tesseract  4.1.1
control.cpp
Go to the documentation of this file.
1 /******************************************************************
2  * File: control.cpp (Formerly control.c)
3  * Description: Module-independent matcher controller.
4  * Author: Ray Smith
5  *
6  * (C) Copyright 1992, Hewlett-Packard Ltd.
7  ** Licensed under the Apache License, Version 2.0 (the "License");
8  ** you may not use this file except in compliance with the License.
9  ** You may obtain a copy of the License at
10  ** http://www.apache.org/licenses/LICENSE-2.0
11  ** Unless required by applicable law or agreed to in writing, software
12  ** distributed under the License is distributed on an "AS IS" BASIS,
13  ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14  ** See the License for the specific language governing permissions and
15  ** limitations under the License.
16  *
17  **********************************************************************/
18 
19 // Include automatically generated configuration file if running autoconf.
20 #ifdef HAVE_CONFIG_H
21 #include "config_auto.h"
22 #endif
23 
24 #include <cmath>
25 #include <cstdint> // for int16_t, int32_t
26 #include <cstdio> // for fclose, fopen, FILE
27 #include <ctime> // for clock
28 #include <cctype>
29 #include "callcpp.h"
30 #include "control.h"
31 #ifndef DISABLED_LEGACY_ENGINE
32 #include "docqual.h"
33 #include "drawfx.h"
34 #include "fixspace.h"
35 #endif
36 #include "lstmrecognizer.h"
37 #include "ocrclass.h"
38 #include "output.h"
39 #include "pageres.h" // for WERD_RES, PAGE_RES_IT, PAGE_RES, BLO...
40 #ifndef DISABLED_LEGACY_ENGINE
41 #include "reject.h"
42 #endif
43 #include "sorthelper.h"
44 #include "tesseractclass.h"
45 #include "tessvars.h"
46 #include "werdit.h"
47 
48 const char* const kBackUpConfigFile = "tempconfigdata.config";
49 // Min believable x-height for any text when refitting as a fraction of
50 // original x-height
51 const double kMinRefitXHeightFraction = 0.5;
52 
53 
60 namespace tesseract {
61 
63  TBOX &selection_box) {
64  PAGE_RES_IT* it = make_pseudo_word(page_res, selection_box);
65  if (it != nullptr) {
67  it->DeleteCurrentWord();
68  delete it;
69  }
70 }
71 
78  int16_t char_qual;
79  int16_t good_char_qual;
80 
81  WordData word_data(*pr_it);
82  SetupWordPassN(2, &word_data);
83  // LSTM doesn't run on pass2, but we want to run pass2 for tesseract.
84  if (lstm_recognizer_ == nullptr) {
85 #ifndef DISABLED_LEGACY_ENGINE
86  classify_word_and_language(2, pr_it, &word_data);
87 #endif // ndef DISABLED_LEGACY_ENGINE
88  } else {
89  classify_word_and_language(1, pr_it, &word_data);
90  }
91 #ifndef DISABLED_LEGACY_ENGINE
93  WERD_RES* word_res = pr_it->word();
94  word_char_quality(word_res, pr_it->row()->row, &char_qual, &good_char_qual);
95  tprintf("\n%d chars; word_blob_quality: %d; outline_errs: %d; "
96  "char_quality: %d; good_char_quality: %d\n",
97  word_res->reject_map.length(),
98  word_blob_quality(word_res, pr_it->row()->row),
99  word_outline_errs(word_res), char_qual, good_char_qual);
100  }
101 #endif // ndef DISABLED_LEGACY_ENGINE
102  return true;
103 }
104 
105 // Helper function to check for a target word and handle it appropriately.
106 // Inspired by Jetsoft's requirement to process only single words on pass2
107 // and beyond.
108 // If word_config is not null:
109 // If the word_box and target_word_box overlap, read the word_config file
110 // else reset to previous config data.
111 // return true.
112 // else
113 // If the word_box and target_word_box overlap or pass <= 1, return true.
114 // Note that this function uses a fixed temporary file for storing the previous
115 // configs, so it is neither thread-safe, nor process-safe, but the assumption
116 // is that it will only be used for one debug window at a time.
117 //
118 // Since this function is used for debugging (and not to change OCR results)
119 // set only debug params from the word config file.
120 bool Tesseract::ProcessTargetWord(const TBOX& word_box,
121  const TBOX& target_word_box,
122  const char* word_config,
123  int pass) {
124  if (word_config != nullptr) {
125  if (word_box.major_overlap(target_word_box)) {
126  if (backup_config_file_ == nullptr) {
127  backup_config_file_ = kBackUpConfigFile;
128  FILE* config_fp = fopen(backup_config_file_, "wb");
129  if (config_fp == nullptr) {
130  tprintf("Error, failed to open file \"%s\"\n", backup_config_file_);
131  } else {
132  ParamUtils::PrintParams(config_fp, params());
133  fclose(config_fp);
134  }
135  ParamUtils::ReadParamsFile(word_config,
137  params());
138  }
139  } else {
140  if (backup_config_file_ != nullptr) {
141  ParamUtils::ReadParamsFile(backup_config_file_,
143  params());
144  backup_config_file_ = nullptr;
145  }
146  }
147  } else if (pass > 1 && !word_box.major_overlap(target_word_box)) {
148  return false;
149  }
150  return true;
151 }
152 
155  const TBOX* target_word_box,
156  const char* word_config,
157  PAGE_RES* page_res,
158  GenericVector<WordData>* words) {
159  // Prepare all the words.
160  PAGE_RES_IT page_res_it(page_res);
161  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
162  page_res_it.forward()) {
163  if (target_word_box == nullptr ||
164  ProcessTargetWord(page_res_it.word()->word->bounding_box(),
165  *target_word_box, word_config, 1)) {
166  words->push_back(WordData(page_res_it));
167  }
168  }
169  // Setup all the words for recognition with polygonal approximation.
170  for (int w = 0; w < words->size(); ++w) {
171  SetupWordPassN(pass_n, &(*words)[w]);
172  if (w > 0) (*words)[w].prev_word = &(*words)[w - 1];
173  }
174 }
175 
176 // Sets up the single word ready for whichever engine is to be run.
177 void Tesseract::SetupWordPassN(int pass_n, WordData* word) {
178  if (pass_n == 1 || !word->word->done) {
179  if (pass_n == 1) {
180  word->word->SetupForRecognition(unicharset, this, BestPix(),
181  tessedit_ocr_engine_mode, nullptr,
185  word->row, word->block);
186  } else if (pass_n == 2) {
187  // TODO(rays) Should we do this on pass1 too?
188  word->word->caps_height = 0.0;
189  if (word->word->x_height == 0.0f)
190  word->word->x_height = word->row->x_height();
191  }
192  word->lang_words.truncate(0);
193  for (int s = 0; s <= sub_langs_.size(); ++s) {
194  // The sub_langs_.size() entry is for the master language.
195  Tesseract* lang_t = s < sub_langs_.size() ? sub_langs_[s] : this;
196  auto* word_res = new WERD_RES;
197  word_res->InitForRetryRecognition(*word->word);
198  word->lang_words.push_back(word_res);
199  // LSTM doesn't get setup for pass2.
200  if (pass_n == 1 || lang_t->tessedit_ocr_engine_mode != OEM_LSTM_ONLY) {
201  word_res->SetupForRecognition(
202  lang_t->unicharset, lang_t, BestPix(),
203  lang_t->tessedit_ocr_engine_mode, nullptr,
205  lang_t->textord_use_cjk_fp_model,
206  lang_t->poly_allow_detailed_fx, word->row, word->block);
207  }
208  }
209  }
210 }
211 
212 // Runs word recognition on all the words.
213 bool Tesseract::RecogAllWordsPassN(int pass_n, ETEXT_DESC* monitor,
214  PAGE_RES_IT* pr_it,
215  GenericVector<WordData>* words) {
216  // TODO(rays) Before this loop can be parallelized (it would yield a massive
217  // speed-up) all remaining member globals need to be converted to local/heap
218  // (eg set_pass1 and set_pass2) and an intermediate adaption pass needs to be
219  // added. The results will be significantly different with adaption on, and
220  // deterioration will need investigation.
221  pr_it->restart_page();
222  for (int w = 0; w < words->size(); ++w) {
223  WordData* word = &(*words)[w];
224  if (w > 0) word->prev_word = &(*words)[w - 1];
225  if (monitor != nullptr) {
226  monitor->ocr_alive = true;
227  if (pass_n == 1) {
228  monitor->progress = 70 * w / words->size();
229  } else {
230  monitor->progress = 70 + 30 * w / words->size();
231  }
232  if (monitor->progress_callback2 != nullptr) {
233  TBOX box = pr_it->word()->word->bounding_box();
234  (*monitor->progress_callback2)(monitor, box.left(),
235  box.right(), box.top(), box.bottom());
236  }
237  if (monitor->deadline_exceeded() ||
238  (monitor->cancel != nullptr && (*monitor->cancel)(monitor->cancel_this,
239  words->size()))) {
240  // Timeout. Fake out the rest of the words.
241  for (; w < words->size(); ++w) {
242  (*words)[w].word->SetupFake(unicharset);
243  }
244  return false;
245  }
246  }
247  if (word->word->tess_failed) {
248  int s;
249  for (s = 0; s < word->lang_words.size() &&
250  word->lang_words[s]->tess_failed; ++s) {}
251  // If all are failed, skip it. Image words are skipped by this test.
252  if (s > word->lang_words.size()) continue;
253  }
254  // Sync pr_it with the wth WordData.
255  while (pr_it->word() != nullptr && pr_it->word() != word->word)
256  pr_it->forward();
257  ASSERT_HOST(pr_it->word() != nullptr);
258  bool make_next_word_fuzzy = false;
259  #ifndef DISABLED_LEGACY_ENGINE
260  if (!AnyLSTMLang() &&
261  ReassignDiacritics(pass_n, pr_it, &make_next_word_fuzzy)) {
262  // Needs to be setup again to see the new outlines in the chopped_word.
263  SetupWordPassN(pass_n, word);
264  }
265  #endif // ndef DISABLED_LEGACY_ENGINE
266 
267  classify_word_and_language(pass_n, pr_it, word);
269  tprintf("Pass%d: %s [%s]\n", pass_n,
271  word->word->best_choice->debug_string().string());
272  }
273  pr_it->forward();
274  if (make_next_word_fuzzy && pr_it->word() != nullptr) {
275  pr_it->MakeCurrentWordFuzzy();
276  }
277  }
278  return true;
279 }
280 
303  ETEXT_DESC* monitor,
304  const TBOX* target_word_box,
305  const char* word_config,
306  int dopasses) {
307  PAGE_RES_IT page_res_it(page_res);
308 
310  tessedit_test_adaption.set_value (true);
311  tessedit_minimal_rejection.set_value (true);
312  }
313 
314  if (dopasses==0 || dopasses==1) {
315  page_res_it.restart_page();
316  // ****************** Pass 1 *******************
317 
318  #ifndef DISABLED_LEGACY_ENGINE
319  // If the adaptive classifier is full switch to one we prepared earlier,
320  // ie on the previous page. If the current adaptive classifier is non-empty,
321  // prepare a backup starting at this page, in case it fills up. Do all this
322  // independently for each language.
323  if (AdaptiveClassifierIsFull()) {
325  } else if (!AdaptiveClassifierIsEmpty()) {
327  }
328  // Now check the sub-langs as well.
329  for (int i = 0; i < sub_langs_.size(); ++i) {
330  if (sub_langs_[i]->AdaptiveClassifierIsFull()) {
331  sub_langs_[i]->SwitchAdaptiveClassifier();
332  } else if (!sub_langs_[i]->AdaptiveClassifierIsEmpty()) {
333  sub_langs_[i]->StartBackupAdaptiveClassifier();
334  }
335  }
336 
337  #endif // ndef DISABLED_LEGACY_ENGINE
338 
339  // Set up all words ready for recognition, so that if parallelism is on
340  // all the input and output classes are ready to run the classifier.
342  SetupAllWordsPassN(1, target_word_box, word_config, page_res, &words);
343  #ifndef DISABLED_LEGACY_ENGINE
344  if (tessedit_parallelize) {
345  PrerecAllWordsPar(words);
346  }
347  #endif // ndef DISABLED_LEGACY_ENGINE
348 
349  stats_.word_count = words.size();
350 
351  stats_.dict_words = 0;
352  stats_.doc_blob_quality = 0;
353  stats_.doc_outline_errs = 0;
354  stats_.doc_char_quality = 0;
355  stats_.good_char_count = 0;
356  stats_.doc_good_char_quality = 0;
357 
358  most_recently_used_ = this;
359  // Run pass 1 word recognition.
360  if (!RecogAllWordsPassN(1, monitor, &page_res_it, &words)) return false;
361  // Pass 1 post-processing.
362  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
363  page_res_it.forward()) {
364  if (page_res_it.word()->word->flag(W_REP_CHAR)) {
365  fix_rep_char(&page_res_it);
366  continue;
367  }
368 
369  // Count dict words.
370  if (page_res_it.word()->best_choice->permuter() == USER_DAWG_PERM)
371  ++(stats_.dict_words);
372 
373  // Update misadaption log (we only need to do it on pass 1, since
374  // adaption only happens on this pass).
375  if (page_res_it.word()->blamer_bundle != nullptr &&
376  page_res_it.word()->blamer_bundle->misadaption_debug().length() > 0) {
377  page_res->misadaption_log.push_back(
378  page_res_it.word()->blamer_bundle->misadaption_debug());
379  }
380  }
381  }
382 
383  if (dopasses == 1) return true;
384 
385  #ifndef DISABLED_LEGACY_ENGINE
386 
387  // ****************** Pass 2 *******************
389  AnyTessLang()) {
390  page_res_it.restart_page();
392  SetupAllWordsPassN(2, target_word_box, word_config, page_res, &words);
393  if (tessedit_parallelize) {
394  PrerecAllWordsPar(words);
395  }
396  most_recently_used_ = this;
397  // Run pass 2 word recognition.
398  if (!RecogAllWordsPassN(2, monitor, &page_res_it, &words)) return false;
399  }
400 
401  // The next passes are only required for Tess-only.
402  if (AnyTessLang() && !AnyLSTMLang()) {
403  // ****************** Pass 3 *******************
404  // Fix fuzzy spaces.
406 
409  fix_fuzzy_spaces(monitor, stats_.word_count, page_res);
410 
411  // ****************** Pass 4 *******************
414 
415  // ****************** Pass 5,6 *******************
416  rejection_passes(page_res, monitor, target_word_box, word_config);
417 
418  // ****************** Pass 8 *******************
419  font_recognition_pass(page_res);
420 
421  // ****************** Pass 9 *******************
422  // Check the correctness of the final results.
423  blamer_pass(page_res);
424  script_pos_pass(page_res);
425  }
426 
427  #endif // ndef DISABLED_LEGACY_ENGINE
428 
429  // Write results pass.
431  // This is now redundant, but retained commented so show how to obtain
432  // bounding boxes and style information.
433 
434  #ifndef DISABLED_LEGACY_ENGINE
435  // changed by jetsoft
436  // needed for dll to output memory structure
437  if ((dopasses == 0 || dopasses == 2) && (monitor || tessedit_write_unlv))
438  output_pass(page_res_it, target_word_box);
439  // end jetsoft
440  #endif //ndef DISABLED_LEGACY_ENGINE
441 
442  const auto pageseg_mode = static_cast<PageSegMode>(
443  static_cast<int>(tessedit_pageseg_mode));
444  textord_.CleanupSingleRowResult(pageseg_mode, page_res);
445 
446  // Remove empty words, as these mess up the result iterators.
447  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
448  page_res_it.forward()) {
449  const WERD_RES* word = page_res_it.word();
450  const POLY_BLOCK* pb = page_res_it.block()->block != nullptr
451  ? page_res_it.block()->block->pdblk.poly_block()
452  : nullptr;
453  if (word->best_choice == nullptr || word->best_choice->length() == 0 ||
454  (word->best_choice->IsAllSpaces() && (pb == nullptr || pb->IsText()))) {
455  page_res_it.DeleteCurrentWord();
456  }
457  }
458 
459  if (monitor != nullptr) {
460  monitor->progress = 100;
461  }
462  return true;
463 }
464 
465 #ifndef DISABLED_LEGACY_ENGINE
466 
468  PAGE_RES_IT word_it(page_res);
469 
470  WERD_RES *w_prev = nullptr;
471  WERD_RES *w = word_it.word();
472  while (true) {
473  w_prev = w;
474  while (word_it.forward() != nullptr &&
475  (!word_it.word() || word_it.word()->part_of_combo)) {
476  // advance word_it, skipping over parts of combos
477  }
478  if (!word_it.word()) break;
479  w = word_it.word();
480  if (!w || !w_prev || w->uch_set != w_prev->uch_set) {
481  continue;
482  }
483  if (w_prev->word->flag(W_REP_CHAR) || w->word->flag(W_REP_CHAR)) {
484  if (tessedit_bigram_debug) {
485  tprintf("Skipping because one of the words is W_REP_CHAR\n");
486  }
487  continue;
488  }
489  // Two words sharing the same language model, excellent!
490  GenericVector<WERD_CHOICE *> overrides_word1;
491  GenericVector<WERD_CHOICE *> overrides_word2;
492 
493  const STRING orig_w1_str = w_prev->best_choice->unichar_string();
494  const STRING orig_w2_str = w->best_choice->unichar_string();
495  WERD_CHOICE prev_best(w->uch_set);
496  {
497  int w1start, w1end;
498  w_prev->best_choice->GetNonSuperscriptSpan(&w1start, &w1end);
499  prev_best = w_prev->best_choice->shallow_copy(w1start, w1end);
500  }
501  WERD_CHOICE this_best(w->uch_set);
502  {
503  int w2start, w2end;
504  w->best_choice->GetNonSuperscriptSpan(&w2start, &w2end);
505  this_best = w->best_choice->shallow_copy(w2start, w2end);
506  }
507 
508  if (w->tesseract->getDict().valid_bigram(prev_best, this_best)) {
509  if (tessedit_bigram_debug) {
510  tprintf("Top choice \"%s %s\" verified by bigram model.\n",
511  orig_w1_str.string(), orig_w2_str.string());
512  }
513  continue;
514  }
515  if (tessedit_bigram_debug > 2) {
516  tprintf("Examining alt choices for \"%s %s\".\n",
517  orig_w1_str.string(), orig_w2_str.string());
518  }
519  if (tessedit_bigram_debug > 1) {
520  if (!w_prev->best_choices.singleton()) {
521  w_prev->PrintBestChoices();
522  }
523  if (!w->best_choices.singleton()) {
524  w->PrintBestChoices();
525  }
526  }
527  float best_rating = 0.0;
528  int best_idx = 0;
529  WERD_CHOICE_IT prev_it(&w_prev->best_choices);
530  for (prev_it.mark_cycle_pt(); !prev_it.cycled_list(); prev_it.forward()) {
531  WERD_CHOICE *p1 = prev_it.data();
532  WERD_CHOICE strip1(w->uch_set);
533  {
534  int p1start, p1end;
535  p1->GetNonSuperscriptSpan(&p1start, &p1end);
536  strip1 = p1->shallow_copy(p1start, p1end);
537  }
538  WERD_CHOICE_IT w_it(&w->best_choices);
539  for (w_it.mark_cycle_pt(); !w_it.cycled_list(); w_it.forward()) {
540  WERD_CHOICE *p2 = w_it.data();
541  WERD_CHOICE strip2(w->uch_set);
542  {
543  int p2start, p2end;
544  p2->GetNonSuperscriptSpan(&p2start, &p2end);
545  strip2 = p2->shallow_copy(p2start, p2end);
546  }
547  if (w->tesseract->getDict().valid_bigram(strip1, strip2)) {
548  overrides_word1.push_back(p1);
549  overrides_word2.push_back(p2);
550  if (overrides_word1.size() == 1 ||
551  p1->rating() + p2->rating() < best_rating) {
552  best_rating = p1->rating() + p2->rating();
553  best_idx = overrides_word1.size() - 1;
554  }
555  }
556  }
557  }
558  if (!overrides_word1.empty()) {
559  // Excellent, we have some bigram matches.
561  *overrides_word1[best_idx]) &&
563  *overrides_word2[best_idx])) {
564  if (tessedit_bigram_debug > 1) {
565  tprintf("Top choice \"%s %s\" verified (sans case) by bigram "
566  "model.\n", orig_w1_str.string(), orig_w2_str.string());
567  }
568  continue;
569  }
570  const STRING new_w1_str = overrides_word1[best_idx]->unichar_string();
571  const STRING new_w2_str = overrides_word2[best_idx]->unichar_string();
572  if (new_w1_str != orig_w1_str) {
573  w_prev->ReplaceBestChoice(overrides_word1[best_idx]);
574  }
575  if (new_w2_str != orig_w2_str) {
576  w->ReplaceBestChoice(overrides_word2[best_idx]);
577  }
578  if (tessedit_bigram_debug > 0) {
579  STRING choices_description;
580  int num_bigram_choices
581  = overrides_word1.size() * overrides_word2.size();
582  if (num_bigram_choices == 1) {
583  choices_description = "This was the unique bigram choice.";
584  } else {
585  if (tessedit_bigram_debug > 1) {
586  STRING bigrams_list;
587  const int kMaxChoicesToPrint = 20;
588  for (int i = 0; i < overrides_word1.size() &&
589  i < kMaxChoicesToPrint; i++) {
590  if (i > 0) { bigrams_list += ", "; }
591  WERD_CHOICE *p1 = overrides_word1[i];
592  WERD_CHOICE *p2 = overrides_word2[i];
593  bigrams_list += p1->unichar_string() + " " + p2->unichar_string();
594  }
595  choices_description = "There were many choices: {";
596  choices_description += bigrams_list;
597  choices_description += "}";
598  } else {
599  choices_description.add_str_int("There were ", num_bigram_choices);
600  choices_description += " compatible bigrams.";
601  }
602  }
603  tprintf("Replaced \"%s %s\" with \"%s %s\" with bigram model. %s\n",
604  orig_w1_str.string(), orig_w2_str.string(),
605  new_w1_str.string(), new_w2_str.string(),
606  choices_description.string());
607  }
608  }
609  }
610 }
611 
613  ETEXT_DESC* monitor,
614  const TBOX* target_word_box,
615  const char* word_config) {
616  PAGE_RES_IT page_res_it(page_res);
617  // ****************** Pass 5 *******************
618  // Gather statistics on rejects.
619  int word_index = 0;
620  while (!tessedit_test_adaption && page_res_it.word() != nullptr) {
622  WERD_RES* word = page_res_it.word();
623  word_index++;
624  if (monitor != nullptr) {
625  monitor->ocr_alive = true;
626  monitor->progress = 95 + 5 * word_index / stats_.word_count;
627  }
628  if (word->rebuild_word == nullptr) {
629  // Word was not processed by tesseract.
630  page_res_it.forward();
631  continue;
632  }
633  check_debug_pt(word, 70);
634 
635  // changed by jetsoft
636  // specific to its needs to extract one word when need
637  if (target_word_box &&
639  *target_word_box, word_config, 4)) {
640  page_res_it.forward();
641  continue;
642  }
643  // end jetsoft
644 
645  page_res_it.rej_stat_word();
646  const int chars_in_word = word->reject_map.length();
647  const int rejects_in_word = word->reject_map.reject_count();
648 
649  const int blob_quality = word_blob_quality(word, page_res_it.row()->row);
650  stats_.doc_blob_quality += blob_quality;
651  const int outline_errs = word_outline_errs(word);
652  stats_.doc_outline_errs += outline_errs;
653  int16_t all_char_quality;
654  int16_t accepted_all_char_quality;
655  word_char_quality(word, page_res_it.row()->row,
656  &all_char_quality, &accepted_all_char_quality);
657  stats_.doc_char_quality += all_char_quality;
658  const uint8_t permuter_type = word->best_choice->permuter();
659  if ((permuter_type == SYSTEM_DAWG_PERM) ||
660  (permuter_type == FREQ_DAWG_PERM) ||
661  (permuter_type == USER_DAWG_PERM)) {
662  stats_.good_char_count += chars_in_word - rejects_in_word;
663  stats_.doc_good_char_quality += accepted_all_char_quality;
664  }
665  check_debug_pt(word, 80);
667  (blob_quality == 0) && (outline_errs >= chars_in_word))
669  check_debug_pt(word, 90);
670  page_res_it.forward();
671  }
672 
674  tprintf
675  ("QUALITY: num_chs= %d num_rejs= %d %5.3f blob_qual= %d %5.3f"
676  " outline_errs= %d %5.3f char_qual= %d %5.3f good_ch_qual= %d %5.3f\n",
677  page_res->char_count, page_res->rej_count,
678  page_res->rej_count / static_cast<float>(page_res->char_count),
679  stats_.doc_blob_quality,
680  stats_.doc_blob_quality / static_cast<float>(page_res->char_count),
681  stats_.doc_outline_errs,
682  stats_.doc_outline_errs / static_cast<float>(page_res->char_count),
683  stats_.doc_char_quality,
684  stats_.doc_char_quality / static_cast<float>(page_res->char_count),
685  stats_.doc_good_char_quality,
686  (stats_.good_char_count > 0) ?
687  (stats_.doc_good_char_quality /
688  static_cast<float>(stats_.good_char_count)) : 0.0);
689  }
690  bool good_quality_doc =
691  ((page_res->rej_count / static_cast<float>(page_res->char_count)) <=
692  quality_rej_pc) &&
693  (stats_.doc_blob_quality / static_cast<float>(page_res->char_count) >=
694  quality_blob_pc) &&
695  (stats_.doc_outline_errs / static_cast<float>(page_res->char_count) <=
697  (stats_.doc_char_quality / static_cast<float>(page_res->char_count) >=
699 
700  // ****************** Pass 6 *******************
701  // Do whole document or whole block rejection pass
702  if (!tessedit_test_adaption) {
704  quality_based_rejection(page_res_it, good_quality_doc);
705  }
706 }
707 
708 #endif // ndef DISABLED_LEGACY_ENGINE
709 
711  if (!wordrec_run_blamer) return;
712  PAGE_RES_IT page_res_it(page_res);
713  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
714  page_res_it.forward()) {
715  WERD_RES *word = page_res_it.word();
718  }
719  tprintf("Blame reasons:\n");
720  for (int bl = 0; bl < IRR_NUM_REASONS; ++bl) {
722  static_cast<IncorrectResultReason>(bl)),
723  page_res->blame_reasons[bl]);
724  }
725  if (page_res->misadaption_log.length() > 0) {
726  tprintf("Misadaption log:\n");
727  for (int i = 0; i < page_res->misadaption_log.length(); ++i) {
728  tprintf("%s\n", page_res->misadaption_log[i].string());
729  }
730  }
731 }
732 
733 // Sets script positions and detects smallcaps on all output words.
735  PAGE_RES_IT page_res_it(page_res);
736  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
737  page_res_it.forward()) {
738  WERD_RES* word = page_res_it.word();
739  if (word->word->flag(W_REP_CHAR)) {
740  page_res_it.forward();
741  continue;
742  }
743  const float x_height = page_res_it.block()->block->x_height();
744  float word_x_height = word->x_height;
745  if (word_x_height < word->best_choice->min_x_height() ||
746  word_x_height > word->best_choice->max_x_height()) {
747  word_x_height = (word->best_choice->min_x_height() +
748  word->best_choice->max_x_height()) / 2.0f;
749  }
750  // Test for small caps. Word capheight must be close to block xheight,
751  // and word must contain no lower case letters, and at least one upper case.
752  const double small_cap_xheight = x_height * kXHeightCapRatio;
753  const double small_cap_delta = (x_height - small_cap_xheight) / 2.0;
754  if (word->uch_set->script_has_xheight() &&
755  small_cap_xheight - small_cap_delta <= word_x_height &&
756  word_x_height <= small_cap_xheight + small_cap_delta) {
757  // Scan for upper/lower.
758  int num_upper = 0;
759  int num_lower = 0;
760  for (int i = 0; i < word->best_choice->length(); ++i) {
761  if (word->uch_set->get_isupper(word->best_choice->unichar_id(i)))
762  ++num_upper;
763  else if (word->uch_set->get_islower(word->best_choice->unichar_id(i)))
764  ++num_lower;
765  }
766  if (num_upper > 0 && num_lower == 0)
767  word->small_caps = true;
768  }
769  word->SetScriptPositions();
770  }
771 }
772 
773 // Helper finds the gap between the index word and the next.
774 static void WordGap(const PointerVector<WERD_RES>& words, int index, int* right,
775  int* next_left) {
776  *right = -INT32_MAX;
777  *next_left = INT32_MAX;
778  if (index < words.size()) {
779  *right = words[index]->word->bounding_box().right();
780  if (index + 1 < words.size())
781  *next_left = words[index + 1]->word->bounding_box().left();
782  }
783 }
784 
785 // Factored helper computes the rating, certainty, badness and validity of
786 // the permuter of the words in [first_index, end_index).
787 static void EvaluateWordSpan(const PointerVector<WERD_RES>& words,
788  int first_index, int end_index, float* rating,
789  float* certainty, bool* bad,
790  bool* valid_permuter) {
791  if (end_index <= first_index) {
792  *bad = true;
793  *valid_permuter = false;
794  }
795  for (int index = first_index; index < end_index && index < words.size();
796  ++index) {
797  WERD_CHOICE* choice = words[index]->best_choice;
798  if (choice == nullptr) {
799  *bad = true;
800  } else {
801  *rating += choice->rating();
802  *certainty = std::min(*certainty, choice->certainty());
803  if (!Dict::valid_word_permuter(choice->permuter(), false))
804  *valid_permuter = false;
805  }
806  }
807 }
808 
809 // Helper chooses the best combination of words, transferring good ones from
810 // new_words to best_words. To win, a new word must have (better rating and
811 // certainty) or (better permuter status and rating within rating ratio and
812 // certainty within certainty margin) than current best.
813 // All the new_words are consumed (moved to best_words or deleted.)
814 // The return value is the number of new_words used minus the number of
815 // best_words that remain in the output.
816 static int SelectBestWords(double rating_ratio,
817  double certainty_margin,
818  bool debug,
819  PointerVector<WERD_RES>* new_words,
820  PointerVector<WERD_RES>* best_words) {
821  // Process the smallest groups of words that have an overlapping word
822  // boundary at the end.
823  GenericVector<WERD_RES*> out_words;
824  // Index into each word vector (best, new).
825  int b = 0, n = 0;
826  int num_best = 0, num_new = 0;
827  while (b < best_words->size() || n < new_words->size()) {
828  // Start of the current run in each.
829  int start_b = b, start_n = n;
830  while (b < best_words->size() || n < new_words->size()) {
831  int b_right = -INT32_MAX;
832  int next_b_left = INT32_MAX;
833  WordGap(*best_words, b, &b_right, &next_b_left);
834  int n_right = -INT32_MAX;
835  int next_n_left = INT32_MAX;
836  WordGap(*new_words, n, &n_right, &next_n_left);
837  if (std::max(b_right, n_right) < std::min(next_b_left, next_n_left)) {
838  // The word breaks overlap. [start_b,b] and [start_n, n] match.
839  break;
840  }
841  // Keep searching for the matching word break.
842  if ((b_right < n_right && b < best_words->size()) ||
843  n == new_words->size())
844  ++b;
845  else
846  ++n;
847  }
848  // Rating of the current run in each.
849  float b_rating = 0.0f, n_rating = 0.0f;
850  // Certainty of the current run in each.
851  float b_certainty = 0.0f, n_certainty = 0.0f;
852  // True if any word is missing its best choice.
853  bool b_bad = false, n_bad = false;
854  // True if all words have a valid permuter.
855  bool b_valid_permuter = true, n_valid_permuter = true;
856  const int end_b = b < best_words->size() ? b + 1 : b;
857  const int end_n = n < new_words->size() ? n + 1 : n;
858  EvaluateWordSpan(*best_words, start_b, end_b, &b_rating, &b_certainty,
859  &b_bad, &b_valid_permuter);
860  EvaluateWordSpan(*new_words, start_n, end_n, &n_rating, &n_certainty,
861  &n_bad, &n_valid_permuter);
862  bool new_better = false;
863  if (!n_bad && (b_bad || (n_certainty > b_certainty &&
864  n_rating < b_rating) ||
865  (!b_valid_permuter && n_valid_permuter &&
866  n_rating < b_rating * rating_ratio &&
867  n_certainty > b_certainty - certainty_margin))) {
868  // New is better.
869  for (int i = start_n; i < end_n; ++i) {
870  out_words.push_back((*new_words)[i]);
871  (*new_words)[i] = nullptr;
872  ++num_new;
873  }
874  new_better = true;
875  } else if (!b_bad) {
876  // Current best is better.
877  for (int i = start_b; i < end_b; ++i) {
878  out_words.push_back((*best_words)[i]);
879  (*best_words)[i] = nullptr;
880  ++num_best;
881  }
882  }
883  if (debug) {
884  tprintf("%d new words %s than %d old words: r: %g v %g c: %g v %g"
885  " valid dict: %d v %d\n",
886  end_n - start_n, new_better ? "better" : "worse",
887  end_b - start_b, n_rating, b_rating,
888  n_certainty, b_certainty, n_valid_permuter, b_valid_permuter);
889  }
890  // Move on to the next group.
891  b = end_b;
892  n = end_n;
893  }
894  // Transfer from out_words to best_words.
895  best_words->clear();
896  for (int i = 0; i < out_words.size(); ++i)
897  best_words->push_back(out_words[i]);
898  return num_new - num_best;
899 }
900 
901 // Helper to recognize the word using the given (language-specific) tesseract.
902 // Returns positive if this recognizer found more new best words than the
903 // number kept from best_words.
905  WordRecognizer recognizer, bool debug,
906  WERD_RES** in_word,
907  PointerVector<WERD_RES>* best_words) {
908  if (debug) {
909  tprintf("Trying word using lang %s, oem %d\n",
910  lang.string(), static_cast<int>(tessedit_ocr_engine_mode));
911  }
912  // Run the recognizer on the word.
913  PointerVector<WERD_RES> new_words;
914  (this->*recognizer)(word_data, in_word, &new_words);
915  if (new_words.empty()) {
916  // Transfer input word to new_words, as the classifier must have put
917  // the result back in the input.
918  new_words.push_back(*in_word);
919  *in_word = nullptr;
920  }
921  if (debug) {
922  for (int i = 0; i < new_words.size(); ++i)
923  new_words[i]->DebugTopChoice("Lang result");
924  }
925  // Initial version is a bit of a hack based on better certainty and rating
926  // or a dictionary vs non-dictionary word.
927  return SelectBestWords(classify_max_rating_ratio,
929  debug, &new_words, best_words);
930 }
931 
932 // Helper returns true if all the words are acceptable.
933 static bool WordsAcceptable(const PointerVector<WERD_RES>& words) {
934  for (int w = 0; w < words.size(); ++w) {
935  if (words[w]->tess_failed || !words[w]->tess_accepted) return false;
936  }
937  return true;
938 }
939 
940 #ifndef DISABLED_LEGACY_ENGINE
941 
942 // Moves good-looking "noise"/diacritics from the reject list to the main
943 // blob list on the current word. Returns true if anything was done, and
944 // sets make_next_word_fuzzy if blob(s) were added to the end of the word.
946  bool* make_next_word_fuzzy) {
947  *make_next_word_fuzzy = false;
948  WERD* real_word = pr_it->word()->word;
949  if (real_word->rej_cblob_list()->empty() ||
950  real_word->cblob_list()->empty() ||
951  real_word->rej_cblob_list()->length() > noise_maxperword)
952  return false;
953  real_word->rej_cblob_list()->sort(&C_BLOB::SortByXMiddle);
954  // Get the noise outlines into a vector with matching bool map.
955  GenericVector<C_OUTLINE*> outlines;
956  real_word->GetNoiseOutlines(&outlines);
957  GenericVector<bool> word_wanted;
958  GenericVector<bool> overlapped_any_blob;
959  GenericVector<C_BLOB*> target_blobs;
960  AssignDiacriticsToOverlappingBlobs(outlines, pass, real_word, pr_it,
961  &word_wanted, &overlapped_any_blob,
962  &target_blobs);
963  // Filter the outlines that overlapped any blob and put them into the word
964  // now. This simplifies the remaining task and also makes it more accurate
965  // as it has more completed blobs to work on.
966  GenericVector<bool> wanted;
967  GenericVector<C_BLOB*> wanted_blobs;
968  GenericVector<C_OUTLINE*> wanted_outlines;
969  int num_overlapped = 0;
970  int num_overlapped_used = 0;
971  for (int i = 0; i < overlapped_any_blob.size(); ++i) {
972  if (overlapped_any_blob[i]) {
973  ++num_overlapped;
974  if (word_wanted[i]) ++num_overlapped_used;
975  wanted.push_back(word_wanted[i]);
976  wanted_blobs.push_back(target_blobs[i]);
977  wanted_outlines.push_back(outlines[i]);
978  outlines[i] = nullptr;
979  }
980  }
981  real_word->AddSelectedOutlines(wanted, wanted_blobs, wanted_outlines, nullptr);
982  AssignDiacriticsToNewBlobs(outlines, pass, real_word, pr_it, &word_wanted,
983  &target_blobs);
984  int non_overlapped = 0;
985  int non_overlapped_used = 0;
986  for (int i = 0; i < word_wanted.size(); ++i) {
987  if (word_wanted[i]) ++non_overlapped_used;
988  if (outlines[i] != nullptr) ++non_overlapped_used;
989  }
990  if (debug_noise_removal) {
991  tprintf("Used %d/%d overlapped %d/%d non-overlaped diacritics on word:",
992  num_overlapped_used, num_overlapped, non_overlapped_used,
993  non_overlapped);
994  real_word->bounding_box().print();
995  }
996  // Now we have decided which outlines we want, put them into the real_word.
997  if (real_word->AddSelectedOutlines(word_wanted, target_blobs, outlines,
998  make_next_word_fuzzy)) {
999  pr_it->MakeCurrentWordFuzzy();
1000  }
1001  // TODO(rays) Parts of combos have a deep copy of the real word, and need
1002  // to have their noise outlines moved/assigned in the same way!!
1003  return num_overlapped_used != 0 || non_overlapped_used != 0;
1004 }
1005 
1006 // Attempts to put noise/diacritic outlines into the blobs that they overlap.
1007 // Input: a set of noisy outlines that probably belong to the real_word.
1008 // Output: word_wanted indicates which outlines are to be assigned to a blob,
1009 // target_blobs indicates which to assign to, and overlapped_any_blob is
1010 // true for all outlines that overlapped a blob.
1012  const GenericVector<C_OUTLINE*>& outlines, int pass, WERD* real_word,
1013  PAGE_RES_IT* pr_it, GenericVector<bool>* word_wanted,
1014  GenericVector<bool>* overlapped_any_blob,
1015  GenericVector<C_BLOB*>* target_blobs) {
1016  GenericVector<bool> blob_wanted;
1017  word_wanted->init_to_size(outlines.size(), false);
1018  overlapped_any_blob->init_to_size(outlines.size(), false);
1019  target_blobs->init_to_size(outlines.size(), nullptr);
1020  // For each real blob, find the outlines that seriously overlap it.
1021  // A single blob could be several merged characters, so there can be quite
1022  // a few outlines overlapping, and the full engine needs to be used to chop
1023  // and join to get a sensible result.
1024  C_BLOB_IT blob_it(real_word->cblob_list());
1025  for (blob_it.mark_cycle_pt(); !blob_it.cycled_list(); blob_it.forward()) {
1026  C_BLOB* blob = blob_it.data();
1027  const TBOX blob_box = blob->bounding_box();
1028  blob_wanted.init_to_size(outlines.size(), false);
1029  int num_blob_outlines = 0;
1030  for (int i = 0; i < outlines.size(); ++i) {
1031  if (blob_box.major_x_overlap(outlines[i]->bounding_box()) &&
1032  !(*word_wanted)[i]) {
1033  blob_wanted[i] = true;
1034  (*overlapped_any_blob)[i] = true;
1035  ++num_blob_outlines;
1036  }
1037  }
1038  if (debug_noise_removal) {
1039  tprintf("%d noise outlines overlap blob at:", num_blob_outlines);
1040  blob_box.print();
1041  }
1042  // If any outlines overlap the blob, and not too many, classify the blob
1043  // (using the full engine, languages and all), and choose the maximal
1044  // combination of outlines that doesn't hurt the end-result classification
1045  // by too much. Mark them as wanted.
1046  if (0 < num_blob_outlines && num_blob_outlines < noise_maxperblob) {
1047  if (SelectGoodDiacriticOutlines(pass, noise_cert_basechar, pr_it, blob,
1048  outlines, num_blob_outlines,
1049  &blob_wanted)) {
1050  for (int i = 0; i < blob_wanted.size(); ++i) {
1051  if (blob_wanted[i]) {
1052  // Claim the outline and record where it is going.
1053  (*word_wanted)[i] = true;
1054  (*target_blobs)[i] = blob;
1055  }
1056  }
1057  }
1058  }
1059  }
1060 }
1061 
1062 // Attempts to assign non-overlapping outlines to their nearest blobs or
1063 // make new blobs out of them.
1065  const GenericVector<C_OUTLINE*>& outlines, int pass, WERD* real_word,
1066  PAGE_RES_IT* pr_it, GenericVector<bool>* word_wanted,
1067  GenericVector<C_BLOB*>* target_blobs) {
1068  GenericVector<bool> blob_wanted;
1069  word_wanted->init_to_size(outlines.size(), false);
1070  target_blobs->init_to_size(outlines.size(), nullptr);
1071  // Check for outlines that need to be turned into stand-alone blobs.
1072  for (int i = 0; i < outlines.size(); ++i) {
1073  if (outlines[i] == nullptr) continue;
1074  // Get a set of adjacent outlines that don't overlap any existing blob.
1075  blob_wanted.init_to_size(outlines.size(), false);
1076  int num_blob_outlines = 0;
1077  TBOX total_ol_box(outlines[i]->bounding_box());
1078  while (i < outlines.size() && outlines[i] != nullptr) {
1079  blob_wanted[i] = true;
1080  total_ol_box += outlines[i]->bounding_box();
1081  ++i;
1082  ++num_blob_outlines;
1083  }
1084  // Find the insertion point.
1085  C_BLOB_IT blob_it(real_word->cblob_list());
1086  while (!blob_it.at_last() &&
1087  blob_it.data_relative(1)->bounding_box().left() <=
1088  total_ol_box.left()) {
1089  blob_it.forward();
1090  }
1091  // Choose which combination of them we actually want and where to put
1092  // them.
1093  if (debug_noise_removal)
1094  tprintf("Num blobless outlines = %d\n", num_blob_outlines);
1095  C_BLOB* left_blob = blob_it.data();
1096  TBOX left_box = left_blob->bounding_box();
1097  C_BLOB* right_blob = blob_it.at_last() ? nullptr : blob_it.data_relative(1);
1098  if ((left_box.x_overlap(total_ol_box) || right_blob == nullptr ||
1099  !right_blob->bounding_box().x_overlap(total_ol_box)) &&
1100  SelectGoodDiacriticOutlines(pass, noise_cert_disjoint, pr_it, left_blob,
1101  outlines, num_blob_outlines,
1102  &blob_wanted)) {
1103  if (debug_noise_removal) tprintf("Added to left blob\n");
1104  for (int j = 0; j < blob_wanted.size(); ++j) {
1105  if (blob_wanted[j]) {
1106  (*word_wanted)[j] = true;
1107  (*target_blobs)[j] = left_blob;
1108  }
1109  }
1110  } else if (right_blob != nullptr &&
1111  (!left_box.x_overlap(total_ol_box) ||
1112  right_blob->bounding_box().x_overlap(total_ol_box)) &&
1114  right_blob, outlines,
1115  num_blob_outlines, &blob_wanted)) {
1116  if (debug_noise_removal) tprintf("Added to right blob\n");
1117  for (int j = 0; j < blob_wanted.size(); ++j) {
1118  if (blob_wanted[j]) {
1119  (*word_wanted)[j] = true;
1120  (*target_blobs)[j] = right_blob;
1121  }
1122  }
1123  } else if (SelectGoodDiacriticOutlines(pass, noise_cert_punc, pr_it, nullptr,
1124  outlines, num_blob_outlines,
1125  &blob_wanted)) {
1126  if (debug_noise_removal) tprintf("Fitted between blobs\n");
1127  for (int j = 0; j < blob_wanted.size(); ++j) {
1128  if (blob_wanted[j]) {
1129  (*word_wanted)[j] = true;
1130  (*target_blobs)[j] = nullptr;
1131  }
1132  }
1133  }
1134  }
1135 }
1136 
1137 // Starting with ok_outlines set to indicate which outlines overlap the blob,
1138 // chooses the optimal set (approximately) and returns true if any outlines
1139 // are desired, in which case ok_outlines indicates which ones.
1141  int pass, float certainty_threshold, PAGE_RES_IT* pr_it, C_BLOB* blob,
1142  const GenericVector<C_OUTLINE*>& outlines, int num_outlines,
1143  GenericVector<bool>* ok_outlines) {
1144  STRING best_str;
1145  float target_cert = certainty_threshold;
1146  if (blob != nullptr) {
1147  float target_c2;
1148  target_cert = ClassifyBlobAsWord(pass, pr_it, blob, &best_str, &target_c2);
1149  if (debug_noise_removal) {
1150  tprintf("No Noise blob classified as %s=%g(%g) at:", best_str.string(),
1151  target_cert, target_c2);
1152  blob->bounding_box().print();
1153  }
1154  target_cert -= (target_cert - certainty_threshold) * noise_cert_factor;
1155  }
1156  GenericVector<bool> test_outlines = *ok_outlines;
1157  // Start with all the outlines in.
1158  STRING all_str;
1159  GenericVector<bool> best_outlines = *ok_outlines;
1160  float best_cert = ClassifyBlobPlusOutlines(test_outlines, outlines, pass,
1161  pr_it, blob, &all_str);
1162  if (debug_noise_removal) {
1163  TBOX ol_box;
1164  for (int i = 0; i < test_outlines.size(); ++i) {
1165  if (test_outlines[i]) ol_box += outlines[i]->bounding_box();
1166  }
1167  tprintf("All Noise blob classified as %s=%g, delta=%g at:",
1168  all_str.string(), best_cert, best_cert - target_cert);
1169  ol_box.print();
1170  }
1171  // Iteratively zero out the bit that improves the certainty the most, until
1172  // we get past the threshold, have zero bits, or fail to improve.
1173  int best_index = 0; // To zero out.
1174  while (num_outlines > 1 && best_index >= 0 &&
1175  (blob == nullptr || best_cert < target_cert || blob != nullptr)) {
1176  // Find the best bit to zero out.
1177  best_index = -1;
1178  for (int i = 0; i < outlines.size(); ++i) {
1179  if (test_outlines[i]) {
1180  test_outlines[i] = false;
1181  STRING str;
1182  float cert = ClassifyBlobPlusOutlines(test_outlines, outlines, pass,
1183  pr_it, blob, &str);
1184  if (debug_noise_removal) {
1185  TBOX ol_box;
1186  for (int j = 0; j < outlines.size(); ++j) {
1187  if (test_outlines[j]) ol_box += outlines[j]->bounding_box();
1188  tprintf("%d", test_outlines[j]);
1189  }
1190  tprintf(" blob classified as %s=%g, delta=%g) at:", str.string(),
1191  cert, cert - target_cert);
1192  ol_box.print();
1193  }
1194  if (cert > best_cert) {
1195  best_cert = cert;
1196  best_index = i;
1197  best_outlines = test_outlines;
1198  }
1199  test_outlines[i] = true;
1200  }
1201  }
1202  if (best_index >= 0) {
1203  test_outlines[best_index] = false;
1204  --num_outlines;
1205  }
1206  }
1207  if (best_cert >= target_cert) {
1208  // Save the best combination.
1209  *ok_outlines = best_outlines;
1210  if (debug_noise_removal) {
1211  tprintf("%s noise combination ", blob ? "Adding" : "New");
1212  for (int i = 0; i < best_outlines.size(); ++i) {
1213  tprintf("%d", best_outlines[i]);
1214  }
1215  tprintf(" yields certainty %g, beating target of %g\n", best_cert,
1216  target_cert);
1217  }
1218  return true;
1219  }
1220 
1221  return false;
1222 }
1223 
1224 // Classifies the given blob plus the outlines flagged by ok_outlines, undoes
1225 // the inclusion of the outlines, and returns the certainty of the raw choice.
1227  const GenericVector<bool>& ok_outlines,
1228  const GenericVector<C_OUTLINE*>& outlines, int pass_n, PAGE_RES_IT* pr_it,
1229  C_BLOB* blob, STRING* best_str) {
1230  C_OUTLINE_IT ol_it;
1231  C_OUTLINE* first_to_keep = nullptr;
1232  C_BLOB* local_blob = nullptr;
1233  if (blob != nullptr) {
1234  // Add the required outlines to the blob.
1235  ol_it.set_to_list(blob->out_list());
1236  first_to_keep = ol_it.data();
1237  }
1238  for (int i = 0; i < ok_outlines.size(); ++i) {
1239  if (ok_outlines[i]) {
1240  // This outline is to be added.
1241  if (blob == nullptr) {
1242  local_blob = new C_BLOB(outlines[i]);
1243  blob = local_blob;
1244  ol_it.set_to_list(blob->out_list());
1245  } else {
1246  ol_it.add_before_stay_put(outlines[i]);
1247  }
1248  }
1249  }
1250  float c2;
1251  float cert = ClassifyBlobAsWord(pass_n, pr_it, blob, best_str, &c2);
1252  ol_it.move_to_first();
1253  if (first_to_keep == nullptr) {
1254  // We created blob. Empty its outlines and delete it.
1255  for (; !ol_it.empty(); ol_it.forward()) ol_it.extract();
1256  delete local_blob;
1257  cert = -c2;
1258  } else {
1259  // Remove the outlines that we put in.
1260  for (; ol_it.data() != first_to_keep; ol_it.forward()) {
1261  ol_it.extract();
1262  }
1263  }
1264  return cert;
1265 }
1266 
1267 // Classifies the given blob (part of word_data->word->word) as an individual
1268 // word, using languages, chopper etc, returning only the certainty of the
1269 // best raw choice, and undoing all the work done to fake out the word.
1271  C_BLOB* blob, STRING* best_str, float* c2) {
1272  WERD* real_word = pr_it->word()->word;
1273  WERD* word = real_word->ConstructFromSingleBlob(
1274  real_word->flag(W_BOL), real_word->flag(W_EOL), C_BLOB::deep_copy(blob));
1275  WERD_RES* word_res = pr_it->InsertSimpleCloneWord(*pr_it->word(), word);
1276  // Get a new iterator that points to the new word.
1277  PAGE_RES_IT it(pr_it->page_res);
1278  while (it.word() != word_res && it.word() != nullptr) it.forward();
1279  ASSERT_HOST(it.word() == word_res);
1280  WordData wd(it);
1281  // Force full initialization.
1282  SetupWordPassN(1, &wd);
1283  classify_word_and_language(pass_n, &it, &wd);
1284  if (debug_noise_removal) {
1285  if (wd.word->raw_choice != nullptr) {
1286  tprintf("word xheight=%g, row=%g, range=[%g,%g]\n", word_res->x_height,
1287  wd.row->x_height(), wd.word->raw_choice->min_x_height(),
1288  wd.word->raw_choice->max_x_height());
1289  } else {
1290  tprintf("Got word with null raw choice xheight=%g, row=%g\n", word_res->x_height,
1291  wd.row->x_height());
1292  }
1293  }
1294  float cert = 0.0f;
1295  if (wd.word->raw_choice != nullptr) { // This probably shouldn't happen, but...
1296  cert = wd.word->raw_choice->certainty();
1297  float rat = wd.word->raw_choice->rating();
1298  *c2 = rat > 0.0f ? cert * cert / rat : 0.0f;
1299  *best_str = wd.word->raw_choice->unichar_string();
1300  } else {
1301  *c2 = 0.0f;
1302  *best_str = "";
1303  }
1304  it.DeleteCurrentWord();
1305  pr_it->ResetWordIterator();
1306  return cert;
1307 }
1308 
1309 #endif // ndef DISABLED_LEGACY_ENGINE
1310 
1311 // Generic function for classifying a word. Can be used either for pass1 or
1312 // pass2 according to the function passed to recognizer.
1313 // word_data holds the word to be recognized, and its block and row, and
1314 // pr_it points to the word as well, in case we are running LSTM and it wants
1315 // to output multiple words.
1316 // Recognizes in the current language, and if successful that is all.
1317 // If recognition was not successful, tries all available languages until
1318 // it gets a successful result or runs out of languages. Keeps the best result.
1320  WordData* word_data) {
1321 #ifdef DISABLED_LEGACY_ENGINE
1323 #else
1324  WordRecognizer recognizer = pass_n == 1 ? &Tesseract::classify_word_pass1
1326 #endif // def DISABLED_LEGACY_ENGINE
1327 
1328  // Best result so far.
1329  PointerVector<WERD_RES> best_words;
1330  // Points to the best result. May be word or in lang_words.
1331  const WERD_RES* word = word_data->word;
1332  clock_t start_t = clock();
1333  const bool debug = classify_debug_level > 0 || multilang_debug_level > 0;
1334  if (debug) {
1335  tprintf("%s word with lang %s at:",
1336  word->done ? "Already done" : "Processing",
1337  most_recently_used_->lang.string());
1338  word->word->bounding_box().print();
1339  }
1340  if (word->done) {
1341  // If done on pass1, leave it as-is.
1342  if (!word->tess_failed)
1343  most_recently_used_ = word->tesseract;
1344  return;
1345  }
1346  int sub = sub_langs_.size();
1347  if (most_recently_used_ != this) {
1348  // Get the index of the most_recently_used_.
1349  for (sub = 0; sub < sub_langs_.size() &&
1350  most_recently_used_ != sub_langs_[sub]; ++sub) {}
1351  }
1352  most_recently_used_->RetryWithLanguage(
1353  *word_data, recognizer, debug, &word_data->lang_words[sub], &best_words);
1354  Tesseract* best_lang_tess = most_recently_used_;
1355  if (!WordsAcceptable(best_words)) {
1356  // Try all the other languages to see if they are any better.
1357  if (most_recently_used_ != this &&
1358  this->RetryWithLanguage(*word_data, recognizer, debug,
1359  &word_data->lang_words[sub_langs_.size()],
1360  &best_words) > 0) {
1361  best_lang_tess = this;
1362  }
1363  for (int i = 0; !WordsAcceptable(best_words) && i < sub_langs_.size();
1364  ++i) {
1365  if (most_recently_used_ != sub_langs_[i] &&
1366  sub_langs_[i]->RetryWithLanguage(*word_data, recognizer, debug,
1367  &word_data->lang_words[i],
1368  &best_words) > 0) {
1369  best_lang_tess = sub_langs_[i];
1370  }
1371  }
1372  }
1373  most_recently_used_ = best_lang_tess;
1374  if (!best_words.empty()) {
1375  if (best_words.size() == 1 && !best_words[0]->combination) {
1376  // Move the best single result to the main word.
1377  word_data->word->ConsumeWordResults(best_words[0]);
1378  } else {
1379  // Words came from LSTM, and must be moved to the PAGE_RES properly.
1380  word_data->word = best_words.back();
1381  pr_it->ReplaceCurrentWord(&best_words);
1382  }
1383  ASSERT_HOST(word_data->word->box_word != nullptr);
1384  } else {
1385  tprintf("no best words!!\n");
1386  }
1387  clock_t ocr_t = clock();
1388  if (tessedit_timing_debug) {
1389  tprintf("%s (ocr took %.2f sec)\n",
1390  word_data->word->best_choice->unichar_string().string(),
1391  static_cast<double>(ocr_t-start_t)/CLOCKS_PER_SEC);
1392  }
1393 }
1394 
1402  WERD_RES** in_word,
1403  PointerVector<WERD_RES>* out_words) {
1404  ROW* row = word_data.row;
1405  BLOCK* block = word_data.block;
1406  prev_word_best_choice_ = word_data.prev_word != nullptr
1407  ? word_data.prev_word->word->best_choice : nullptr;
1408 #ifndef ANDROID_BUILD
1409 #ifdef DISABLED_LEGACY_ENGINE
1411 #else
1414 #endif // def DISABLED_LEGACY_ENGINE
1415  if (!(*in_word)->odd_size || tessedit_ocr_engine_mode == OEM_LSTM_ONLY) {
1416  LSTMRecognizeWord(*block, row, *in_word, out_words);
1417  if (!out_words->empty())
1418  return; // Successful lstm recognition.
1419  }
1421  // No fallback allowed, so use a fake.
1422  (*in_word)->SetupFake(lstm_recognizer_->GetUnicharset());
1423  return;
1424  }
1425 
1426  #ifndef DISABLED_LEGACY_ENGINE
1427  // Fall back to tesseract for failed words or odd words.
1428  (*in_word)->SetupForRecognition(unicharset, this, BestPix(),
1429  OEM_TESSERACT_ONLY, nullptr,
1432  poly_allow_detailed_fx, row, block);
1433 #endif // ndef DISABLED_LEGACY_ENGINE
1434  }
1435 #endif // ndef ANDROID_BUILD
1436 
1437 #ifndef DISABLED_LEGACY_ENGINE
1438  WERD_RES* word = *in_word;
1439  match_word_pass_n(1, word, row, block);
1440  if (!word->tess_failed && !word->word->flag(W_REP_CHAR)) {
1441  word->tess_would_adapt = AdaptableWord(word);
1442  bool adapt_ok = word_adaptable(word, tessedit_tess_adaption_mode);
1443 
1444  if (adapt_ok) {
1445  // Send word to adaptive classifier for training.
1446  word->BestChoiceToCorrectText();
1447  LearnWord(nullptr, word);
1448  // Mark misadaptions if running blamer.
1449  if (word->blamer_bundle != nullptr) {
1452  }
1453  }
1454 
1455  if (tessedit_enable_doc_dict && !word->IsAmbiguous())
1457  }
1458 #endif // ndef DISABLED_LEGACY_ENGINE
1459 }
1460 
1461 // Helper to report the result of the xheight fix.
1462 void Tesseract::ReportXhtFixResult(bool accept_new_word, float new_x_ht,
1463  WERD_RES* word, WERD_RES* new_word) {
1464  tprintf("New XHT Match:%s = %s ",
1465  word->best_choice->unichar_string().string(),
1466  word->best_choice->debug_string().string());
1467  word->reject_map.print(debug_fp);
1468  tprintf(" -> %s = %s ",
1469  new_word->best_choice->unichar_string().string(),
1470  new_word->best_choice->debug_string().string());
1471  new_word->reject_map.print(debug_fp);
1472  tprintf(" %s->%s %s %s\n",
1473  word->guessed_x_ht ? "GUESS" : "CERT",
1474  new_word->guessed_x_ht ? "GUESS" : "CERT",
1475  new_x_ht > 0.1 ? "STILL DOUBT" : "OK",
1476  accept_new_word ? "ACCEPTED" : "");
1477 }
1478 
1479 #ifndef DISABLED_LEGACY_ENGINE
1480 
1481 // Run the x-height fix-up, based on min/max top/bottom information in
1482 // unicharset.
1483 // Returns true if the word was changed.
1484 // See the comment in fixxht.cpp for a description of the overall process.
1485 bool Tesseract::TrainedXheightFix(WERD_RES *word, BLOCK* block, ROW *row) {
1486  int original_misfits = CountMisfitTops(word);
1487  if (original_misfits == 0)
1488  return false;
1489  float baseline_shift = 0.0f;
1490  float new_x_ht = ComputeCompatibleXheight(word, &baseline_shift);
1491  if (baseline_shift != 0.0f) {
1492  // Try the shift on its own first.
1493  if (!TestNewNormalization(original_misfits, baseline_shift, word->x_height,
1494  word, block, row))
1495  return false;
1496  original_misfits = CountMisfitTops(word);
1497  if (original_misfits > 0) {
1498  float new_baseline_shift;
1499  // Now recompute the new x_height.
1500  new_x_ht = ComputeCompatibleXheight(word, &new_baseline_shift);
1501  if (new_x_ht >= kMinRefitXHeightFraction * word->x_height) {
1502  // No test of return value here, as we are definitely making a change
1503  // to the word by shifting the baseline.
1504  TestNewNormalization(original_misfits, baseline_shift, new_x_ht,
1505  word, block, row);
1506  }
1507  }
1508  return true;
1509  } else if (new_x_ht >= kMinRefitXHeightFraction * word->x_height) {
1510  return TestNewNormalization(original_misfits, 0.0f, new_x_ht,
1511  word, block, row);
1512  } else {
1513  return false;
1514  }
1515 }
1516 
1517 // Runs recognition with the test baseline shift and x-height and returns true
1518 // if there was an improvement in recognition result.
1519 bool Tesseract::TestNewNormalization(int original_misfits,
1520  float baseline_shift, float new_x_ht,
1521  WERD_RES *word, BLOCK* block, ROW *row) {
1522  bool accept_new_x_ht = false;
1523  WERD_RES new_x_ht_word(word->word);
1524  if (word->blamer_bundle != nullptr) {
1525  new_x_ht_word.blamer_bundle = new BlamerBundle();
1526  new_x_ht_word.blamer_bundle->CopyTruth(*(word->blamer_bundle));
1527  }
1528  new_x_ht_word.x_height = new_x_ht;
1529  new_x_ht_word.baseline_shift = baseline_shift;
1530  new_x_ht_word.caps_height = 0.0;
1531  new_x_ht_word.SetupForRecognition(
1532  unicharset, this, BestPix(), tessedit_ocr_engine_mode, nullptr,
1534  poly_allow_detailed_fx, row, block);
1535  match_word_pass_n(2, &new_x_ht_word, row, block);
1536  if (!new_x_ht_word.tess_failed) {
1537  int new_misfits = CountMisfitTops(&new_x_ht_word);
1538  if (debug_x_ht_level >= 1) {
1539  tprintf("Old misfits=%d with x-height %f, new=%d with x-height %f\n",
1540  original_misfits, word->x_height,
1541  new_misfits, new_x_ht);
1542  tprintf("Old rating= %f, certainty=%f, new=%f, %f\n",
1543  word->best_choice->rating(), word->best_choice->certainty(),
1544  new_x_ht_word.best_choice->rating(),
1545  new_x_ht_word.best_choice->certainty());
1546  }
1547  // The misfits must improve and either the rating or certainty.
1548  accept_new_x_ht = new_misfits < original_misfits &&
1549  (new_x_ht_word.best_choice->certainty() >
1550  word->best_choice->certainty() ||
1551  new_x_ht_word.best_choice->rating() <
1552  word->best_choice->rating());
1553  if (debug_x_ht_level >= 1) {
1554  ReportXhtFixResult(accept_new_x_ht, new_x_ht, word, &new_x_ht_word);
1555  }
1556  }
1557  if (accept_new_x_ht) {
1558  word->ConsumeWordResults(&new_x_ht_word);
1559  return true;
1560  }
1561  return false;
1562 }
1563 
1564 #endif // ndef DISABLED_LEGACY_ENGINE
1565 
1573  WERD_RES** in_word,
1574  PointerVector<WERD_RES>* out_words) {
1575  // Return if we do not want to run Tesseract.
1577  return;
1578  }
1579 #ifndef DISABLED_LEGACY_ENGINE
1580  ROW* row = word_data.row;
1581  BLOCK* block = word_data.block;
1582  WERD_RES* word = *in_word;
1583  prev_word_best_choice_ = word_data.prev_word != nullptr
1584  ? word_data.prev_word->word->best_choice : nullptr;
1585 
1587  check_debug_pt(word, 30);
1588  if (!word->done) {
1589  word->caps_height = 0.0;
1590  if (word->x_height == 0.0f)
1591  word->x_height = row->x_height();
1592  match_word_pass_n(2, word, row, block);
1593  check_debug_pt(word, 40);
1594  }
1595 
1596  SubAndSuperscriptFix(word);
1597 
1598  if (!word->tess_failed && !word->word->flag(W_REP_CHAR)) {
1600  block->classify_rotation().y() == 0.0f) {
1601  // Use the tops and bottoms since they are available.
1602  TrainedXheightFix(word, block, row);
1603  }
1604 
1606  }
1607 #ifndef GRAPHICS_DISABLED
1609  if (fx_win == nullptr)
1610  create_fx_win();
1611  clear_fx_win();
1612  word->rebuild_word->plot(fx_win);
1613  TBOX wbox = word->rebuild_word->bounding_box();
1614  fx_win->ZoomToRectangle(wbox.left(), wbox.top(),
1615  wbox.right(), wbox.bottom());
1617  }
1618 #endif
1620  check_debug_pt(word, 50);
1621 #endif // ndef DISABLED_LEGACY_ENGINE
1622 }
1623 
1624 #ifndef DISABLED_LEGACY_ENGINE
1631  ROW *row, BLOCK* block) {
1632  if (word->tess_failed) return;
1633  tess_segment_pass_n(pass_n, word);
1634 
1635  if (!word->tess_failed) {
1636  if (!word->word->flag (W_REP_CHAR)) {
1637  word->fix_quotes();
1639  word->fix_hyphens();
1640  /* Don't trust fix_quotes! - though I think I've fixed the bug */
1641  if (word->best_choice->length() != word->box_word->length()) {
1642  tprintf("POST FIX_QUOTES FAIL String:\"%s\"; Strlen=%d;"
1643  " #Blobs=%d\n",
1644  word->best_choice->debug_string().string(),
1645  word->best_choice->length(),
1646  word->box_word->length());
1647 
1648  }
1649  word->tess_accepted = tess_acceptable_word(word);
1650 
1651  // Also sets word->done flag
1652  make_reject_map(word, row, pass_n);
1653  }
1654  }
1655  set_word_fonts(word);
1656 
1657  ASSERT_HOST(word->raw_choice != nullptr);
1658 }
1659 #endif // ndef DISABLED_LEGACY_ENGINE
1660 
1661 // Helper to return the best rated BLOB_CHOICE in the whole word that matches
1662 // the given char_id, or nullptr if none can be found.
1663 static BLOB_CHOICE* FindBestMatchingChoice(UNICHAR_ID char_id,
1664  WERD_RES* word_res) {
1665  // Find the corresponding best BLOB_CHOICE from any position in the word_res.
1666  BLOB_CHOICE* best_choice = nullptr;
1667  for (int i = 0; i < word_res->best_choice->length(); ++i) {
1668  BLOB_CHOICE* choice = FindMatchingChoice(char_id,
1669  word_res->GetBlobChoices(i));
1670  if (choice != nullptr) {
1671  if (best_choice == nullptr || choice->rating() < best_choice->rating())
1672  best_choice = choice;
1673  }
1674  }
1675  return best_choice;
1676 }
1677 
1678 // Helper to insert blob_choice in each location in the leader word if there is
1679 // no matching BLOB_CHOICE there already, and correct any incorrect results
1680 // in the best_choice.
1681 static void CorrectRepcharChoices(BLOB_CHOICE* blob_choice,
1682  WERD_RES* word_res) {
1683  WERD_CHOICE* word = word_res->best_choice;
1684  for (int i = 0; i < word_res->best_choice->length(); ++i) {
1685  BLOB_CHOICE* choice = FindMatchingChoice(blob_choice->unichar_id(),
1686  word_res->GetBlobChoices(i));
1687  if (choice == nullptr) {
1688  BLOB_CHOICE_IT choice_it(word_res->GetBlobChoices(i));
1689  choice_it.add_before_stay_put(new BLOB_CHOICE(*blob_choice));
1690  }
1691  }
1692  // Correct any incorrect results in word.
1693  for (int i = 0; i < word->length(); ++i) {
1694  if (word->unichar_id(i) != blob_choice->unichar_id())
1695  word->set_unichar_id(blob_choice->unichar_id(), i);
1696  }
1697 }
1698 
1707  WERD_RES *word_res = page_res_it->word();
1708  const WERD_CHOICE &word = *(word_res->best_choice);
1709 
1710  // Find the frequency of each unique character in the word.
1711  SortHelper<UNICHAR_ID> rep_ch(word.length());
1712  for (int i = 0; i < word.length(); ++i) {
1713  rep_ch.Add(word.unichar_id(i), 1);
1714  }
1715 
1716  // Find the most frequent result.
1717  UNICHAR_ID maxch_id = INVALID_UNICHAR_ID; // most common char
1718  int max_count = rep_ch.MaxCount(&maxch_id);
1719  // Find the best exemplar of a classifier result for maxch_id.
1720  BLOB_CHOICE* best_choice = FindBestMatchingChoice(maxch_id, word_res);
1721  if (best_choice == nullptr) {
1722  tprintf("Failed to find a choice for %s, occurring %d times\n",
1723  word_res->uch_set->debug_str(maxch_id).string(), max_count);
1724  return;
1725  }
1726  word_res->done = true;
1727 
1728  // Measure the mean space.
1729  int gap_count = 0;
1730  WERD* werd = word_res->word;
1731  C_BLOB_IT blob_it(werd->cblob_list());
1732  C_BLOB* prev_blob = blob_it.data();
1733  for (blob_it.forward(); !blob_it.at_first(); blob_it.forward()) {
1734  C_BLOB* blob = blob_it.data();
1735  int gap = blob->bounding_box().left();
1736  gap -= prev_blob->bounding_box().right();
1737  ++gap_count;
1738  prev_blob = blob;
1739  }
1740  // Just correct existing classification.
1741  CorrectRepcharChoices(best_choice, word_res);
1742  word_res->reject_map.initialise(word.length());
1743 }
1744 
1746  const UNICHARSET& char_set, const char *s, const char *lengths) {
1747  int i = 0;
1748  int offset = 0;
1749  int leading_punct_count;
1750  int upper_count = 0;
1751  int hyphen_pos = -1;
1753 
1754  if (strlen (lengths) > 20)
1755  return word_type;
1756 
1757  /* Single Leading punctuation char*/
1758 
1759  if (s[offset] != '\0' && STRING(chs_leading_punct).contains(s[offset]))
1760  offset += lengths[i++];
1761  leading_punct_count = i;
1762 
1763  /* Initial cap */
1764  while (s[offset] != '\0' && char_set.get_isupper(s + offset, lengths[i])) {
1765  offset += lengths[i++];
1766  upper_count++;
1767  }
1768  if (upper_count > 1) {
1769  word_type = AC_UPPER_CASE;
1770  } else {
1771  /* Lower case word, possibly with an initial cap */
1772  while (s[offset] != '\0' && char_set.get_islower(s + offset, lengths[i])) {
1773  offset += lengths[i++];
1774  }
1775  if (i - leading_punct_count < quality_min_initial_alphas_reqd)
1776  goto not_a_word;
1777  /*
1778  Allow a single hyphen in a lower case word
1779  - don't trust upper case - I've seen several cases of "H" -> "I-I"
1780  */
1781  if (lengths[i] == 1 && s[offset] == '-') {
1782  hyphen_pos = i;
1783  offset += lengths[i++];
1784  if (s[offset] != '\0') {
1785  while ((s[offset] != '\0') &&
1786  char_set.get_islower(s + offset, lengths[i])) {
1787  offset += lengths[i++];
1788  }
1789  if (i < hyphen_pos + 3)
1790  goto not_a_word;
1791  }
1792  } else {
1793  /* Allow "'s" in NON hyphenated lower case words */
1794  if (lengths[i] == 1 && (s[offset] == '\'') &&
1795  lengths[i + 1] == 1 && (s[offset + lengths[i]] == 's')) {
1796  offset += lengths[i++];
1797  offset += lengths[i++];
1798  }
1799  }
1800  if (upper_count > 0)
1801  word_type = AC_INITIAL_CAP;
1802  else
1803  word_type = AC_LOWER_CASE;
1804  }
1805 
1806  /* Up to two different, constrained trailing punctuation chars */
1807  if (lengths[i] == 1 && s[offset] != '\0' &&
1808  STRING(chs_trailing_punct1).contains(s[offset]))
1809  offset += lengths[i++];
1810  if (lengths[i] == 1 && s[offset] != '\0' && i > 0 &&
1811  s[offset - lengths[i - 1]] != s[offset] &&
1812  STRING(chs_trailing_punct2).contains (s[offset]))
1813  offset += lengths[i++];
1814 
1815  if (s[offset] != '\0')
1816  word_type = AC_UNACCEPTABLE;
1817 
1818  not_a_word:
1819 
1820  if (word_type == AC_UNACCEPTABLE) {
1821  /* Look for abbreviation string */
1822  i = 0;
1823  offset = 0;
1824  if (s[0] != '\0' && char_set.get_isupper(s, lengths[0])) {
1825  word_type = AC_UC_ABBREV;
1826  while (s[offset] != '\0' &&
1827  char_set.get_isupper(s + offset, lengths[i]) &&
1828  lengths[i + 1] == 1 && s[offset + lengths[i]] == '.') {
1829  offset += lengths[i++];
1830  offset += lengths[i++];
1831  }
1832  }
1833  else if (s[0] != '\0' && char_set.get_islower(s, lengths[0])) {
1834  word_type = AC_LC_ABBREV;
1835  while (s[offset] != '\0' &&
1836  char_set.get_islower(s + offset, lengths[i]) &&
1837  lengths[i + 1] == 1 && s[offset + lengths[i]] == '.') {
1838  offset += lengths[i++];
1839  offset += lengths[i++];
1840  }
1841  }
1842  if (s[offset] != '\0')
1843  word_type = AC_UNACCEPTABLE;
1844  }
1845 
1846  return word_type;
1847 }
1848 
1849 bool Tesseract::check_debug_pt(WERD_RES* word, int location) {
1850  bool show_map_detail = false;
1851  int16_t i;
1852 
1853  if (!test_pt)
1854  return false;
1855 
1856  tessedit_rejection_debug.set_value (false);
1857  debug_x_ht_level.set_value(0);
1858 
1859  if (word->word->bounding_box().contains(FCOORD (test_pt_x, test_pt_y))) {
1860  if (location < 0)
1861  return true; // For breakpoint use
1862  tessedit_rejection_debug.set_value(true);
1863  debug_x_ht_level.set_value(2);
1864  tprintf ("\n\nTESTWD::");
1865  switch (location) {
1866  case 0:
1867  tprintf ("classify_word_pass1 start\n");
1868  word->word->print();
1869  break;
1870  case 10:
1871  tprintf ("make_reject_map: initial map");
1872  break;
1873  case 20:
1874  tprintf ("make_reject_map: after NN");
1875  break;
1876  case 30:
1877  tprintf ("classify_word_pass2 - START");
1878  break;
1879  case 40:
1880  tprintf ("classify_word_pass2 - Pre Xht");
1881  break;
1882  case 50:
1883  tprintf ("classify_word_pass2 - END");
1884  show_map_detail = true;
1885  break;
1886  case 60:
1887  tprintf ("fixspace");
1888  break;
1889  case 70:
1890  tprintf ("MM pass START");
1891  break;
1892  case 80:
1893  tprintf ("MM pass END");
1894  break;
1895  case 90:
1896  tprintf ("After Poor quality rejection");
1897  break;
1898  case 100:
1899  tprintf ("unrej_good_quality_words - START");
1900  break;
1901  case 110:
1902  tprintf ("unrej_good_quality_words - END");
1903  break;
1904  case 120:
1905  tprintf ("Write results pass");
1906  show_map_detail = true;
1907  break;
1908  }
1909  if (word->best_choice != nullptr) {
1910  tprintf(" \"%s\" ", word->best_choice->unichar_string().string());
1911  word->reject_map.print(debug_fp);
1912  tprintf("\n");
1913  if (show_map_detail) {
1914  tprintf("\"%s\"\n", word->best_choice->unichar_string().string());
1915  for (i = 0; word->best_choice->unichar_string()[i] != '\0'; i++) {
1916  tprintf("**** \"%c\" ****\n", word->best_choice->unichar_string()[i]);
1917  word->reject_map[i].full_print(debug_fp);
1918  }
1919  }
1920  } else {
1921  tprintf("null best choice\n");
1922  }
1923  tprintf ("Tess Accepted: %s\n", word->tess_accepted ? "TRUE" : "FALSE");
1924  tprintf ("Done flag: %s\n\n", word->done ? "TRUE" : "FALSE");
1925  return true;
1926  } else {
1927  return false;
1928  }
1929 }
1930 
1936 static void find_modal_font( // good chars in word
1937  STATS* fonts, // font stats
1938  int16_t* font_out, // output font
1939  int8_t* font_count // output count
1940 ) {
1941  int16_t font; //font index
1942  int32_t count; //pile count
1943 
1944  if (fonts->get_total () > 0) {
1945  font = static_cast<int16_t>(fonts->mode ());
1946  *font_out = font;
1947  count = fonts->pile_count (font);
1948  *font_count = count < INT8_MAX ? count : INT8_MAX;
1949  fonts->add (font, -*font_count);
1950  }
1951  else {
1952  *font_out = -1;
1953  *font_count = 0;
1954  }
1955 }
1956 
1963  // Don't try to set the word fonts for an lstm word, as the configs
1964  // will be meaningless.
1965  if (word->chopped_word == nullptr) return;
1966  ASSERT_HOST(word->best_choice != nullptr);
1967 
1968 #ifndef DISABLED_LEGACY_ENGINE
1969  const int fontinfo_size = get_fontinfo_table().size();
1970  if (fontinfo_size == 0) return;
1971  GenericVector<int> font_total_score;
1972  font_total_score.init_to_size(fontinfo_size, 0);
1973 
1974  // Compute the font scores for the word
1975  if (tessedit_debug_fonts) {
1976  tprintf("Examining fonts in %s\n",
1977  word->best_choice->debug_string().string());
1978  }
1979  for (int b = 0; b < word->best_choice->length(); ++b) {
1980  const BLOB_CHOICE* choice = word->GetBlobChoice(b);
1981  if (choice == nullptr) continue;
1982  const GenericVector<ScoredFont>& fonts = choice->fonts();
1983  for (int f = 0; f < fonts.size(); ++f) {
1984  const int fontinfo_id = fonts[f].fontinfo_id;
1985  if (0 <= fontinfo_id && fontinfo_id < fontinfo_size) {
1986  font_total_score[fontinfo_id] += fonts[f].score;
1987  }
1988  }
1989  }
1990  // Find the top and 2nd choice for the word.
1991  int score1 = 0, score2 = 0;
1992  int16_t font_id1 = -1, font_id2 = -1;
1993  for (int f = 0; f < fontinfo_size; ++f) {
1994  if (tessedit_debug_fonts && font_total_score[f] > 0) {
1995  tprintf("Font %s, total score = %d\n",
1996  fontinfo_table_.get(f).name, font_total_score[f]);
1997  }
1998  if (font_total_score[f] > score1) {
1999  score2 = score1;
2000  font_id2 = font_id1;
2001  score1 = font_total_score[f];
2002  font_id1 = f;
2003  } else if (font_total_score[f] > score2) {
2004  score2 = font_total_score[f];
2005  font_id2 = f;
2006  }
2007  }
2008  word->fontinfo = font_id1 >= 0 ? &fontinfo_table_.get(font_id1) : nullptr;
2009  word->fontinfo2 = font_id2 >= 0 ? &fontinfo_table_.get(font_id2) : nullptr;
2010  // Each score has a limit of UINT16_MAX, so divide by that to get the number
2011  // of "votes" for that font, ie number of perfect scores.
2012  word->fontinfo_id_count = ClipToRange<int>(score1 / UINT16_MAX, 1, INT8_MAX);
2013  word->fontinfo_id2_count = ClipToRange<int>(score2 / UINT16_MAX, 0, INT8_MAX);
2014  if (score1 > 0) {
2015  const FontInfo fi = fontinfo_table_.get(font_id1);
2016  if (tessedit_debug_fonts) {
2017  if (word->fontinfo_id2_count > 0 && font_id2 >= 0) {
2018  tprintf("Word modal font=%s, score=%d, 2nd choice %s/%d\n",
2019  fi.name, word->fontinfo_id_count,
2020  fontinfo_table_.get(font_id2).name,
2021  word->fontinfo_id2_count);
2022  } else {
2023  tprintf("Word modal font=%s, score=%d. No 2nd choice\n",
2024  fi.name, word->fontinfo_id_count);
2025  }
2026  }
2027  }
2028 #endif // ndef DISABLED_LEGACY_ENGINE
2029 }
2030 
2031 #ifndef DISABLED_LEGACY_ENGINE
2038  PAGE_RES_IT page_res_it(page_res);
2039  WERD_RES *word; // current word
2040  STATS doc_fonts(0, font_table_size_); // font counters
2041 
2042  // Gather font id statistics.
2043  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
2044  page_res_it.forward()) {
2045  word = page_res_it.word();
2046  if (word->fontinfo != nullptr) {
2047  doc_fonts.add(word->fontinfo->universal_id, word->fontinfo_id_count);
2048  }
2049  if (word->fontinfo2 != nullptr) {
2050  doc_fonts.add(word->fontinfo2->universal_id, word->fontinfo_id2_count);
2051  }
2052  }
2053  int16_t doc_font; // modal font
2054  int8_t doc_font_count; // modal font
2055  find_modal_font(&doc_fonts, &doc_font, &doc_font_count);
2056  if (doc_font_count == 0)
2057  return;
2058  // Get the modal font pointer.
2059  const FontInfo* modal_font = nullptr;
2060  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
2061  page_res_it.forward()) {
2062  word = page_res_it.word();
2063  if (word->fontinfo != nullptr && word->fontinfo->universal_id == doc_font) {
2064  modal_font = word->fontinfo;
2065  break;
2066  }
2067  if (word->fontinfo2 != nullptr && word->fontinfo2->universal_id == doc_font) {
2068  modal_font = word->fontinfo2;
2069  break;
2070  }
2071  }
2072  ASSERT_HOST(modal_font != nullptr);
2073 
2074  // Assign modal font to weak words.
2075  for (page_res_it.restart_page(); page_res_it.word() != nullptr;
2076  page_res_it.forward()) {
2077  word = page_res_it.word();
2078  const int length = word->best_choice->length();
2079 
2080  const int count = word->fontinfo_id_count;
2081  if (!(count == length || (length > 3 && count >= length * 3 / 4))) {
2082  word->fontinfo = modal_font;
2083  // Counts only get 1 as it came from the doc.
2084  word->fontinfo_id_count = 1;
2085  }
2086  }
2087 }
2088 #endif // ndef DISABLED_LEGACY_ENGINE
2089 
2090 // If a word has multiple alternates check if the best choice is in the
2091 // dictionary. If not, replace it with an alternate that exists in the
2092 // dictionary.
2094  PAGE_RES_IT word_it(page_res);
2095  for (WERD_RES* word = word_it.word(); word != nullptr;
2096  word = word_it.forward()) {
2097  if (word->best_choices.singleton())
2098  continue; // There are no alternates.
2099 
2100  const WERD_CHOICE* best = word->best_choice;
2101  if (word->tesseract->getDict().valid_word(*best) != 0)
2102  continue; // The best choice is in the dictionary.
2103 
2104  WERD_CHOICE_IT choice_it(&word->best_choices);
2105  for (choice_it.mark_cycle_pt(); !choice_it.cycled_list();
2106  choice_it.forward()) {
2107  WERD_CHOICE* alternate = choice_it.data();
2108  if (word->tesseract->getDict().valid_word(*alternate)) {
2109  // The alternate choice is in the dictionary.
2110  if (tessedit_bigram_debug) {
2111  tprintf("Dictionary correction replaces best choice '%s' with '%s'\n",
2112  best->unichar_string().string(),
2113  alternate->unichar_string().string());
2114  }
2115  // Replace the 'best' choice with a better choice.
2116  word->ReplaceBestChoice(alternate);
2117  break;
2118  }
2119  }
2120  }
2121 }
2122 
2123 } // namespace tesseract
const char *const kBackUpConfigFile
Definition: control.cpp:48
const double kMinRefitXHeightFraction
Definition: control.cpp:51
ACCEPTABLE_WERD_TYPE
Definition: control.h:29
@ AC_UC_ABBREV
A.B.C.
Definition: control.h:35
@ AC_INITIAL_CAP
ALL but initial lc.
Definition: control.h:33
@ AC_LC_ABBREV
a.b.c.
Definition: control.h:34
@ AC_UNACCEPTABLE
Unacceptable word.
Definition: control.h:30
@ AC_UPPER_CASE
ALL upper case.
Definition: control.h:32
@ AC_LOWER_CASE
ALL lower case.
Definition: control.h:31
FILE * debug_fp
Definition: tessvars.cpp:24
PAGE_RES_IT * make_pseudo_word(PAGE_RES *page_res, const TBOX &selection_box)
Definition: werdit.cpp:35
IncorrectResultReason
Definition: blamer.h:51
@ IRR_NUM_REASONS
Definition: blamer.h:98
bool EqualIgnoringCaseAndTerminalPunct(const WERD_CHOICE &word1, const WERD_CHOICE &word2)
Definition: ratngs.cpp:809
BLOB_CHOICE * FindMatchingChoice(UNICHAR_ID char_id, BLOB_CHOICE_LIST *bc_list)
Definition: ratngs.cpp:184
@ FREQ_DAWG_PERM
Definition: ratngs.h:244
@ USER_DAWG_PERM
Definition: ratngs.h:243
@ SYSTEM_DAWG_PERM
Definition: ratngs.h:241
@ W_EOL
end of line
Definition: werd.h:33
@ W_REP_CHAR
repeated character
Definition: werd.h:38
@ W_BOL
start of line
Definition: werd.h:32
#define LOC_DOC_BLK_REJ
Definition: errcode.h:52
#define SUBLOC_NORM
Definition: errcode.h:58
#define LOC_MM_ADAPT
Definition: errcode.h:51
void set_global_subloc_code(int loc_code)
Definition: globaloc.cpp:30
#define LOC_FUZZY_SPACE
Definition: errcode.h:49
#define ASSERT_HOST(x)
Definition: errcode.h:88
#define LOC_WRITE_RESULTS
Definition: errcode.h:53
void set_global_loc_code(int loc_code)
Definition: globaloc.cpp:25
DLLSYM void tprintf(const char *format,...)
Definition: tprintf.cpp:35
int UNICHAR_ID
Definition: unichar.h:34
int count(LIST var_list)
Definition: oldlist.cpp:95
ScrollView * fx_win
Definition: drawfx.cpp:40
void clear_fx_win()
Definition: drawfx.cpp:62
void create_fx_win()
Definition: drawfx.cpp:49
@ OEM_TESSERACT_LSTM_COMBINED
Definition: publictypes.h:271
@ OEM_TESSERACT_ONLY
Definition: publictypes.h:269
@ SET_PARAM_CONSTRAINT_DEBUG_ONLY
Definition: params.h:37
void(Tesseract::*)(const WordData &, WERD_RES **, PointerVector< WERD_RES > *) WordRecognizer
void init_to_size(int size, const T &t)
int push_back(T object)
bool empty() const
Definition: genericvector.h:91
int size() const
Definition: genericvector.h:72
T & back() const
int length() const
Definition: genericvector.h:86
PointerVector< WERD_RES > lang_words
void LSTMRecognizeWord(const BLOCK &block, ROW *row, WERD_RES *word, PointerVector< WERD_RES > *words)
Definition: linerec.cpp:229
bool recog_interactive(PAGE_RES_IT *pr_it)
Definition: control.cpp:77
void classify_word_and_language(int pass_n, PAGE_RES_IT *pr_it, WordData *word_data)
Definition: control.cpp:1319
void match_word_pass_n(int pass_n, WERD_RES *word, ROW *row, BLOCK *block)
Definition: control.cpp:1630
void bigram_correction_pass(PAGE_RES *page_res)
Definition: control.cpp:467
float ComputeCompatibleXheight(WERD_RES *word_res, float *baseline_shift)
Definition: fixxht.cpp:102
void SetupAllWordsPassN(int pass_n, const TBOX *target_word_box, const char *word_config, PAGE_RES *page_res, GenericVector< WordData > *words)
Definition: control.cpp:154
void fix_rep_char(PAGE_RES_IT *page_res_it)
Definition: control.cpp:1706
void SetupWordPassN(int pass_n, WordData *word)
Definition: control.cpp:177
void recog_pseudo_word(PAGE_RES *page_res, TBOX &selection_box)
Definition: control.cpp:62
int16_t word_outline_errs(WERD_RES *word)
Definition: docqual.cpp:72
void rejection_passes(PAGE_RES *page_res, ETEXT_DESC *monitor, const TBOX *target_word_box, const char *word_config)
Definition: control.cpp:612
void tess_segment_pass_n(int pass_n, WERD_RES *word)
Definition: tessbox.cpp:32
bool SelectGoodDiacriticOutlines(int pass, float certainty_threshold, PAGE_RES_IT *pr_it, C_BLOB *blob, const GenericVector< C_OUTLINE * > &outlines, int num_outlines, GenericVector< bool > *ok_outlines)
Definition: control.cpp:1140
bool tessedit_enable_bigram_correction
ACCEPTABLE_WERD_TYPE acceptable_word_string(const UNICHARSET &char_set, const char *s, const char *lengths)
Definition: control.cpp:1745
void output_pass(PAGE_RES_IT &page_res_it, const TBOX *target_word_box)
Definition: output.cpp:36
Dict & getDict() override
bool TestNewNormalization(int original_misfits, float baseline_shift, float new_x_ht, WERD_RES *word, BLOCK *block, ROW *row)
Definition: control.cpp:1519
void quality_based_rejection(PAGE_RES_IT &page_res_it, bool good_quality_doc)
Definition: docqual.cpp:138
void classify_word_pass1(const WordData &word_data, WERD_RES **in_word, PointerVector< WERD_RES > *out_words)
Definition: control.cpp:1401
bool SubAndSuperscriptFix(WERD_RES *word_res)
float ClassifyBlobAsWord(int pass_n, PAGE_RES_IT *pr_it, C_BLOB *blob, STRING *best_str, float *c2)
Definition: control.cpp:1270
int RetryWithLanguage(const WordData &word_data, WordRecognizer recognizer, bool debug, WERD_RES **in_word, PointerVector< WERD_RES > *best_words)
Definition: control.cpp:904
void dictionary_correction_pass(PAGE_RES *page_res)
Definition: control.cpp:2093
bool check_debug_pt(WERD_RES *word, int location)
Definition: control.cpp:1849
void word_char_quality(WERD_RES *word, ROW *row, int16_t *match_count, int16_t *accepted_match_count)
Definition: docqual.cpp:92
void tess_add_doc_word(WERD_CHOICE *word_choice)
Definition: tessbox.cpp:72
void font_recognition_pass(PAGE_RES *page_res)
Definition: control.cpp:2037
bool ReassignDiacritics(int pass, PAGE_RES_IT *pr_it, bool *make_next_word_fuzzy)
Definition: control.cpp:945
bool AnyTessLang() const
bool ProcessTargetWord(const TBOX &word_box, const TBOX &target_word_box, const char *word_config, int pass)
Definition: control.cpp:120
void AssignDiacriticsToOverlappingBlobs(const GenericVector< C_OUTLINE * > &outlines, int pass, WERD *real_word, PAGE_RES_IT *pr_it, GenericVector< bool > *word_wanted, GenericVector< bool > *overlapped_any_blob, GenericVector< C_BLOB * > *target_blobs)
Definition: control.cpp:1011
int CountMisfitTops(WERD_RES *word_res)
Definition: fixxht.cpp:70
int16_t word_blob_quality(WERD_RES *word, ROW *row)
Definition: docqual.cpp:60
bool TrainedXheightFix(WERD_RES *word, BLOCK *block, ROW *row)
Definition: control.cpp:1485
bool word_adaptable(WERD_RES *word, uint16_t mode)
Definition: adaptions.cpp:34
void set_word_fonts(WERD_RES *word)
Definition: control.cpp:1962
void fix_fuzzy_spaces(ETEXT_DESC *monitor, int32_t word_count, PAGE_RES *page_res)
Definition: fixspace.cpp:75
bool AnyLSTMLang() const
bool tess_acceptable_word(WERD_RES *word)
Definition: tessbox.cpp:62
bool right_to_left() const
void script_pos_pass(PAGE_RES *page_res)
Definition: control.cpp:734
bool RecogAllWordsPassN(int pass_n, ETEXT_DESC *monitor, PAGE_RES_IT *pr_it, GenericVector< WordData > *words)
Definition: control.cpp:213
Pix * BestPix() const
void PrerecAllWordsPar(const GenericVector< WordData > &words)
Definition: par_control.cpp:38
void blamer_pass(PAGE_RES *page_res)
Definition: control.cpp:710
void classify_word_pass2(const WordData &word_data, WERD_RES **in_word, PointerVector< WERD_RES > *out_words)
Definition: control.cpp:1572
void AssignDiacriticsToNewBlobs(const GenericVector< C_OUTLINE * > &outlines, int pass, WERD *real_word, PAGE_RES_IT *pr_it, GenericVector< bool > *word_wanted, GenericVector< C_BLOB * > *target_blobs)
Definition: control.cpp:1064
void make_reject_map(WERD_RES *word, ROW *row, int16_t pass)
bool recog_all_words(PAGE_RES *page_res, ETEXT_DESC *monitor, const TBOX *target_word_box, const char *word_config, int dopasses)
Definition: control.cpp:302
float ClassifyBlobPlusOutlines(const GenericVector< bool > &ok_outlines, const GenericVector< C_OUTLINE * > &outlines, int pass_n, PAGE_RES_IT *pr_it, C_BLOB *blob, STRING *best_str)
Definition: control.cpp:1226
void ReportXhtFixResult(bool accept_new_word, float new_x_ht, WERD_RES *word, WERD_RES *new_word)
Definition: control.cpp:1462
static void LastChanceBlame(bool debug, WERD_RES *word)
Definition: blamer.cpp:560
const STRING & misadaption_debug() const
Definition: blamer.h:133
void SetMisAdaptionDebug(const WERD_CHOICE *best_choice, bool debug)
Definition: blamer.cpp:587
void CopyTruth(const BlamerBundle &other)
Definition: blamer.h:203
static const char * IncorrectReasonName(IncorrectResultReason irr)
Definition: blamer.cpp:64
IncorrectResultReason incorrect_result_reason() const
Definition: blamer.h:120
TBOX bounding_box() const
Definition: blobs.cpp:861
void plot(ScrollView *window)
Definition: blobs.cpp:897
int length() const
Definition: boxword.h:83
static const double kXHeightCapRatio
Definition: ccstruct.h:37
int32_t universal_id
Definition: fontinfo.h:123
Definition: ocrblock.h:31
PDBLK pdblk
Page Description Block.
Definition: ocrblock.h:190
FCOORD classify_rotation() const
Definition: ocrblock.h:140
int32_t x_height() const
return xheight
Definition: ocrblock.h:106
Definition: ocrrow.h:37
float x_height() const
Definition: ocrrow.h:64
GenericVector< int > blame_reasons
Definition: pageres.h:86
int32_t rej_count
Definition: pageres.h:79
int32_t char_count
Definition: pageres.h:78
GenericVector< STRING > misadaption_log
Definition: pageres.h:91
BLOCK * block
Definition: pageres.h:116
ROW * row
Definition: pageres.h:140
const UNICHARSET * uch_set
Definition: pageres.h:203
bool tess_would_adapt
Definition: pageres.h:304
TWERD * rebuild_word
Definition: pageres.h:266
WERD_CHOICE_LIST best_choices
Definition: pageres.h:249
bool guessed_x_ht
Definition: pageres.h:313
BlamerBundle * blamer_bundle
Definition: pageres.h:252
bool done
Definition: pageres.h:305
tesseract::Tesseract * tesseract
Definition: pageres.h:280
int8_t fontinfo_id2_count
Definition: pageres.h:312
const FontInfo * fontinfo
Definition: pageres.h:309
tesseract::BoxWord * box_word
Definition: pageres.h:272
WERD_CHOICE * best_choice
Definition: pageres.h:241
void ReplaceBestChoice(WERD_CHOICE *choice)
Definition: pageres.cpp:795
const FontInfo * fontinfo2
Definition: pageres.h:310
float x_height
Definition: pageres.h:316
bool part_of_combo
Definition: pageres.h:340
void PrintBestChoices() const
Definition: pageres.cpp:717
bool IsAmbiguous()
Definition: pageres.cpp:452
void ConsumeWordResults(WERD_RES *word)
Definition: pageres.cpp:765
void BestChoiceToCorrectText()
Definition: pageres.cpp:923
bool tess_failed
Definition: pageres.h:295
void SetScriptPositions()
Definition: pageres.cpp:858
bool tess_accepted
Definition: pageres.h:303
BLOB_CHOICE_LIST * GetBlobChoices(int index) const
Definition: pageres.cpp:759
WERD_CHOICE * raw_choice
Definition: pageres.h:246
void fix_hyphens()
Definition: pageres.cpp:1047
bool small_caps
Definition: pageres.h:306
int8_t fontinfo_id_count
Definition: pageres.h:311
TWERD * chopped_word
Definition: pageres.h:212
void fix_quotes()
Definition: pageres.cpp:1018
bool SetupForRecognition(const UNICHARSET &unicharset_in, tesseract::Tesseract *tesseract, Pix *pix, int norm_mode, const TBOX *norm_box, bool numeric_mode, bool use_body_size, bool allow_detailed_fx, ROW *row, const BLOCK *block)
Definition: pageres.cpp:302
REJMAP reject_map
Definition: pageres.h:294
float caps_height
Definition: pageres.h:317
void InitForRetryRecognition(const WERD_RES &source)
Definition: pageres.cpp:277
WERD * word
Definition: pageres.h:186
float baseline_shift
Definition: pageres.h:318
BLOB_CHOICE * GetBlobChoice(int index) const
Definition: pageres.cpp:750
WERD_RES * restart_page()
Definition: pageres.h:701
void MakeCurrentWordFuzzy()
Definition: pageres.cpp:1473
void rej_stat_word()
Definition: pageres.cpp:1667
ROW_RES * row() const
Definition: pageres.h:757
void ReplaceCurrentWord(tesseract::PointerVector< WERD_RES > *words)
Definition: pageres.cpp:1333
void ResetWordIterator()
Definition: pageres.cpp:1523
WERD_RES * word() const
Definition: pageres.h:754
BLOCK_RES * block() const
Definition: pageres.h:760
PAGE_RES * page_res
Definition: pageres.h:677
WERD_RES * forward()
Definition: pageres.h:734
WERD_RES * InsertSimpleCloneWord(const WERD_RES &clone_res, WERD *new_word)
Definition: pageres.cpp:1213
void DeleteCurrentWord()
Definition: pageres.cpp:1440
POLY_BLOCK * poly_block() const
Definition: pdblock.h:55
Definition: points.h:189
float y() const
Definition: points.h:210
bool IsText() const
Definition: polyblk.h:49
const GenericVector< tesseract::ScoredFont > & fonts() const
Definition: ratngs.h:93
float rating() const
Definition: ratngs.h:80
UNICHAR_ID unichar_id() const
Definition: ratngs.h:77
WERD_CHOICE shallow_copy(int start, int end) const
Definition: ratngs.cpp:418
const STRING debug_string() const
Definition: ratngs.h:495
bool IsAllSpaces() const
Definition: ratngs.h:511
UNICHAR_ID unichar_id(int index) const
Definition: ratngs.h:305
uint8_t permuter() const
Definition: ratngs.h:336
void set_unichar_id(UNICHAR_ID unichar_id, int index)
Definition: ratngs.h:349
float min_x_height() const
Definition: ratngs.h:326
float certainty() const
Definition: ratngs.h:320
int length() const
Definition: ratngs.h:293
float max_x_height() const
Definition: ratngs.h:329
const STRING & unichar_string() const
Definition: ratngs.h:531
void GetNonSuperscriptSpan(int *start, int *end) const
Definition: ratngs.cpp:401
float rating() const
Definition: ratngs.h:317
Definition: rect.h:34
bool x_overlap(const TBOX &box) const
Definition: rect.h:401
int16_t top() const
Definition: rect.h:58
void print() const
Definition: rect.h:278
bool major_overlap(const TBOX &box) const
Definition: rect.h:368
int16_t left() const
Definition: rect.h:72
int16_t bottom() const
Definition: rect.h:65
bool contains(const FCOORD pt) const
Definition: rect.h:333
bool major_x_overlap(const TBOX &box) const
Definition: rect.h:412
int16_t right() const
Definition: rect.h:79
void print(FILE *fp)
Definition: rejctmap.cpp:321
int16_t reject_count()
Definition: rejctmap.h:229
void full_print(FILE *fp)
Definition: rejctmap.cpp:333
void initialise(int16_t length)
Definition: rejctmap.cpp:273
void rej_word_bad_quality()
Definition: rejctmap.cpp:415
int32_t length() const
Definition: rejctmap.h:223
Definition: statistc.h:31
int32_t pile_count(int32_t value) const
Definition: statistc.h:76
void add(int32_t value, int32_t count)
Definition: statistc.cpp:93
int32_t get_total() const
Definition: statistc.h:84
int32_t mode() const
Definition: statistc.cpp:107
TBOX bounding_box() const
Definition: stepblob.cpp:253
static int SortByXMiddle(const void *v1, const void *v2)
Definition: stepblob.h:125
static C_BLOB * deep_copy(const C_BLOB *src)
Definition: stepblob.h:119
C_OUTLINE_LIST * out_list()
Definition: stepblob.h:70
Definition: werd.h:56
C_BLOB_LIST * rej_cblob_list()
Definition: werd.h:90
C_BLOB_LIST * cblob_list()
Definition: werd.h:95
WERD * ConstructFromSingleBlob(bool bol, bool eol, C_BLOB *blob)
Definition: werd.cpp:125
bool AddSelectedOutlines(const GenericVector< bool > &wanted, const GenericVector< C_BLOB * > &target_blobs, const GenericVector< C_OUTLINE * > &outlines, bool *make_next_word_fuzzy)
Definition: werd.cpp:524
void print()
Definition: werd.cpp:253
TBOX bounding_box() const
Definition: werd.cpp:148
bool flag(WERD_FLAGS mask) const
Definition: werd.h:117
void GetNoiseOutlines(GenericVector< C_OUTLINE * > *outlines)
Definition: werd.cpp:506
UNICHARSET unicharset
Definition: ccutil.h:73
ParamsVectors * params()
Definition: ccutil.h:67
STRING lang
Definition: ccutil.h:71
volatile int8_t ocr_alive
true if not last
Definition: ocrclass.h:110
int16_t progress
chars in this buffer(0)
Definition: ocrclass.h:105
void * cancel_this
monitor-aware progress callback
Definition: ocrclass.h:116
PROGRESS_FUNC2 progress_callback2
called whenever progress increases
Definition: ocrclass.h:115
bool deadline_exceeded() const
Definition: ocrclass.h:138
CANCEL_FUNC cancel
for errcode use
Definition: ocrclass.h:112
static bool ReadParamsFile(const char *file, SetParamConstraint constraint, ParamsVectors *member_params)
Definition: params.cpp:39
static void PrintParams(FILE *fp, const ParamsVectors *member_params)
Definition: params.cpp:168
void Add(T value, int count)
Definition: sorthelper.h:65
int MaxCount(T *max_value) const
Definition: sorthelper.h:80
Definition: strngs.h:45
void add_str_int(const char *str, int number)
Definition: strngs.cpp:377
int32_t length() const
Definition: strngs.cpp:189
const char * string() const
Definition: strngs.cpp:194
bool top_bottom_useful() const
Definition: unicharset.h:537
bool get_isupper(UNICHAR_ID unichar_id) const
Definition: unicharset.h:505
STRING debug_str(UNICHAR_ID id) const
Definition: unicharset.cpp:343
bool script_has_xheight() const
Definition: unicharset.h:904
bool get_islower(UNICHAR_ID unichar_id) const
Definition: unicharset.h:498
double classify_max_rating_ratio
Definition: classify.h:438
bool AdaptiveClassifierIsEmpty() const
Definition: classify.h:326
bool AdaptableWord(WERD_RES *word)
Definition: adaptmatch.cpp:821
void LearnWord(const char *fontname, WERD_RES *word)
Definition: adaptmatch.cpp:250
void StartBackupAdaptiveClassifier()
Definition: adaptmatch.cpp:629
bool classify_bln_numeric_mode
Definition: classify.h:508
UnicityTable< FontInfo > & get_fontinfo_table()
Definition: classify.h:386
double classify_max_certainty_margin
Definition: classify.h:440
void SwitchAdaptiveClassifier()
Definition: adaptmatch.cpp:613
bool AdaptiveClassifierIsFull() const
Definition: classify.h:325
UnicityTable< FontInfo > fontinfo_table_
Definition: classify.h:529
static bool valid_word_permuter(uint8_t perm, bool numbers_ok)
Check all the DAWGs to see if this word is in any of them.
Definition: dict.h:474
bool valid_bigram(const WERD_CHOICE &word1, const WERD_CHOICE &word2) const
Definition: dict.cpp:813
const UNICHARSET & GetUnicharset() const
void CleanupSingleRowResult(PageSegMode pageseg_mode, PAGE_RES *page_res)
Definition: textord.cpp:318
static void Update()
Definition: scrollview.cpp:709
void ZoomToRectangle(int x1, int y1, int x2, int y2)
Definition: scrollview.cpp:757
WERD_CHOICE * prev_word_best_choice_
Definition: wordrec.h:476
bool wordrec_run_blamer
Definition: wordrec.h:232
bool wordrec_debug_blamer
Definition: wordrec.h:231