38 #include "config_auto.h"
43 #include "allheaders.h"
54 "Take segmentation and labeling from box file",
56 BOOL_MEMBER(tessedit_resegment_from_line_boxes, false,
57 "Conversion of word/line box file to char box file",
60 "Generate training data from boxed chars", this->params()),
62 "Generate more boxes from boxed chars", this->params()),
64 "Break input into lines and remap boxes if present",
67 "Dump intermediate images made during page segmentation",
70 "Try inverting the image in `LSTMRecognizeWord`", this->params()),
75 "Page seg mode: 0=osd only, 1=auto+osd, 2=auto_only, 3=auto, 4=column,"
76 " 5=block_vert, 6=block, 7=line, 8=word, 9=word_circle, 10=char,"
77 "11=sparse_text, 12=sparse_text+osd, 13=raw_line"
78 " (Values from PageSegMode enum in tesseract/publictypes.h)",
81 "Which OCR engine(s) to run (Tesseract, LSTM, both)."
82 " Defaults to loading and running the most accurate"
86 "Blacklist of chars not to recognize", this->params()),
88 "Whitelist of chars to recognize", this->params()),
90 "List of chars to override tessedit_char_blacklist",
93 "Perform training for ambiguities", this->params()),
96 "Whether to use the top-line splitting process for Devanagari "
97 "documents while performing page-segmentation.",
101 "Whether to use the top-line splitting process for Devanagari "
102 "documents while performing ocr.",
105 "Write all parameters to the given file.", this->params()),
107 "Generate and print debug"
108 " information for adaption",
110 INT_MEMBER(bidi_debug, 0,
"Debug level for BiDi", this->params()),
111 INT_MEMBER(applybox_debug, 1,
"Debug level", this->params()),
112 INT_MEMBER(applybox_page, 0,
"Page number to apply boxes from",
115 "Exposure value follows"
116 " this pattern in the image filename. The name of the image"
117 " files are expected to be in the form"
118 " [lang].[fontname].exp[num].tif",
120 BOOL_MEMBER(applybox_learn_chars_and_char_frags_mode, false,
121 "Learn both character fragments (as is done in the"
122 " special low exposure mode) as well as unfragmented"
127 " is assumed to contain ngrams. Only learn the ngrams"
128 " whose outlines overlap horizontally.",
130 BOOL_MEMBER(tessedit_display_outwords, false,
"Draw output words",
132 BOOL_MEMBER(tessedit_dump_choices, false,
"Dump char choices",
134 BOOL_MEMBER(tessedit_timing_debug, false,
"Print timing stats",
137 "Try to improve fuzzy spaces", this->params()),
139 "Don't bother with word plausibility", this->params()),
140 BOOL_MEMBER(tessedit_fix_hyphens, true,
"Crunch double hyphens?",
143 "Add words to the document dictionary", this->params()),
144 BOOL_MEMBER(tessedit_debug_fonts, false,
"Output font info per char",
146 BOOL_MEMBER(tessedit_debug_block_rejection, false,
"Block and Row stats",
148 BOOL_MEMBER(tessedit_enable_bigram_correction, true,
149 "Enable correction based on the word bigram dictionary.",
151 BOOL_MEMBER(tessedit_enable_dict_correction, false,
152 "Enable single word correction based on the dictionary.",
155 "Amount of debug output for bigram correction.",
158 "Remove and conditionally reassign small outlines when they"
159 " confuse layout analysis, determining diacritics vs noise",
161 INT_MEMBER(debug_noise_removal, 0,
"Debug reassignment of small outlines",
167 "Hingepoint for base char certainty", this->params()),
171 "Hingepoint for disjoint certainty", this->params()),
175 "Threshold for new punc char certainty", this->params()),
178 "Scaling on certainty diff from Hingepoint",
180 INT_MEMBER(noise_maxperblob, 8,
"Max diacritics to apply to a blob",
182 INT_MEMBER(noise_maxperword, 16,
"Max diacritics to apply to a word",
184 INT_MEMBER(debug_x_ht_level, 0,
"Reestimate debug", this->params()),
185 STRING_MEMBER(chs_leading_punct,
"('`\"",
"Leading punctuation",
187 STRING_MEMBER(chs_trailing_punct1,
").,;:?!",
"1st Trailing punctuation",
189 STRING_MEMBER(chs_trailing_punct2,
")'`\"",
"2nd Trailing punctuation",
192 "good_quality_doc lte rejection limit", this->params()),
194 "good_quality_doc gte good blobs limit", this->params()),
196 "good_quality_doc lte outline error limit", this->params()),
198 "good_quality_doc gte good char limit", this->params()),
199 INT_MEMBER(quality_min_initial_alphas_reqd, 2,
"alphas in a good word",
202 "Adaptation decision algorithm for tess", this->params()),
204 "Do minimal rejection on pass 1 output", this->params()),
205 BOOL_MEMBER(tessedit_test_adaption, false,
"Test adaption criteria",
207 BOOL_MEMBER(test_pt, false,
"Test for point", this->params()),
208 double_MEMBER(test_pt_x, 99999.99,
"xcoord", this->params()),
209 double_MEMBER(test_pt_y, 99999.99,
"ycoord", this->params()),
210 INT_MEMBER(multilang_debug_level, 0,
"Print multilang debug info.",
212 INT_MEMBER(paragraph_debug_level, 0,
"Print paragraph debug info.",
215 "Run paragraph detection on the post-text-recognition "
219 "Use ratings matrix/beam search with lstm", this->params()),
220 STRING_MEMBER(outlines_odd,
"%| ",
"Non standard number of outlines",
222 STRING_MEMBER(outlines_2,
"ij!?%\":;",
"Non standard number of outlines",
225 "Reduce rejection on good docs", this->params()),
226 BOOL_MEMBER(tessedit_use_reject_spaces, true,
"Reject spaces?",
229 "%rej allowed before rej whole doc", this->params()),
231 "%rej allowed before rej whole block", this->params()),
233 "%rej allowed before rej whole row", this->params()),
235 "Number of row rejects in whole word rejects"
236 " which prevents whole row rejection",
238 BOOL_MEMBER(tessedit_preserve_blk_rej_perfect_wds, true,
239 "Only rej partially rejected words in block rejection",
241 BOOL_MEMBER(tessedit_preserve_row_rej_perfect_wds, true,
242 "Only rej partially rejected words in row rejection",
245 "Use word segmentation quality metric", this->params()),
247 "Use word segmentation quality metric", this->params()),
249 "Only preserve wds longer than this", this->params()),
251 "Apply row rejection to good docs", this->params()),
253 "rej good doc wd if more than this fraction rejected",
256 "Reject all bad quality wds", this->params()),
257 BOOL_MEMBER(tessedit_debug_doc_rejection, false,
"Page stats",
260 "Output data to debug file", this->params()),
261 BOOL_MEMBER(bland_unrej, false,
"unrej potential with no checks",
264 "good_quality_doc gte good char limit", this->params()),
266 "Mark v.bad words for tilde crunch", this->params()),
267 BOOL_MEMBER(hocr_font_info, false,
"Add font info to hocr output",
269 BOOL_MEMBER(hocr_char_boxes, false,
"Add coordinates for each character to hocr output",
271 BOOL_MEMBER(crunch_early_merge_tess_fails, true,
"Before word crunch?",
273 BOOL_MEMBER(crunch_early_convert_bad_unlv_chs, false,
274 "Take out ~^ early?", this->params()),
275 double_MEMBER(crunch_terrible_rating, 80.0,
"crunch rating lt this",
277 BOOL_MEMBER(crunch_terrible_garbage, true,
"As it says", this->params()),
279 "crunch garbage cert lt this", this->params()),
281 "crunch garbage rating lt this", this->params()),
282 double_MEMBER(crunch_pot_poor_rate, 40,
"POTENTIAL crunch rating lt this",
284 double_MEMBER(crunch_pot_poor_cert, -8.0,
"POTENTIAL crunch cert lt this",
286 double_MEMBER(crunch_del_rating, 60,
"POTENTIAL crunch rating lt this",
288 double_MEMBER(crunch_del_cert, -10.0,
"POTENTIAL crunch cert lt this",
290 double_MEMBER(crunch_del_min_ht, 0.7,
"Del if word ht lt xht x this",
292 double_MEMBER(crunch_del_max_ht, 3.0,
"Del if word ht gt xht x this",
295 "Del if word width lt xht x this", this->params()),
297 "Del if word gt xht x this above bl", this->params()),
299 "Del if word gt xht x this below bl", this->params()),
300 double_MEMBER(crunch_small_outlines_size, 0.6,
"Small if lt xht x this",
302 INT_MEMBER(crunch_rating_max, 10,
"For adj length in rating per ch",
305 "How many potential indicators needed", this->params()),
306 BOOL_MEMBER(crunch_leave_ok_strings, true,
"Don't touch sensible strings",
308 BOOL_MEMBER(crunch_accept_ok, true,
"Use acceptability in okstring",
311 "Don't pot crunch sensible strings", this->params()),
312 BOOL_MEMBER(crunch_include_numerals, false,
"Fiddle alpha figures",
315 "Don't crunch words with long lower case strings",
318 "Don't crunch words with long lower case strings",
321 "Crunch words with long repetitions", this->params()),
322 INT_MEMBER(crunch_debug, 0,
"As it says", this->params()),
324 "How many non-noise blbs either side?", this->params()),
325 double_MEMBER(fixsp_small_outlines_size, 0.28,
"Small if lt xht x this",
328 "Reward punctuation joins", this->params()),
329 INT_MEMBER(fixsp_done_mode, 1,
"What constitues done for spacing",
331 INT_MEMBER(debug_fix_space_level, 0,
"Contextual fixspace debug",
334 "Punct. chs expected WITHIN numbers", this->params()),
336 "Max allowed deviation of blob top outside of font data",
339 "Min change in xht before actually trying it", this->params()),
341 "Debug level for sub & superscript fixer", this->params()),
343 superscript_worse_certainty, 2.0,
344 "How many times worse "
345 "certainty does a superscript position glyph need to be for "
346 "us to try classifying it as a char with a different "
350 superscript_bettered_certainty, 0.97,
352 "badness do we think sufficient to choose a superscript "
353 "over what we'd thought. For example, a value of 0.6 means "
354 "we want to reduce badness of certainty by at least 40%",
357 "A superscript scaled down more than this is unbelievably "
358 "small. For example, 0.3 means we expect the font size to "
359 "be no smaller than 30% of the text line font size.",
362 "Maximum top of a character measured as a multiple of "
363 "x-height above the baseline for us to reconsider whether "
367 "Minimum bottom of a character measured as a multiple of "
368 "x-height above the baseline for us to reconsider whether "
369 "it's a superscript.",
371 BOOL_MEMBER(tessedit_write_block_separators, false,
372 "Write block separators in output", this->params()),
373 BOOL_MEMBER(tessedit_write_rep_codes, false,
"Write repetition char code",
375 BOOL_MEMBER(tessedit_write_unlv, false,
"Write .unlv output file",
377 BOOL_MEMBER(tessedit_create_txt, false,
"Write .txt output file",
379 BOOL_MEMBER(tessedit_create_hocr, false,
"Write .html hOCR output file",
381 BOOL_MEMBER(tessedit_create_alto, false,
"Write .xml ALTO file",
383 BOOL_MEMBER(tessedit_create_lstmbox, false,
"Write .box file for LSTM training",
385 BOOL_MEMBER(tessedit_create_tsv, false,
"Write .tsv output file",
387 BOOL_MEMBER(tessedit_create_wordstrbox, false,
"Write WordStr format .box output file",
389 BOOL_MEMBER(tessedit_create_pdf, false,
"Write .pdf output file",
392 "Create PDF with only one invisible text layer",
394 INT_MEMBER(jpg_quality, 85,
"Set JPEG quality level", this->params()),
395 INT_MEMBER(user_defined_dpi, 0,
"Specify DPI for input image",
398 "Specify minimum characters to try during OSD",
401 "Output char for unidentified blobs", this->params()),
402 INT_MEMBER(suspect_level, 99,
"Suspect marker level", this->params()),
404 "Don't suspect dict wds longer than this", this->params()),
405 BOOL_MEMBER(suspect_constrain_1Il, false,
"UNLV keep 1Il chars rejected",
408 "Don't touch bad rating limit", this->params()),
409 double_MEMBER(suspect_accept_rating, -999.9,
"Accept good rating limit",
412 "Only reject tess failures", this->params()),
413 BOOL_MEMBER(tessedit_zero_rejection, false,
"Don't reject ANYTHING",
416 "Make output have exactly one word per WERD", this->params()),
418 "Don't reject ANYTHING AT ALL", this->params()),
419 INT_MEMBER(tessedit_reject_mode, 0,
"Rejection algorithm",
421 BOOL_MEMBER(tessedit_rejection_debug, false,
"Adaption debug",
423 BOOL_MEMBER(tessedit_flip_0O, true,
"Contextual 0O O0 flips",
426 "Aspect ratio dot/hyphen test", this->params()),
428 "Aspect ratio dot/hyphen test", this->params()),
430 "Use DOC dawg in 11l conf. detector", this->params()),
431 BOOL_MEMBER(rej_1Il_use_dict_word, false,
"Use dictword test",
433 BOOL_MEMBER(rej_1Il_trust_permuter_type, true,
"Don't double check",
435 BOOL_MEMBER(rej_use_tess_accepted, true,
"Individual rejection control",
437 BOOL_MEMBER(rej_use_tess_blanks, true,
"Individual rejection control",
439 BOOL_MEMBER(rej_use_good_perm, true,
"Individual rejection control",
441 BOOL_MEMBER(rej_use_sensible_wd, false,
"Extend permuter check",
443 BOOL_MEMBER(rej_alphas_in_number_perm, false,
"Extend permuter check",
446 "if >this fract", this->params()),
447 INT_MEMBER(tessedit_image_border, 2,
"Rej blbs near image edge limit",
450 "Allow NN to unrej", this->params()),
451 STRING_MEMBER(conflict_set_I_l_1,
"Il1[]",
"Il1 conflict set",
453 INT_MEMBER(min_sane_x_ht_pixels, 8,
"Reject any x-ht lt or eq than this",
455 BOOL_MEMBER(tessedit_create_boxfile, false,
"Output text with boxes",
458 "-1 -> All pages, else specific page to process",
461 "Capture the image from the IPE", this->params()),
462 BOOL_MEMBER(interactive_display_mode, false,
"Run interactively?",
464 STRING_MEMBER(file_type,
".tif",
"Filename extension", this->params()),
465 BOOL_MEMBER(tessedit_override_permuter, true,
"According to dict_word",
468 "List of languages to load with this one", this->params()),
469 BOOL_MEMBER(tessedit_use_primary_params_model, false,
470 "In multilingual mode use params model of the"
474 "Min acceptable orientation margin", this->params()),
475 BOOL_MEMBER(textord_tabfind_show_vlines, false,
"Debug line finding",
477 BOOL_MEMBER(textord_use_cjk_fp_model, false,
"Use CJK fixed pitch model",
480 "Allow feature extractors to see the original outline",
483 "Only initialize with the config file. Useful if the "
484 "instance is not going to be used for OCR but say only "
485 "for layout analysis.",
487 BOOL_MEMBER(textord_equation_detect, false,
"Turn on equation detector",
490 "Enable vertical detection", this->params()),
491 BOOL_MEMBER(textord_tabfind_force_vertical_text, false,
492 "Force using vertical text page mode", this->params()),
494 textord_tabfind_vertical_text_ratio, 0.5,
495 "Fraction of textlines deemed vertical to use vertical page "
499 textord_tabfind_aligned_gap_fraction, 0.75,
500 "Fraction of height used as a minimum gap for aligned blobs.",
502 INT_MEMBER(tessedit_parallelize, 0,
"Run in parallel where possible",
505 "Preserve multiple interword spaces", this->params()),
507 "Page separator (default is form feed control character)",
510 "Allows to include alternative symbols choices in the hOCR output. "
511 "Valid input values are 0, 1 and 2. 0 is the default value. "
512 "With 1 the alternative symbol choices per timestep are included. "
513 "With 2 alternative symbol choices are extracted from the CTC "
514 "process instead of the lattice. The choices are mapped per "
518 lstm_choice_iterations, 5,
519 "Sets the number of cascading iterations for the Beamsearch in "
520 "lstm_choice_mode. Note that lstm_choice_mode must be set to a "
521 "value greater than 0 to produce results.",
524 lstm_rating_coefficient, 5,
525 "Sets the rating coefficient for the lstm choices. The smaller the "
526 "coefficient, the better are the ratings for each choice and less "
527 "information is lost due to the cut off at 0. The standard value is "
528 "5", this->params()),
530 "Detect music staff and remove intersecting components", this->params()),
532 backup_config_file_(nullptr),
533 pix_binary_(nullptr),
535 pix_original_(nullptr),
536 pix_thresholds_(nullptr),
537 source_resolution_(0),
539 right_to_left_(false),
540 scaled_color_(nullptr),
544 most_recently_used_(this),
546 equ_detect_(nullptr),
547 #ifndef ANDROID_BUILD
548 lstm_recognizer_(nullptr),
550 train_line_page_num_(0) {
555 pixDestroy(&pix_original_);
557 sub_langs_.delete_data_pointers();
558 #ifndef ANDROID_BUILD
559 delete lstm_recognizer_;
560 lstm_recognizer_ =
nullptr;
566 if (lstm_recognizer_ && lstm_recognizer_->
GetDict()) {
567 return *lstm_recognizer_->
GetDict();
577 pixDestroy(&pix_binary_);
578 pixDestroy(&pix_grey_);
579 pixDestroy(&pix_thresholds_);
580 pixDestroy(&scaled_color_);
581 deskew_ =
FCOORD(1.0f, 0.0f);
582 reskew_ =
FCOORD(1.0f, 0.0f);
585 for (
int i = 0; i < sub_langs_.size(); ++i)
586 sub_langs_[i]->
Clear();
589 #ifndef DISABLED_LEGACY_ENGINE
592 equ_detect_ = detector;
599 for (
int i = 0; i < sub_langs_.size(); ++i) {
600 sub_langs_[i]->ResetAdaptiveClassifierInternal();
604 #endif //ndef DISABLED_LEGACY_ENGINE
609 for (
int i = 0; i < sub_langs_.size(); ++i) {
610 sub_langs_[i]->getDict().ResetDocumentDictionary();
619 if (lstm_recognizer_) {
626 for (
int i = 0; i < sub_langs_.size(); ++i) {
627 sub_langs_[i]->unicharset.set_black_and_whitelist(
630 if (sub_langs_[i]->lstm_recognizer_) {
631 UNICHARSET& lstm_unicharset = sub_langs_[i]->lstm_recognizer_->GetUnicharset();
644 auto max_pageseg_strategy =
645 static_cast<ShiroRekhaSplitter::SplitStrategy>(
647 for (
int i = 0; i < sub_langs_.size(); ++i) {
648 auto pageseg_strategy =
649 static_cast<ShiroRekhaSplitter::SplitStrategy>(
651 if (pageseg_strategy > max_pageseg_strategy)
652 max_pageseg_strategy = pageseg_strategy;
653 pixDestroy(&sub_langs_[i]->pix_binary_);
654 sub_langs_[i]->pix_binary_ = pixClone(
pix_binary());
660 if (splitter_.
Split(
true, &pixa_debug_)) {
662 pixDestroy(&pix_binary_);
675 auto max_ocr_strategy =
676 static_cast<ShiroRekhaSplitter::SplitStrategy>(
678 for (
int i = 0; i < sub_langs_.size(); ++i) {
680 static_cast<ShiroRekhaSplitter::SplitStrategy>(
682 if (ocr_strategy > max_ocr_strategy)
683 max_ocr_strategy = ocr_strategy;
689 bool split_for_ocr = splitter_.
Split(
false, &pixa_debug_);
692 pixDestroy(&pix_binary_);
693 pix_binary_ = pixClone(splitter_.
orig_pix());
698 BLOCK block(
"",
true, 0, 0, 0, 0, pixGetWidth(pix_binary_),
699 pixGetHeight(pix_binary_));