MEMBER(textord_single_height_mode, false,
"Script has no xheight, so use a single mode",
ccstruct_->params()),
// tospace.cpp ///////////////////////////////////////////
BOOL_MEMBER(tosp_old_to_method, false, "Space stats use prechopping?",
ccstruct_->params()),
BOOL_MEMBER(tosp_old_to_constrain_sp_kn, false,
"Constrain relative values of inter and intra-word gaps for "
"old_to_method.",
ccstruct_->params()),
BOOL_MEMBER(tosp_only_use_prop_rows, true,
"Block stats to use fixed pitch rows?",
ccstruct_->params()),
BOOL_MEMBER(tosp_force_wordbreak_on_punct, false,
"Force word breaks on punct to break long lines in non-space "
"delimited langs",
ccstruct_->params()),
BOOL_MEMBER(tosp_use_pre_chopping, false,
"Space stats use prechopping?",
ccstruct_->params()),
BOOL_MEMBER(tosp_old_to_bug_fix, false, "Fix suspected bug in old code",
ccstruct_->params()),
BOOL_MEMBER(tosp_block_use_cert_spaces, true,
"Only stat OBVIOUS spaces",
ccstruct_->params()),
BOOL_MEMBER(tosp_row_use_cert_spaces, true, "Only stat OBVIOUS spaces",
ccstruct_->params()),
BOOL_MEMBER(tosp_narrow_blobs_not_cert, true,
"Only stat OBVIOUS spaces",
ccstruct_->params()),
BOOL_MEMBER(tosp_row_use_cert_spaces1, true, "Only stat OBVIOUS spaces",
ccstruct_->params()),
BOOL_MEMBER(tosp_recovery_isolated_row_stats, true,
"Use row alone when inadequate cert spaces",
ccstruct_->params()),
BOOL_MEMBER(tosp_only_small_gaps_for_kern, false, "Better guess",
ccstruct_->params()),
BOOL_MEMBER(tosp_all_flips_fuzzy, false, "Pass ANY flip to context?",
ccstruct_->params()),
BOOL_MEMBER(tosp_fuzzy_limit_all, true,
"Dont restrict kn->sp fuzzy limit to tables",
ccstruct_->params()),
BOOL_MEMBER(tosp_stats_use_xht_gaps, true,
"Use within xht gap for wd breaks",
ccstruct_->params()),
BOOL_MEMBER(tosp_use_xht_gaps, true, "Use within xht gap for wd breaks",
ccstruct_->params()),
BOOL_MEMBER(tosp_only_use_xht_gaps, false,
"Only use within xht gap for wd breaks",
ccstruct_->params()),
BOOL_MEMBER(tosp_rule_9_test_punct, false,
"Dont chng kn to space next to punct",
ccstruct_->params()),
BOOL_MEMBER(tosp_flip_fuzz_kn_to_sp, true, "Default flip",
ccstruct_->params()),
BOOL_MEMBER(tosp_flip_fuzz_sp_to_kn, true, "Default flip",
ccstruct_->params()),
BOOL_MEMBER(tosp_improve_thresh, false, "Enable improvement heuristic",
ccstruct_->params()),
INT_MEMBER(tosp_debug_level, 0, "Debug data",
ccstruct_->params()),
INT_MEMBER(tosp_enough_space_samples_for_median, 3,
"or should we use mean",
ccstruct_->params()),
INT_MEMBER(tosp_redo_kern_limit, 10,
"No.samples reqd to reestimate for row",
ccstruct_->params()),
INT_MEMBER(tosp_few_samples, 40,
"No.gaps reqd with 1 large gap to treat as a table",
ccstruct_->params()),
INT_MEMBER(tosp_short_row, 20,
"No.gaps reqd with few cert spaces to use certs",
ccstruct_->params()),
INT_MEMBER(tosp_sanity_method, 1, "How to avoid being silly",
ccstruct_->params()),
double_MEMBER(tosp_old_sp_kn_th_factor, 2.0,
"Factor for defining space threshold in terms of space and "
"kern sizes",
ccstruct_->params()),
double_MEMBER(tosp_threshold_bias1, 0,
"how far between kern and space?",
ccstruct_->params()),
double_MEMBER(tosp_threshold_bias2, 0,
"how far between kern and space?",
ccstruct_->params()),
double_MEMBER(tosp_narrow_fraction, 0.3, "Fract of xheight for narrow",
ccstruct_->params()),
double_MEMBER(tosp_narrow_aspect_ratio, 0.48,
"narrow if w/h less than this",
ccstruct_->params()),
double_MEMBER(tosp_wide_fraction, 0.52, "Fract of xheight for wide",
ccstruct_->params()),
double_MEMBER(tosp_wide_aspect_ratio, 0.0, "wide if w/h less than this",
ccstruct_->params()),
double_MEMBER(tosp_fuzzy_space_factor, 0.6,
"Fract of xheight for fuzz sp",
ccstruct_->params()),
double_MEMBER(tosp_fuzzy_space_factor1, 0.5,
"Fract of xheight for fuzz sp",
ccstruct_->params()),
double_MEMBER(tosp_fuzzy_space_factor2, 0.72,
"Fract of xheight for fuzz sp",
ccstruct_->params()),
double_MEMBER(tosp_gap_factor, 0.83, "gap ratio to flip sp->kern",
ccstruct_->params()),
double_MEMBER(tosp_kern_gap_factor1, 2.0, "gap ratio to flip kern->sp",
ccstruct_->params()),
double_MEMBER(tosp_kern_gap_factor2, 1.3, "gap ratio to flip kern->sp",
ccstruct_->params()),
double_MEMBER(tosp_kern_gap_factor3, 2.5, "gap ratio to flip kern->sp",
ccstruct_->params()),
double_MEMBER(tosp_ignore_big_gaps, -1, "xht multiplier",
ccstruct_->params()),
double_MEMBER(tosp_ignore_very_big_gaps, 3.5, "xht multiplier",
ccstruct_->params()),
double_MEMBER(tosp_rep_space, 1.6, "rep gap multiplier for space",
ccstruct_->params()),
double_MEMBER(tosp_enough_small_gaps, 0.65,
"Fract of kerns reqd for isolated row stats",
ccstruct_->params()),
double_MEMBER(tosp_table_kn_sp_ratio, 2.25,
"Min difference of kn & sp in table",
ccstruct_->params()),
double_MEMBER(tosp_table_xht_sp_ratio, 0.33,
"Expect spaces bigger than this",
ccstruct_->params()),
double_MEMBER(tosp_table_fuzzy_kn_sp_ratio, 3.0,
"Fuzzy if less than this",
ccstruct_->params()),
double_MEMBER(tosp_fuzzy_kn_fraction, 0.5, "New fuzzy kn alg",
ccstruct_->params()),
double_MEMBER(tosp_fuzzy_sp_fraction, 0.5, "New fuzzy sp alg",
ccstruct_->params()),
double_MEMBER(tosp_min_sane_kn_sp, 1.5,
"Dont trust spaces less than this time kn",
ccstruct_->params()),
double_MEMBER(tosp_init_guess_kn_mult, 2.2,
"Thresh guess - mult kn by this",
ccstruct_->params()),
double_MEMBER(tosp_init_guess_xht_mult, 0.28,
"Thresh guess - mult xht by this",
ccstruct_->params()),
double_MEMBER(tosp_max_sane_kn_thresh, 5.0,
"Multiplier on kn to limit thresh",
ccstruct_->params()),
double_MEMBER(tosp_flip_caution, 0.0,
"Dont autoflip kn to sp when large separation",
ccstruct_->params()),
double_MEMBER(tosp_large_kerning, 0.19,
"Limit use of xht gap with large kns",
ccstruct_->params()),
double_MEMBER(tosp_dont_fool_with_small_kerns, -1,
"Limit use of xht gap with odd small kns",
ccstruct_->params()),
double_MEMBER(tosp_near_lh_edge, 0,
"Dont reduce box if the top left is non blank",
ccstruct_->params()),
double_MEMBER(tosp_silly_kn_sp_gap, 0.2,
"Dont let sp minus kn get too small",
ccstruct_->params()),
double_MEMBER(tosp_pass_wide_fuzz_sp_to_context, 0.75,
"How wide fuzzies need context",
ccstruct_->params()),
// tordmain.cpp ///////////////////////////////////////////
BOOL_MEMBER(textord_no_rejects, false, "Don't remove noise blobs",
ccstruct_->params()),
BOOL_MEMBER(textord_show_blobs, false, "Display unsorted blobs",
ccstruct_->params()),
BOOL_MEMBER(textord_show_boxes, false, "Display unsorted blobs",
ccstruct_->params()),
INT_MEMBER(textord_max_noise_size, 7, "Pixel size of noise",
ccstruct_->params()),
INT_MEMBER(textord_baseline_debug, 0, "Baseline debug level",
ccstruct_->params()),
double_MEMBER(textord_blob_size_bigile, 95, "Percentile for large blobs",
ccstruct_->params()),
double_MEMBER(textord_noise_area_ratio, 0.7,
"Fraction of bounding box for noise",
ccstruct_->params()),
double_MEMBER(textord_blob_size_smallile, 20,
"Percentile for small blobs",
ccstruct_->params()),
double_MEMBER(textord_initialx_ile, 0.75,
"Ile of sizes for xheight guess",
ccstruct_->params()),
double_MEMBER(textord_initialasc_ile, 0.90,
"Ile of sizes for xheight guess",
ccstruct_->params()),
INT_MEMBER(textord_noise_sizefraction, 10,
"Fraction of size for maxima",
ccstruct_->params()),
double_MEMBER(textord_noise_sizelimit, 0.5,
"Fraction of x for big t count",
ccstruct_->params()),
INT_MEMBER(textord_noise_translimit, 16, "Transitions for normal blob",
ccstruct_->params()),
double_MEMBER(textord_noise_normratio, 2.0,
"Dot to norm ratio for deletion",
ccstruct_->params()),
BOOL_MEMBER(textord_noise_rejwords, true, "Reject noise-like words",
ccstruct_->params()),
BOOL_MEMBER(textord_noise_rejrows, true, "Reject noise-like rows",
ccstruct_->params()),
double_MEMBER(textord_noise_syfract, 0.2,
"xh fract height error for norm blobs",
ccstruct_->params()),
double_MEMBER(textord_noise_sxfract, 0.4,
"xh fract width error for norm blobs",
ccstruct_->params()),
double_MEMBER(textord_noise_hfract, 1.0/64,
"Height fraction to discard outlines as speckle noise",
ccstruct_->params()),
INT_MEMBER(textord_noise_sncount, 1, "super norm blobs to save row",
ccstruct_->params()),
double_MEMBER(textord_noise_rowratio, 6.0,
"Dot to norm ratio for deletion",
ccstruct_->params()),
BOOL_MEMBER(textord_noise_debug, false, "Debug row garbage detector",
ccstruct_->params()),
double_MEMBER(textord_blshift_maxshift, 0.00, "Max baseline shift",
ccstruct_->params()),
double_MEMBER(textord_blshift_xfraction, 9.99,
"Min size of baseline shift",
ccstruct_->params()) {
}
Textord::~Textord() {
}
// Make the textlines and words inside each block.
void Textord::TextordPage(PageSegMode pageseg_mode, const FCOORD& reskew,
int width, int height, Pix* binary_pix,
Pix* thresholds_pix, Pix* grey_pix,
bool use_box_bottoms, BLOBNBOX_LIST* diacritic_blobs,
BLOCK_LIST* blocks, TO_BLOCK_LIST* to_blocks) {
page_tr_.set_x(width);
page_tr_.set_y(height);
if (to_blocks->empty()) {
// AutoPageSeg was not used, so we need to find_components first.
find_components(binary_pix, blocks, to_blocks);
TO_BLOCK_IT it(to_blocks);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
TO_BLOCK* to_block = it.data();
// Compute the edge offsets whether or not there is a grey_pix.
// We have by-passed auto page seg, so we have to run it here.
// By page segmentation mode there is no non-text to avoid running on.
to_block->ComputeEdgeOffsets(thresholds_pix, grey_pix);
}
} else if (!PSM_SPARSE(pageseg_mode)) {
// AutoPageSeg does not need to find_components as it did that already.
// Filter_blobs sets up the TO_BLOCKs the same as find_components does.
filter_blobs(page_tr_, to_blocks, true);
}
ASSERT_HOST(!to_blocks->empty());
if (pageseg_mode == PSM_SINGLE_BLOCK_VERT_TEXT) {
const FCOORD anticlockwise90(0.0f, 1.0f);
const FCOORD clockwise90(0.0f, -1.0f);
TO_BLOCK_IT it(to_blocks);
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward()) {
TO_BLOCK* to_block = it.data();
BLOCK* block = to_block->block;
// Create a fake poly_block in block from its bounding box.
block->set_poly_block(new POLY_BLOCK(block->bounding_box(),
PT_VERTICAL_TEXT));
// Rotate the to_block along with its contained block and blobnbox lists.
to_block->rotate(anticlockwise90);
// Set the block's rotation values to obey the convention followed in
// layout analysis for vertical text.
block->set_re_rotation(clockwise90);
block->set_classify_rotation(clockwise90);
}
}
TO_BLOCK_IT to_block_it(to_blocks);
TO_BLOCK* to_block = to_block_it.data();
// Make the rows in the block.
float gradient;
// Do it the old fashioned way.
if (PSM_LINE_FIND_ENABLED(pageseg_mode)) {
gradient = make_rows(page_tr_, to_blocks);
} else if (!PSM_SPARSE(pageseg_mode)) {
// RAW_LINE, SINGLE_LINE, SINGLE_WORD and SINGLE_CHAR all need a single row.
gradient = make_single_row(page_tr_, pageseg_mode != PSM_RAW_LINE,
to_block, to_blocks);
}
BaselineDetect baseline_detector(textord_baseline_debug,
reskew, to_blocks);
baseline_detector.ComputeStraightBaselines(use_box_bottoms);
baseline_detector.ComputeBaselineSplinesAndXheights(page_tr_, true,
textord_heavy_nr,
textord_show_final_rows,
this);
// Now make the words in the lines.
if (PSM_WORD_FIND_ENABLED(pageseg_mode)) {
// SINGLE_LINE uses the old word maker on the single line.
make_words(this, page_tr_, gradient, blocks, to_blocks);
} else {
// SINGLE_WORD and SINGLE_CHAR cram all the blobs into a
// single word, and in SINGLE_CHAR mode, all the outlines
// go in a single blob.
TO_BLOCK* to_block = to_block_it.data();
make_single_word(pageseg_mode == PSM_SINGLE_CHAR,
to_block->get_rows(), to_block->block->row_list());
}
// Remove empties.
cleanup_blocks(PSM_WORD_FIND_ENABLED(pageseg_mode), blocks);
TransferDiacriticsToBlockGroups(diacritic_blobs, blocks);
// Compute the margins for each row in the block, to be used later for
// paragraph detection.
BLOCK_IT b_it(blocks);
for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
b_it.data()->compute_row_margins();
}
#ifndef GRAPHICS_DISABLED
close_to_win();
#endif
}
// If we were supposed to return only a single textline, and there is more
// than one, clean up and leave only the best.
void Textord::CleanupSingleRowResult(PageSegMode pageseg_mode,
PAGE_RES* page_res) {
if (PSM_LINE_FIND_ENABLED(pageseg_mode) || PSM_SPARSE(pageseg_mode))
return; // No cleanup required.
PAGE_RES_IT it(page_res);
// Find the best row, being the greatest mean word conf.
float row_total_conf = 0.0f;
int row_word_count = 0;
ROW_RES* best_row = NULL;
float best_conf = 0.0f;
for (it.restart_page(); it.word() != NULL; it.forward()) {
WERD_RES* word = it.word();
row_total_conf += word->best_choice->certainty();
++row_word_count;
if (it.next_row() != it.row()) {
row_total_conf /= row_word_count;
if (best_row == NULL || best_conf < row_total_conf) {
best_row = it.row();
best_conf = row_total_conf;
}
row_total_conf = 0.0f;
row_word_count = 0;
}
}
// Now eliminate any word not in the best row.
for (it.restart_page(); it.word() != NULL; it.forward()) {
if (it.row() != best_row)
it.DeleteCurrentWord();
}
}
} // namespace tesseract.