1 /**********************************************************************
2 * File: adaptions.cpp (Formerly adaptions.c)
3 * Description: Functions used to adapt to blobs already confidently
4 * identified
5 * Author: Chris Newton
6 * Created: Thu Oct 7 10:17:28 BST 1993
7 *
8 * (C) Copyright 1992, Hewlett-Packard Ltd.
9 ** Licensed under the Apache License, Version 2.0 (the "License");
10 ** you may not use this file except in compliance with the License.
11 ** You may obtain a copy of the License at
12 ** http://www.apache.org/licenses/LICENSE-2.0
13 ** Unless required by applicable law or agreed to in writing, software
14 ** distributed under the License is distributed on an "AS IS" BASIS,
15 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 ** See the License for the specific language governing permissions and
17 ** limitations under the License.
18 *
19 **********************************************************************/
20
21 #include "mfcpch.h"
22 #ifdef __UNIX__
23 #include <assert.h>
24 #endif
25 #include <ctype.h>
26 #include <string.h>
27 #include "tessbox.h"
28 #include "tessvars.h"
29 #include "memry.h"
30 #include "mainblk.h"
31 #include "charcut.h"
32 #include "imgs.h"
33 #include "scaleimg.h"
34 #include "reject.h"
35 #include "control.h"
36 #include "adaptions.h"
37 #include "stopper.h"
38 #include "charsample.h"
39 #include "matmatch.h"
40 #include "secname.h"
41 #include "tesseractclass.h"
42
43 inT32 demo_word = 0;
44
45 #define WINDOWNAMESIZE 13 /*max size of name */
46
47 #define EXTERN
48
49 EXTERN BOOL_VAR (tessedit_reject_ems, FALSE, "Reject all m's");
50 EXTERN BOOL_VAR (tessedit_reject_suspect_ems, FALSE, "Reject suspect m's");
51
52 EXTERN double_VAR (tessedit_cluster_t1, 0.20,
53 "t1 threshold for clustering samples");
54 EXTERN double_VAR (tessedit_cluster_t2, 0.40,
55 "t2 threshold for clustering samples");
56 EXTERN double_VAR (tessedit_cluster_t3, 0.12,
57 "Extra threshold for clustering samples, only keep a new sample if best score greater than this value");
58 EXTERN double_VAR (tessedit_cluster_accept_fraction, 0.80,
59 "Largest fraction of characters in cluster for it to be used for adaption");
60 EXTERN INT_VAR (tessedit_cluster_min_size, 3,
61 "Smallest number of samples in a cluster for it to be used for adaption");
62 EXTERN BOOL_VAR (tessedit_cluster_debug, FALSE,
63 "Generate and print debug information for adaption by clustering");
64 EXTERN BOOL_VAR (tessedit_use_best_sample, FALSE,
65 "Use best sample from cluster when adapting");
66 EXTERN BOOL_VAR (tessedit_test_cluster_input, FALSE,
67 "Set reject map to enable cluster input to be measured");
68
69 EXTERN BOOL_VAR (tessedit_matrix_match, TRUE, "Use matrix matcher");
70 EXTERN BOOL_VAR (tessedit_mm_use_non_adaption_set, FALSE,
71 "Don't try to adapt to characters on this list");
72 EXTERN STRING_VAR (tessedit_non_adaption_set, ",.;:'~@*",
73 "Characters to be avoided when adapting");
74 EXTERN BOOL_VAR (tessedit_mm_adapt_using_prototypes, TRUE,
75 "Use prototypes when adapting");
76 EXTERN BOOL_VAR (tessedit_mm_use_prototypes, TRUE,
77 "Use prototypes as clusters are built");
78 EXTERN BOOL_VAR (tessedit_mm_use_rejmap, FALSE,
79 "Adapt to characters using reject map");
80 EXTERN BOOL_VAR (tessedit_mm_all_rejects, FALSE,
81 "Adapt to all characters using, matrix matcher");
82 EXTERN BOOL_VAR (tessedit_mm_only_match_same_char, FALSE,
83 "Only match samples against clusters for the same character");
84 EXTERN BOOL_VAR (tessedit_process_rns, FALSE, "Handle m - rn ambigs");
85
86 EXTERN BOOL_VAR (tessedit_demo_adaption, FALSE,
87 "Display cut images and matrix match for demo purposes");
88 EXTERN INT_VAR (tessedit_demo_word1, 62,
89 "Word number of first word to display");
90 EXTERN INT_VAR (tessedit_demo_word2, 64,
91 "Word number of second word to display");
92 EXTERN STRING_VAR (tessedit_demo_file, "academe",
93 "Name of document containing demo words");
94 EXTERN BOOL_VAR(tessedit_adapt_to_char_fragments, TRUE,
95 "Adapt to words that contain "
96 " a character composed form fragments");
97
98 namespace tesseract {
word_adaptable(WERD_RES * word,uinT16 mode)99 BOOL8 Tesseract::word_adaptable( //should we adapt?
100 WERD_RES *word,
101 uinT16 mode) {
102 if (tessedit_adaption_debug) {
103 tprintf("Running word_adaptable() for %s rating %.4f certainty %.4f\n",
104 word->best_choice == NULL ? "" :
105 word->best_choice->unichar_string().string(),
106 word->best_choice->rating(), word->best_choice->certainty());
107 }
108
109 BOOL8 status = FALSE;
110 BITS16 flags(mode);
111
112 enum MODES
113 {
114 ADAPTABLE_WERD,
115 ACCEPTABLE_WERD,
116 CHECK_DAWGS,
117 CHECK_SPACES,
118 CHECK_ONE_ELL_CONFLICT,
119 CHECK_AMBIG_WERD
120 };
121
122 /*
123 0: NO adaption
124 */
125 if (mode == 0) {
126 if (tessedit_adaption_debug) tprintf("adaption disabled\n");
127 return FALSE;
128 }
129
130 if (flags.bit (ADAPTABLE_WERD)) {
131 status |= word->tess_would_adapt; // result of Classify::AdaptableWord()
132 if (tessedit_adaption_debug && !status) {
133 tprintf("tess_would_adapt bit is false\n");
134 }
135 }
136
137 if (flags.bit (ACCEPTABLE_WERD)) {
138 status |= word->tess_accepted;
139 if (tessedit_adaption_debug && !status) {
140 tprintf("tess_accepted bit is false\n");
141 }
142 }
143
144 if (!status) { // If not set then
145 return FALSE; // ignore other checks
146 }
147
148 if (flags.bit (CHECK_DAWGS) &&
149 (word->best_choice->permuter () != SYSTEM_DAWG_PERM) &&
150 (word->best_choice->permuter () != FREQ_DAWG_PERM) &&
151 (word->best_choice->permuter () != USER_DAWG_PERM) &&
152 (word->best_choice->permuter () != NUMBER_PERM)) {
153 if (tessedit_adaption_debug) tprintf("word not in dawgs\n");
154 return FALSE;
155 }
156
157 if (flags.bit (CHECK_ONE_ELL_CONFLICT) && one_ell_conflict (word, FALSE)) {
158 if (tessedit_adaption_debug) tprintf("word has ell conflict\n");
159 return FALSE;
160 }
161
162 if (flags.bit (CHECK_SPACES) &&
163 (strchr(word->best_choice->unichar_string().string(), ' ') != NULL)) {
164 if (tessedit_adaption_debug) tprintf("word contains spaces\n");
165 return FALSE;
166 }
167
168 // if (flags.bit (CHECK_AMBIG_WERD) && test_ambig_word (word))
169 if (flags.bit (CHECK_AMBIG_WERD) &&
170 !getDict().NoDangerousAmbig(word->best_choice, NULL, false, NULL, NULL)) {
171 if (tessedit_adaption_debug) tprintf("word is ambiguous\n");
172 return FALSE;
173 }
174
175 // Do not adapt to words that are composed from fragments if
176 // tessedit_adapt_to_char_fragments is false.
177 if (!tessedit_adapt_to_char_fragments) {
178 const char *fragment_lengths = word->best_choice->fragment_lengths();
179 if (fragment_lengths != NULL && *fragment_lengths != '\0') {
180 for (int i = 0; i < word->best_choice->length(); ++i) {
181 if (fragment_lengths[i] > 1) {
182 if (tessedit_adaption_debug) tprintf("won't adapt to fragments\n");
183 return false; // found a character composed from fragments
184 }
185 }
186 }
187 }
188
189 if (tessedit_adaption_debug) {
190 tprintf("returning status %d\n", status);
191 }
192 return status;
193
194 }
195
196
collect_ems_for_adaption(WERD_RES * word,CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)197 void Tesseract::collect_ems_for_adaption(WERD_RES *word,
198 CHAR_SAMPLES_LIST *char_clusters,
199 CHAR_SAMPLE_LIST *chars_waiting) {
200 PBLOB_LIST *blobs = word->outword->blob_list ();
201 PBLOB_IT blob_it(blobs);
202 inT16 i;
203 CHAR_SAMPLE *sample;
204 PIXROW_LIST *pixrow_list;
205 PIXROW_IT pixrow_it;
206 IMAGELINE *imlines; // lines of the image
207 TBOX pix_box; // box of imlines
208 // extent
209 WERD copy_outword; // copy to denorm
210 PBLOB_IT copy_blob_it;
211 OUTLINE_IT copy_outline_it;
212 inT32 resolution = page_image.get_res ();
213
214 if (tessedit_reject_ems || tessedit_reject_suspect_ems)
215 return; // Do nothing
216
217 if (word->word->bounding_box ().height () > resolution / 3)
218 return;
219
220 if (tessedit_demo_adaption)
221 // Make sure not set
222 tessedit_display_mm.set_value (FALSE);
223
224 if (word_adaptable (word, tessedit_em_adaption_mode)
225 && word->reject_map.reject_count () == 0
226 && (strchr (word->best_choice->unichar_string().string (), 'm') != NULL
227 || (tessedit_process_rns
228 && strstr (word->best_choice->unichar_string().string (),
229 "rn") != NULL))) {
230 if (tessedit_process_rns
231 && strstr (word->best_choice->unichar_string().string (), "rn") != NULL) {
232 copy_outword = *(word->outword);
233 copy_blob_it.set_to_list (copy_outword.blob_list ());
234 i = 0;
235 while (word->best_choice->unichar_string()[i] != '\0') {
236 if (word->best_choice->unichar_string()[i] == 'r'
237 && word->best_choice->unichar_string()[i + 1] == 'n') {
238 copy_outline_it.set_to_list (copy_blob_it.data ()->
239 out_list ());
240 copy_outline_it.add_list_after (copy_blob_it.
241 data_relative (1)->
242 out_list ());
243 copy_blob_it.forward ();
244 delete (copy_blob_it.extract ());
245 i++;
246 }
247 copy_blob_it.forward ();
248 i++;
249 }
250 }
251 else
252 copy_outword = *(word->outword);
253
254 copy_outword.baseline_denormalise (&word->denorm);
255 char_clip_word(©_outword, page_image, pixrow_list, imlines, pix_box);
256 pixrow_it.set_to_list (pixrow_list);
257 pixrow_it.move_to_first ();
258
259 blob_it.move_to_first ();
260 for (i = 0;
261 word->best_choice->unichar_string()[i] != '\0';
262 i++, pixrow_it.forward (), blob_it.forward ()) {
263
264 if (word->best_choice->unichar_string()[i] == 'm'
265 || (word->best_choice->unichar_string()[i] == 'r'
266 && word->best_choice->unichar_string()[i + 1] == 'n')) {
267 #ifndef SECURE_NAMES
268 if (tessedit_cluster_debug)
269 tprintf ("Sample %c for adaption found in %s, index %d\n",
270 word->best_choice->unichar_string()[i],
271 word->best_choice->unichar_string().string (), i);
272 #endif
273 if (tessedit_matrix_match) {
274 sample = clip_sample (pixrow_it.data (),
275 imlines,
276 pix_box,
277 copy_outword.flag (W_INVERSE),
278 word->best_choice->unichar_string()[i]);
279
280 if (sample == NULL) { //Clip failed
281 #ifndef SECURE_NAMES
282 tprintf ("Unable to clip sample from %s, index %d\n",
283 word->best_choice->unichar_string().string (), i);
284 #endif
285 if (word->best_choice->unichar_string()[i] == 'r')
286 i++;
287
288 continue;
289 }
290 }
291 else
292 sample = new CHAR_SAMPLE (blob_it.data (),
293 &word->denorm,
294 word->best_choice->unichar_string()[i]);
295
296 cluster_sample(sample, char_clusters, chars_waiting);
297
298 if (word->best_choice->unichar_string()[i] == 'r')
299 i++; // Skip next character
300 }
301 }
302 delete[]imlines; // Free array of imlines
303 delete pixrow_list;
304 }
305 }
306
307
collect_characters_for_adaption(WERD_RES * word,CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)308 void Tesseract::collect_characters_for_adaption(
309 WERD_RES *word,
310 CHAR_SAMPLES_LIST *char_clusters,
311 CHAR_SAMPLE_LIST *chars_waiting) {
312 PBLOB_LIST *blobs = word->outword->blob_list ();
313 PBLOB_IT blob_it(blobs);
314 inT16 i;
315 CHAR_SAMPLE *sample;
316 PIXROW_LIST *pixrow_list;
317 PIXROW_IT pixrow_it;
318 IMAGELINE *imlines; // lines of the image
319 TBOX pix_box; // box of imlines
320 // extent
321 WERD copy_outword; // copy to denorm
322 inT32 resolution = page_image.get_res ();
323
324 if (word->word->bounding_box ().height () > resolution / 3)
325 return;
326
327 if (tessedit_demo_adaption)
328 // Make sure not set
329 tessedit_display_mm.set_value (FALSE);
330
331 if ((word_adaptable (word, tessedit_cluster_adaption_mode)
332 && word->reject_map.reject_count () == 0) || tessedit_mm_use_rejmap) {
333 if (tessedit_test_cluster_input && !tessedit_mm_use_rejmap)
334 return; // Reject map set to acceptable
335 /* Collect information about good matches */
336 copy_outword = *(word->outword);
337 copy_outword.baseline_denormalise (&word->denorm);
338 char_clip_word(©_outword, page_image, pixrow_list, imlines, pix_box);
339 pixrow_it.set_to_list (pixrow_list);
340 pixrow_it.move_to_first ();
341
342 blob_it.move_to_first ();
343 for (i = 0;
344 word->best_choice->unichar_string()[i] != '\0';
345 i++, pixrow_it.forward (), blob_it.forward ()) {
346
347 if (!(tessedit_mm_use_non_adaption_set
348 && STRING(tessedit_non_adaption_set).contains(
349 word->best_choice->unichar_string()[i]))
350 || (tessedit_mm_use_rejmap && word->reject_map[i].accepted ())) {
351 #ifndef SECURE_NAMES
352 if (tessedit_cluster_debug)
353 tprintf ("Sample %c for adaption found in %s, index %d\n",
354 word->best_choice->unichar_string()[i],
355 word->best_choice->unichar_string().string (), i);
356 #endif
357 sample = clip_sample (pixrow_it.data (),
358 imlines,
359 pix_box,
360 copy_outword.flag (W_INVERSE),
361 word->best_choice->unichar_string()[i]);
362
363 if (sample == NULL) { //Clip failed
364 #ifndef SECURE_NAMES
365 tprintf ("Unable to clip sample from %s, index %d\n",
366 word->best_choice->unichar_string().string (), i);
367 #endif
368 continue;
369 }
370 cluster_sample(sample, char_clusters, chars_waiting);
371 }
372 }
373 delete[]imlines; // Free array of imlines
374 delete pixrow_list;
375 }
376 else if (tessedit_test_cluster_input && !tessedit_mm_use_rejmap)
377 // Set word to all rejects
378 word->reject_map.rej_word_tess_failure ();
379
380 }
381
382
cluster_sample(CHAR_SAMPLE * sample,CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)383 void Tesseract::cluster_sample(CHAR_SAMPLE *sample,
384 CHAR_SAMPLES_LIST *char_clusters,
385 CHAR_SAMPLE_LIST *chars_waiting) {
386 CHAR_SAMPLES *best_cluster = NULL;
387 CHAR_SAMPLES_IT c_it = char_clusters;
388 CHAR_SAMPLE_IT cw_it = chars_waiting;
389 float score;
390 float best_score = MAX_INT32;
391
392 if (c_it.empty ())
393 c_it.add_to_end (new CHAR_SAMPLES (sample));
394 else {
395 for (c_it.mark_cycle_pt (); !c_it.cycled_list (); c_it.forward ()) {
396 score = c_it.data ()->match_score (sample, this);
397 if (score < best_score) {
398 best_score = score;
399 best_cluster = c_it.data ();
400 }
401 }
402
403 if (tessedit_cluster_debug)
404 tprintf ("Sample's best score %f\n", best_score);
405
406 if (best_score < tessedit_cluster_t1) {
407 if (best_score > tessedit_cluster_t3 || tessedit_mm_use_prototypes) {
408 best_cluster->add_sample (sample, this);
409 check_wait_list(chars_waiting, sample, best_cluster);
410 #ifndef SECURE_NAMES
411 if (tessedit_cluster_debug)
412 tprintf ("Sample added to an existing cluster\n");
413 #endif
414 }
415 else {
416 #ifndef SECURE_NAMES
417 if (tessedit_cluster_debug)
418 tprintf
419 ("Sample dropped, good match to an existing cluster\n");
420 #endif
421 }
422 }
423 else if (best_score > tessedit_cluster_t2) {
424 c_it.add_to_end (new CHAR_SAMPLES (sample));
425 #ifndef SECURE_NAMES
426 if (tessedit_cluster_debug)
427 tprintf ("New cluster created for this sample\n");
428 #endif
429 }
430 else {
431 cw_it.add_to_end (sample);
432 if (tessedit_cluster_debug)
433 tprintf ("Sample added to the wait list\n");
434 }
435 }
436 }
437
check_wait_list(CHAR_SAMPLE_LIST * chars_waiting,CHAR_SAMPLE * sample,CHAR_SAMPLES * best_cluster)438 void Tesseract::check_wait_list(CHAR_SAMPLE_LIST *chars_waiting,
439 CHAR_SAMPLE *sample,
440 CHAR_SAMPLES *best_cluster) {
441 CHAR_SAMPLE *wait_sample;
442 CHAR_SAMPLE *test_sample = sample;
443 CHAR_SAMPLE_IT cw_it = chars_waiting;
444 CHAR_SAMPLE_LIST add_list; //Samples added to best cluster
445 CHAR_SAMPLE_IT add_it = &add_list;
446 float score;
447
448 add_list.clear ();
449
450 if (!cw_it.empty ()) {
451 do {
452 if (!add_list.empty ()) {
453 add_it.forward ();
454 test_sample = add_it.extract ();
455 best_cluster->add_sample (test_sample, this);
456 }
457
458 for (cw_it.mark_cycle_pt ();
459 !cw_it.cycled_list (); cw_it.forward ()) {
460 wait_sample = cw_it.data ();
461 if (tessedit_mm_use_prototypes)
462 score = best_cluster->match_score (wait_sample, this);
463 else
464 score = sample->match_sample (wait_sample, FALSE, this);
465 if (score < tessedit_cluster_t1) {
466 if (score > tessedit_cluster_t3
467 || tessedit_mm_use_prototypes) {
468 add_it.add_after_stay_put (cw_it.extract ());
469 #ifndef SECURE_NAMES
470 if (tessedit_cluster_debug)
471 tprintf
472 ("Wait sample added to an existing cluster\n");
473 #endif
474 }
475 else {
476 #ifndef SECURE_NAMES
477 if (tessedit_cluster_debug)
478 tprintf
479 ("Wait sample dropped, good match to an existing cluster\n");
480 #endif
481 }
482 }
483 }
484 }
485 while (!add_list.empty ());
486 }
487 }
488
489
complete_clustering(CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)490 void Tesseract::complete_clustering(CHAR_SAMPLES_LIST *char_clusters,
491 CHAR_SAMPLE_LIST *chars_waiting) {
492 CHAR_SAMPLES *best_cluster;
493 CHAR_SAMPLES_IT c_it = char_clusters;
494 CHAR_SAMPLE_IT cw_it = chars_waiting;
495 CHAR_SAMPLE *sample;
496 inT32 total_sample_count = 0;
497
498 while (!cw_it.empty ()) {
499 cw_it.move_to_first ();
500 sample = cw_it.extract ();
501 best_cluster = new CHAR_SAMPLES (sample);
502 c_it.add_to_end (best_cluster);
503 check_wait_list(chars_waiting, sample, best_cluster);
504 }
505
506 for (c_it.mark_cycle_pt (); !c_it.cycled_list (); c_it.forward ()) {
507 c_it.data ()->assign_to_char ();
508 if (tessedit_use_best_sample)
509 c_it.data ()->find_best_sample ();
510 else if (tessedit_mm_adapt_using_prototypes)
511 c_it.data ()->build_prototype ();
512
513 if (tessedit_cluster_debug)
514 total_sample_count += c_it.data ()->n_samples ();
515 }
516 #ifndef SECURE_NAMES
517 if (tessedit_cluster_debug)
518 tprintf ("Clustering completed, %d samples in all\n", total_sample_count);
519 #endif
520
521 #ifndef GRAPHICS_DISABLED
522 if (tessedit_demo_adaption)
523 display_cluster_prototypes(char_clusters);
524 #endif
525
526 }
527
adapt_to_good_ems(WERD_RES * word,CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)528 void Tesseract::adapt_to_good_ems(WERD_RES *word,
529 CHAR_SAMPLES_LIST *char_clusters,
530 CHAR_SAMPLE_LIST *chars_waiting) {
531 PBLOB_LIST *blobs = word->outword->blob_list ();
532 PBLOB_IT blob_it(blobs);
533 inT16 i;
534 CHAR_SAMPLE *sample;
535 CHAR_SAMPLES_IT c_it = char_clusters;
536 CHAR_SAMPLE_IT cw_it = chars_waiting;
537 float score;
538 float best_score;
539 char best_char;
540 CHAR_SAMPLES *best_cluster;
541 PIXROW_LIST *pixrow_list;
542 PIXROW_IT pixrow_it;
543 IMAGELINE *imlines; // lines of the image
544 TBOX pix_box; // box of imlines
545 // extent
546 WERD copy_outword; // copy to denorm
547 TBOX b_box;
548 PBLOB_IT copy_blob_it;
549 OUTLINE_IT copy_outline_it;
550 PIXROW *pixrow = NULL;
551
552 static inT32 word_number = 0;
553
554 #ifndef GRAPHICS_DISABLED
555 ScrollView* demo_win = NULL;
556 #endif
557
558 inT32 resolution = page_image.get_res ();
559
560 if (word->word->bounding_box ().height () > resolution / 3)
561 return;
562
563 word_number++;
564
565 if (strchr (word->best_choice->unichar_string().string (), 'm') == NULL
566 && (tessedit_process_rns
567 && strstr (word->best_choice->unichar_string().string (), "rn") == NULL))
568 return;
569
570 if (tessedit_reject_ems)
571 reject_all_ems(word);
572 else if (tessedit_reject_suspect_ems)
573 reject_suspect_ems(word);
574 else {
575 if (char_clusters->length () == 0) {
576 #ifndef SECURE_NAMES
577 if (tessedit_cluster_debug)
578 tprintf ("No clusters to use for em adaption\n");
579 #endif
580 return;
581 }
582
583 if (!cw_it.empty ()) {
584 complete_clustering(char_clusters, chars_waiting);
585 print_em_stats(char_clusters, chars_waiting);
586 }
587
588 if ((!word_adaptable (word, tessedit_em_adaption_mode) ||
589 word->reject_map.reject_count () != 0)
590 && (strchr (word->best_choice->unichar_string().string (), 'm') != NULL
591 || (tessedit_process_rns
592 && strstr (word->best_choice->unichar_string().string (),
593 "rn") != NULL))) {
594 if (tessedit_process_rns
595 && strstr (word->best_choice->unichar_string().string (),
596 "rn") != NULL) {
597 copy_outword = *(word->outword);
598 copy_blob_it.set_to_list (copy_outword.blob_list ());
599 i = 0;
600 while (word->best_choice->unichar_string()[i] != '\0') {
601 if (word->best_choice->unichar_string()[i] == 'r'
602 && word->best_choice->unichar_string()[i + 1] == 'n') {
603 copy_outline_it.set_to_list (copy_blob_it.data ()->
604 out_list ());
605 copy_outline_it.add_list_after (copy_blob_it.
606 data_relative (1)->
607 out_list ());
608 copy_blob_it.forward ();
609 delete (copy_blob_it.extract ());
610 i++;
611 }
612 copy_blob_it.forward ();
613 i++;
614 }
615 }
616 else
617 copy_outword = *(word->outword);
618
619 copy_outword.baseline_denormalise (&word->denorm);
620 copy_blob_it.set_to_list (copy_outword.blob_list ());
621 char_clip_word(©_outword, page_image, pixrow_list, imlines, pix_box);
622 pixrow_it.set_to_list (pixrow_list);
623 pixrow_it.move_to_first ();
624
625 // For debugging only
626 b_box = copy_outword.bounding_box ();
627 pixrow = pixrow_it.data ();
628
629 blob_it.move_to_first ();
630 copy_blob_it.move_to_first ();
631 for (i = 0;
632 word->best_choice->unichar_string()[i] != '\0';
633 i++, pixrow_it.forward (), blob_it.forward (),
634 copy_blob_it.forward ()) {
635 if ((word->best_choice->unichar_string()[i] == 'm'
636 || (word->best_choice->unichar_string()[i] == 'r'
637 && word->best_choice->unichar_string()[i + 1] == 'n'))
638 && !word->reject_map[i].perm_rejected ()) {
639 if (tessedit_cluster_debug)
640 tprintf ("Sample %c to check found in %s, index %d\n",
641 word->best_choice->unichar_string()[i],
642 word->best_choice->unichar_string().string (), i);
643
644 if (tessedit_demo_adaption)
645 tprintf
646 ("Sample %c to check found in %s (%d), index %d\n",
647 word->best_choice->unichar_string()[i],
648 word->best_choice->unichar_string().string (), word_number,
649 i);
650
651 if (tessedit_matrix_match) {
652 TBOX copy_box = copy_blob_it.data ()->bounding_box ();
653
654 sample = clip_sample (pixrow_it.data (),
655 imlines,
656 pix_box,
657 copy_outword.flag (W_INVERSE),
658 word->best_choice->unichar_string()[i]);
659
660 //Clip failed
661 if (sample == NULL) {
662 tprintf
663 ("Unable to clip sample from %s, index %d\n",
664 word->best_choice->unichar_string().string (), i);
665 #ifndef SECURE_NAMES
666 if (tessedit_cluster_debug)
667 tprintf ("Sample rejected (no sample)\n");
668 #endif
669 word->reject_map[i].setrej_mm_reject ();
670 if (word->best_choice->unichar_string()[i] == 'r') {
671 word->reject_map[i + 1].setrej_mm_reject ();
672 i++;
673 }
674 continue;
675 }
676 }
677 else
678 sample = new CHAR_SAMPLE(blob_it.data(),
679 &word->denorm,
680 word->best_choice->unichar_string()[i]);
681
682 best_score = MAX_INT32;
683 best_char = '\0';
684 best_cluster = NULL;
685
686 for (c_it.mark_cycle_pt ();
687 !c_it.cycled_list (); c_it.forward ()) {
688 if (c_it.data ()->character () != '\0') {
689 score = c_it.data ()->match_score (sample, this);
690 if (score < best_score) {
691 best_cluster = c_it.data ();
692 best_score = score;
693 best_char = c_it.data ()->character ();
694 }
695 }
696 }
697
698 if (best_score > tessedit_cluster_t1) {
699 #ifndef SECURE_NAMES
700 if (tessedit_cluster_debug)
701 tprintf ("Sample rejected (score %f)\n", best_score);
702 if (tessedit_demo_adaption)
703 tprintf ("Sample rejected (score %f)\n", best_score);
704 #endif
705 word->reject_map[i].setrej_mm_reject ();
706 if (word->best_choice->unichar_string()[i] == 'r')
707 word->reject_map[i + 1].setrej_mm_reject ();
708 }
709 else {
710 if (word->best_choice->unichar_string()[i] == best_char) {
711 #ifndef SECURE_NAMES
712 if (tessedit_cluster_debug)
713 tprintf ("Sample accepted (score %f)\n",
714 best_score);
715 if (tessedit_demo_adaption)
716 tprintf ("Sample accepted (score %f)\n",
717 best_score);
718 #endif
719 word->reject_map[i].setrej_mm_accept ();
720 if (word->best_choice->unichar_string()[i] == 'r')
721 word->reject_map[i + 1].setrej_mm_accept ();
722 }
723 else {
724 #ifndef SECURE_NAMES
725 if (tessedit_cluster_debug)
726 tprintf ("Sample rejected (char %c, score %f)\n",
727 best_char, best_score);
728 if (tessedit_demo_adaption)
729 tprintf ("Sample rejected (char %c, score %f)\n",
730 best_char, best_score);
731 #endif
732 word->reject_map[i].setrej_mm_reject ();
733 if (word->best_choice->unichar_string()[i] == 'r')
734 word->reject_map[i + 1].setrej_mm_reject ();
735 }
736 }
737
738 if (tessedit_demo_adaption) {
739 if (strcmp (imagebasename.string (),
740 tessedit_demo_file.string ()) != 0
741 || word_number == tessedit_demo_word1
742 || word_number == tessedit_demo_word2) {
743 #ifndef GRAPHICS_DISABLED
744 demo_win =
745 display_clip_image(©_outword,
746 page_image,
747 pixrow_list,
748 pix_box);
749 #endif
750 demo_word = word_number;
751 best_cluster->match_score (sample, this);
752 demo_word = 0;
753 }
754 }
755 if (word->best_choice->unichar_string()[i] == 'r')
756 i++; // Skip next character
757 }
758 }
759 delete[]imlines; // Free array of imlines
760 delete pixrow_list;
761 }
762 }
763 }
764
765
766
adapt_to_good_samples(WERD_RES * word,CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)767 void Tesseract::adapt_to_good_samples(WERD_RES *word,
768 CHAR_SAMPLES_LIST *char_clusters,
769 CHAR_SAMPLE_LIST *chars_waiting) {
770 PBLOB_LIST *blobs = word->outword->blob_list ();
771 PBLOB_IT blob_it(blobs);
772 inT16 i;
773 CHAR_SAMPLE *sample;
774 CHAR_SAMPLES_IT c_it = char_clusters;
775 CHAR_SAMPLE_IT cw_it = chars_waiting;
776 float score;
777 float best_score;
778 char best_char;
779 CHAR_SAMPLES *best_cluster;
780 PIXROW_LIST *pixrow_list;
781 PIXROW_IT pixrow_it;
782 IMAGELINE *imlines; // lines of the image
783 TBOX pix_box; // box of imlines
784 // extent
785 WERD copy_outword; // copy to denorm
786 TBOX b_box;
787 PBLOB_IT copy_blob_it;
788 PIXROW *pixrow = NULL;
789
790 static inT32 word_number = 0;
791
792 #ifndef GRAPHICS_DISABLED
793 ScrollView* demo_win = NULL;
794 #endif
795
796 inT32 resolution = page_image.get_res ();
797
798 word_number++;
799
800 if (tessedit_test_cluster_input)
801 return;
802
803 if (word->word->bounding_box ().height () > resolution / 3)
804 return;
805
806 if (char_clusters->length () == 0) {
807 #ifndef SECURE_NAMES
808 if (tessedit_cluster_debug)
809 tprintf ("No clusters to use for adaption\n");
810 #endif
811 return;
812 }
813
814 if (!cw_it.empty ()) {
815 complete_clustering(char_clusters, chars_waiting);
816 print_em_stats(char_clusters, chars_waiting);
817 }
818
819 if ((!word_adaptable (word, tessedit_cluster_adaption_mode)
820 && word->reject_map.reject_count () != 0) || tessedit_mm_use_rejmap) {
821 if (tessedit_cluster_debug) {
822 tprintf ("\nChecking: \"%s\" MAP ",
823 word->best_choice->unichar_string().string ());
824 word->reject_map.print (debug_fp);
825 tprintf ("\n");
826 }
827
828 copy_outword = *(word->outword);
829 copy_outword.baseline_denormalise (&word->denorm);
830 copy_blob_it.set_to_list (copy_outword.blob_list ());
831 char_clip_word(©_outword, page_image, pixrow_list, imlines, pix_box);
832 pixrow_it.set_to_list (pixrow_list);
833 pixrow_it.move_to_first ();
834
835 // For debugging only
836 b_box = copy_outword.bounding_box ();
837 pixrow = pixrow_it.data ();
838
839 blob_it.move_to_first ();
840 copy_blob_it.move_to_first ();
841 for (i = 0;
842 word->best_choice->unichar_string()[i] != '\0';
843 i++, pixrow_it.forward (), blob_it.forward (),
844 copy_blob_it.forward ()) {
845 if (word->reject_map[i].recoverable ()
846 || (tessedit_mm_all_rejects && word->reject_map[i].rejected ())) {
847 TBOX copy_box = copy_blob_it.data ()->bounding_box ();
848
849 if (tessedit_cluster_debug)
850 tprintf ("Sample %c to check found in %s, index %d\n",
851 word->best_choice->unichar_string()[i],
852 word->best_choice->unichar_string().string (), i);
853
854 if (tessedit_demo_adaption)
855 tprintf ("Sample %c to check found in %s (%d), index %d\n",
856 word->best_choice->unichar_string()[i],
857 word->best_choice->unichar_string().string (),
858 word_number, i);
859
860 sample = clip_sample (pixrow_it.data (),
861 imlines,
862 pix_box,
863 copy_outword.flag (W_INVERSE),
864 word->best_choice->unichar_string()[i]);
865
866 if (sample == NULL) { //Clip failed
867 tprintf ("Unable to clip sample from %s, index %d\n",
868 word->best_choice->unichar_string().string (), i);
869 #ifndef SECURE_NAMES
870 if (tessedit_cluster_debug)
871 tprintf ("Sample rejected (no sample)\n");
872 #endif
873 word->reject_map[i].setrej_mm_reject ();
874
875 continue;
876 }
877
878 best_score = MAX_INT32;
879 best_char = '\0';
880 best_cluster = NULL;
881
882 for (c_it.mark_cycle_pt ();
883 !c_it.cycled_list (); c_it.forward ()) {
884 if (c_it.data ()->character () != '\0') {
885 score = c_it.data ()->match_score (sample, this);
886 if (score < best_score) {
887 best_cluster = c_it.data ();
888 best_score = score;
889 best_char = c_it.data ()->character ();
890 }
891 }
892 }
893
894 if (best_score > tessedit_cluster_t1) {
895 #ifndef SECURE_NAMES
896 if (tessedit_cluster_debug)
897 tprintf ("Sample rejected (score %f)\n", best_score);
898 if (tessedit_demo_adaption)
899 tprintf ("Sample rejected (score %f)\n", best_score);
900 #endif
901 word->reject_map[i].setrej_mm_reject ();
902 }
903 else {
904 if (word->best_choice->unichar_string()[i] == best_char) {
905 #ifndef SECURE_NAMES
906 if (tessedit_cluster_debug)
907 tprintf ("Sample accepted (score %f)\n", best_score);
908 if (tessedit_demo_adaption)
909 tprintf ("Sample accepted (score %f)\n", best_score);
910 #endif
911 if (tessedit_test_adaption)
912 word->reject_map[i].setrej_minimal_rej_accept ();
913 else
914 word->reject_map[i].setrej_mm_accept ();
915 }
916 else {
917 #ifndef SECURE_NAMES
918 if (tessedit_cluster_debug)
919 tprintf ("Sample rejected (char %c, score %f)\n",
920 best_char, best_score);
921 if (tessedit_demo_adaption)
922 tprintf ("Sample rejected (char %c, score %f)\n",
923 best_char, best_score);
924 #endif
925 word->reject_map[i].setrej_mm_reject ();
926 }
927 }
928
929 if (tessedit_demo_adaption) {
930 if (strcmp (imagebasename.string (),
931 tessedit_demo_file.string ()) != 0
932 || word_number == tessedit_demo_word1
933 || word_number == tessedit_demo_word2) {
934 #ifndef GRAPHICS_DISABLED
935 demo_win =
936 display_clip_image(©_outword,
937 page_image,
938 pixrow_list,
939 pix_box);
940 #endif
941 demo_word = word_number;
942 best_cluster->match_score (sample, this);
943 demo_word = 0;
944 }
945 }
946 }
947 }
948 delete[]imlines; // Free array of imlines
949 delete pixrow_list;
950
951 if (tessedit_cluster_debug) {
952 tprintf ("\nFinal: \"%s\" MAP ",
953 word->best_choice->unichar_string().string ());
954 word->reject_map.print (debug_fp);
955 tprintf ("\n");
956 }
957 }
958 }
959 } // namespace tesseract
960
961
print_em_stats(CHAR_SAMPLES_LIST * char_clusters,CHAR_SAMPLE_LIST * chars_waiting)962 void print_em_stats(CHAR_SAMPLES_LIST *char_clusters,
963 CHAR_SAMPLE_LIST *chars_waiting) {
964 CHAR_SAMPLES_IT c_it = char_clusters;
965
966 if (!tessedit_cluster_debug)
967 return;
968 #ifndef SECURE_NAMES
969 tprintf ("There are %d clusters and %d samples waiting\n",
970 char_clusters->length (), chars_waiting->length ());
971
972 for (c_it.mark_cycle_pt (); !c_it.cycled_list (); c_it.forward ())
973 c_it.data ()->print (debug_fp);
974 #endif
975 tprintf ("\n");
976 }
977
978
clip_sample(PIXROW * pixrow,IMAGELINE * imlines,TBOX pix_box,BOOL8 white_on_black,char c)979 CHAR_SAMPLE *clip_sample( //lines of the image
980 PIXROW *pixrow,
981 IMAGELINE *imlines,
982 TBOX pix_box, //box of imlines extent
983 BOOL8 white_on_black,
984 char c) {
985 TBOX b_box = pixrow->bounding_box ();
986 float baseline_pos = 0;
987 inT32 resolution = page_image.get_res ();
988
989 if (!b_box.null_box ()) {
990 ASSERT_HOST (b_box.width () < page_image.get_xsize () &&
991 b_box.height () < page_image.get_ysize ());
992
993 if (b_box.width () > resolution || b_box.height () > resolution) {
994 tprintf ("clip sample: sample too big (%d x %d)\n",
995 b_box.width (), b_box.height ());
996
997 return NULL;
998 }
999
1000 IMAGE *image = new (IMAGE);
1001 if (image->create (b_box.width (), b_box.height (), 1) == -1) {
1002 tprintf ("clip sample: create image failed (%d x %d)\n",
1003 b_box.width (), b_box.height ());
1004
1005 delete image;
1006 return NULL;
1007 }
1008
1009 if (!white_on_black)
1010 invert_image(image); // Set background to white
1011 pixrow->char_clip_image (imlines, pix_box, NULL, *image, baseline_pos);
1012 if (white_on_black)
1013 invert_image(image); //invert white on black for scaling &NN
1014 return new CHAR_SAMPLE (image, c);
1015 }
1016 else
1017 return NULL;
1018 }
1019
1020
1021 #ifndef GRAPHICS_DISABLED
display_cluster_prototypes(CHAR_SAMPLES_LIST * char_clusters)1022 void display_cluster_prototypes(CHAR_SAMPLES_LIST *char_clusters) {
1023 inT16 proto_number = 0;
1024 CHAR_SAMPLES_IT c_it = char_clusters;
1025 char title[WINDOWNAMESIZE];
1026
1027 for (c_it.mark_cycle_pt (); !c_it.cycled_list (); c_it.forward ()) {
1028 proto_number++;
1029
1030 #ifndef SECURE_NAMES
1031 tprintf ("Displaying proto number %d\n", proto_number);
1032 #endif
1033
1034 if (c_it.data ()->prototype () != NULL) {
1035 sprintf (title, "Proto - %d", proto_number);
1036 display_image (c_it.data ()->prototype ()->make_image (),
1037 title, (proto_number - 1) * 400, 0, FALSE);
1038 }
1039 }
1040 }
1041 #endif
1042
1043 // *********************************************************************
1044 // Simplistic routines to test the effect of rejecting ems and fullstops
1045 // *********************************************************************
1046
reject_all_ems(WERD_RES * word)1047 void reject_all_ems(WERD_RES *word) {
1048 inT16 i;
1049
1050 for (i = 0; word->best_choice->unichar_string()[i] != '\0'; i++) {
1051 if (word->best_choice->unichar_string()[i] == 'm')
1052 // reject all ems
1053 word->reject_map[i].setrej_mm_reject ();
1054 }
1055 }
1056
1057
reject_all_fullstops(WERD_RES * word)1058 void reject_all_fullstops(WERD_RES *word) {
1059 inT16 i;
1060
1061 for (i = 0; word->best_choice->unichar_string()[i] != '\0'; i++) {
1062 if (word->best_choice->unichar_string()[i] == '.')
1063 // reject all fullstops
1064 word->reject_map[i].setrej_mm_reject ();
1065 }
1066 }
1067
1068 namespace tesseract {
reject_suspect_ems(WERD_RES * word)1069 void Tesseract::reject_suspect_ems(WERD_RES *word) {
1070 inT16 i;
1071
1072 if (!word_adaptable (word, tessedit_cluster_adaption_mode))
1073 for (i = 0; word->best_choice->unichar_string()[i] != '\0'; i++) {
1074 if (word->best_choice->unichar_string()[i] == 'm' && suspect_em (word, i))
1075 // reject all ems
1076 word->reject_map[i].setrej_mm_reject ();
1077 }
1078 }
1079 } // namespace tesseract
1080
1081
reject_suspect_fullstops(WERD_RES * word)1082 void reject_suspect_fullstops(WERD_RES *word) {
1083 inT16 i;
1084
1085 for (i = 0; word->best_choice->unichar_string()[i] != '\0'; i++) {
1086 if (word->best_choice->unichar_string()[i] == '.'
1087 && suspect_fullstop (word, i))
1088 // reject all commas
1089 word->reject_map[i].setrej_mm_reject ();
1090 }
1091 }
1092
1093
suspect_em(WERD_RES * word,inT16 index)1094 BOOL8 suspect_em(WERD_RES *word, inT16 index) {
1095 PBLOB_LIST *blobs = word->outword->blob_list ();
1096 PBLOB_IT blob_it(blobs);
1097 inT16 j;
1098
1099 for (j = 0; j < index; j++)
1100 blob_it.forward ();
1101
1102 return (blob_it.data ()->out_list ()->length () != 1);
1103 }
1104
1105
suspect_fullstop(WERD_RES * word,inT16 i)1106 BOOL8 suspect_fullstop(WERD_RES *word, inT16 i) {
1107 float aspect_ratio;
1108 PBLOB_LIST *blobs = word->outword->blob_list ();
1109 PBLOB_IT blob_it(blobs);
1110 inT16 j;
1111 TBOX box;
1112 inT16 width;
1113 inT16 height;
1114
1115 for (j = 0; j < i; j++)
1116 blob_it.forward ();
1117
1118 box = blob_it.data ()->bounding_box ();
1119
1120 width = box.width ();
1121 height = box.height ();
1122
1123 aspect_ratio = ((width > height) ? ((float) width) / height :
1124 ((float) height) / width);
1125
1126 return (aspect_ratio > tessed_fullstop_aspect_ratio);
1127 }
1128