Mercurial > hgrepos > Python2 > PyMuPDF

comparison mupdf-source/thirdparty/tesseract/src/classify/classify.h @ 2:b50eed0cc0ef upstream

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
ADD: MuPDF v1.26.7: the MuPDF source as downloaded by a default build of PyMuPDF 1.26.4. The directory name has changed: no version number in the expanded directory now.
author Franz Glasner <fzglas.hg@dom66.de>
date Mon, 15 Sep 2025 11:43:07 +0200
parents
children
comparison
equal deleted inserted replaced
1:1d09e1dec1d9 2:b50eed0cc0ef
1 ///////////////////////////////////////////////////////////////////////
2 // File: classify.h
3 // Description: classify class.
4 // Author: Samuel Charron
5 //
6 // (C) Copyright 2006, Google Inc.
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 #ifndef TESSERACT_CLASSIFY_CLASSIFY_H_
20 #define TESSERACT_CLASSIFY_CLASSIFY_H_
21
22 // Include automatically generated configuration file if running autoconf.
23 #ifdef HAVE_CONFIG_H
24 # include "config_auto.h"
25 #endif
26
27 #ifdef DISABLED_LEGACY_ENGINE
28
29 # include "ccstruct.h"
30 # include "dict.h"
31
32 namespace tesseract {
33
34 class Classify : public CCStruct {
35 public:
36 Classify();
37 virtual ~Classify();
38 virtual Dict &getDict() {
39 return dict_;
40 }
41
42 // Member variables.
43
44 INT_VAR_H(classify_debug_level);
45 BOOL_VAR_H(classify_bln_numeric_mode);
46 double_VAR_H(classify_max_rating_ratio);
47 double_VAR_H(classify_max_certainty_margin);
48
49 private:
50 Dict dict_;
51 };
52
53 } // namespace tesseract
54
55 #else // DISABLED_LEGACY_ENGINE not defined
56
57 # include "adaptive.h"
58 # include "ccstruct.h"
59 # include "dict.h"
60 # include "featdefs.h"
61 # include "fontinfo.h"
62 # include "intfx.h"
63 # include "intmatcher.h"
64 # include "normalis.h"
65 # include "ocrfeatures.h"
66 # include "ratngs.h"
67 # include "unicity_table.h"
68
69 namespace tesseract {
70
71 class ScrollView;
72 class WERD_CHOICE;
73 class WERD_RES;
74 struct ADAPT_RESULTS;
75 struct NORM_PROTOS;
76
77 static const int kUnknownFontinfoId = -1;
78 static const int kBlankFontinfoId = -2;
79
80 class ShapeClassifier;
81 struct ShapeRating;
82 class ShapeTable;
83 struct UnicharRating;
84
85 // How segmented is a blob. In this enum, character refers to a classifiable
86 // unit, but that is too long and character is usually easier to understand.
87 enum CharSegmentationType {
88 CST_FRAGMENT, // A partial character.
89 CST_WHOLE, // A correctly segmented character.
90 CST_IMPROPER, // More than one but less than 2 characters.
91 CST_NGRAM // Multiple characters.
92 };
93
94 class TESS_API Classify : public CCStruct {
95 public:
96 Classify();
97 ~Classify() override;
98 virtual Dict &getDict() {
99 return dict_;
100 }
101
102 const ShapeTable *shape_table() const {
103 return shape_table_;
104 }
105
106 // Takes ownership of the given classifier, and uses it for future calls
107 // to CharNormClassifier.
108 void SetStaticClassifier(ShapeClassifier *static_classifier);
109
110 // Adds a noise classification result that is a bit worse than the worst
111 // current result, or the worst possible result if no current results.
112 void AddLargeSpeckleTo(int blob_length, BLOB_CHOICE_LIST *choices);
113
114 // Returns true if the blob is small enough to be a large speckle.
115 bool LargeSpeckle(const TBLOB &blob);
116
117 /* adaptive.cpp ************************************************************/
118 int GetFontinfoId(ADAPT_CLASS_STRUCT *Class, uint8_t ConfigId);
119 // Runs the class pruner from int_templates on the given features, returning
120 // the number of classes output in results.
121 // int_templates Class pruner tables
122 // num_features Number of features in blob
123 // features Array of features
124 // normalization_factors (input) Array of int_templates->NumClasses fudge
125 // factors from blob normalization process.
126 // (Indexed by CLASS_INDEX)
127 // expected_num_features (input) Array of int_templates->NumClasses
128 // expected number of features for each class.
129 // (Indexed by CLASS_INDEX)
130 // results (output) Sorted Array of pruned classes.
131 // Array must be sized to take the maximum possible
132 // number of outputs : int_templates->NumClasses.
133 int PruneClasses(const INT_TEMPLATES_STRUCT *int_templates, int num_features, int keep_this,
134 const INT_FEATURE_STRUCT *features, const uint8_t *normalization_factors,
135 const uint16_t *expected_num_features, std::vector<CP_RESULT_STRUCT> *results);
136 void ReadNewCutoffs(TFile *fp, uint16_t *Cutoffs);
137 void PrintAdaptedTemplates(FILE *File, ADAPT_TEMPLATES_STRUCT *Templates);
138 void WriteAdaptedTemplates(FILE *File, ADAPT_TEMPLATES_STRUCT *Templates);
139 ADAPT_TEMPLATES_STRUCT *ReadAdaptedTemplates(TFile *File);
140 /* normmatch.cpp ************************************************************/
141 float ComputeNormMatch(CLASS_ID ClassId, const FEATURE_STRUCT &feature, bool DebugMatch);
142 void FreeNormProtos();
143 NORM_PROTOS *ReadNormProtos(TFile *fp);
144 /* protos.cpp ***************************************************************/
145 void ConvertProto(PROTO_STRUCT *Proto, int ProtoId, INT_CLASS_STRUCT *Class);
146 INT_TEMPLATES_STRUCT *CreateIntTemplates(CLASSES FloatProtos, const UNICHARSET &target_unicharset);
147 /* adaptmatch.cpp ***********************************************************/
148
149 // Learns the given word using its chopped_word, seam_array, denorm,
150 // box_word, best_state, and correct_text to learn both correctly and
151 // incorrectly segmented blobs. If fontname is not nullptr, then LearnBlob
152 // is called and the data will be saved in an internal buffer.
153 // Otherwise AdaptToBlob is called for adaption within a document.
154 void LearnWord(const char *fontname, WERD_RES *word);
155
156 // Builds a blob of length fragments, from the word, starting at start,
157 // and then learns it, as having the given correct_text.
158 // If fontname is not nullptr, then LearnBlob is called and the data will be
159 // saved in an internal buffer for static training.
160 // Otherwise AdaptToBlob is called for adaption within a document.
161 // threshold is a magic number required by AdaptToChar and generated by
162 // ComputeAdaptionThresholds.
163 // Although it can be partly inferred from the string, segmentation is
164 // provided to explicitly clarify the character segmentation.
165 void LearnPieces(const char *fontname, int start, int length, float threshold,
166 CharSegmentationType segmentation, const char *correct_text, WERD_RES *word);
167 void InitAdaptiveClassifier(TessdataManager *mgr);
168 void InitAdaptedClass(TBLOB *Blob, CLASS_ID ClassId, int FontinfoId, ADAPT_CLASS_STRUCT *Class,
169 ADAPT_TEMPLATES_STRUCT *Templates);
170 void AmbigClassifier(const std::vector<INT_FEATURE_STRUCT> &int_features,
171 const INT_FX_RESULT_STRUCT &fx_info, const TBLOB *blob,
172 INT_TEMPLATES_STRUCT *templates, ADAPT_CLASS_STRUCT **classes, UNICHAR_ID *ambiguities,
173 ADAPT_RESULTS *results);
174 void MasterMatcher(INT_TEMPLATES_STRUCT *templates, int16_t num_features,
175 const INT_FEATURE_STRUCT *features, const uint8_t *norm_factors,
176 ADAPT_CLASS_STRUCT **classes, int debug, int matcher_multiplier, const TBOX &blob_box,
177 const std::vector<CP_RESULT_STRUCT> &results, ADAPT_RESULTS *final_results);
178 // Converts configs to fonts, and if the result is not adapted, and a
179 // shape_table_ is present, the shape is expanded to include all
180 // unichar_ids represented, before applying a set of corrections to the
181 // distance rating in int_result, (see ComputeCorrectedRating.)
182 // The results are added to the final_results output.
183 void ExpandShapesAndApplyCorrections(ADAPT_CLASS_STRUCT **classes, bool debug, int class_id, int bottom,
184 int top, float cp_rating, int blob_length,
185 int matcher_multiplier, const uint8_t *cn_factors,
186 UnicharRating *int_result, ADAPT_RESULTS *final_results);
187 // Applies a set of corrections to the distance im_rating,
188 // including the cn_correction, miss penalty and additional penalty
189 // for non-alnums being vertical misfits. Returns the corrected distance.
190 double ComputeCorrectedRating(bool debug, int unichar_id, double cp_rating, double im_rating,
191 int feature_misses, int bottom, int top, int blob_length,
192 int matcher_multiplier, const uint8_t *cn_factors);
193 void ConvertMatchesToChoices(const DENORM &denorm, const TBOX &box, ADAPT_RESULTS *Results,
194 BLOB_CHOICE_LIST *Choices);
195 void AddNewResult(const UnicharRating &new_result, ADAPT_RESULTS *results);
196 int GetAdaptiveFeatures(TBLOB *Blob, INT_FEATURE_ARRAY IntFeatures, FEATURE_SET *FloatFeatures);
197
198 # ifndef GRAPHICS_DISABLED
199 void DebugAdaptiveClassifier(TBLOB *Blob, ADAPT_RESULTS *Results);
200 # endif
201 PROTO_ID MakeNewTempProtos(FEATURE_SET Features, int NumBadFeat, FEATURE_ID BadFeat[],
202 INT_CLASS_STRUCT *IClass, ADAPT_CLASS_STRUCT *Class, BIT_VECTOR TempProtoMask);
203 int MakeNewTemporaryConfig(ADAPT_TEMPLATES_STRUCT *Templates, CLASS_ID ClassId, int FontinfoId,
204 int NumFeatures, INT_FEATURE_ARRAY Features,
205 FEATURE_SET FloatFeatures);
206 void MakePermanent(ADAPT_TEMPLATES_STRUCT *Templates, CLASS_ID ClassId, int ConfigId, TBLOB *Blob);
207 void PrintAdaptiveMatchResults(const ADAPT_RESULTS &results);
208 void RemoveExtraPuncs(ADAPT_RESULTS *Results);
209 void RemoveBadMatches(ADAPT_RESULTS *Results);
210 void SetAdaptiveThreshold(float Threshold);
211 void ShowBestMatchFor(int shape_id, const INT_FEATURE_STRUCT *features, int num_features);
212 // Returns a string for the classifier class_id: either the corresponding
213 // unicharset debug_str or the shape_table_ debug str.
214 std::string ClassIDToDebugStr(const INT_TEMPLATES_STRUCT *templates, int class_id,
215 int config_id) const;
216 // Converts a classifier class_id index with a config ID to:
217 // shape_table_ present: a shape_table_ index OR
218 // No shape_table_: a font ID.
219 // Without shape training, each class_id, config pair represents a single
220 // unichar id/font combination, so this function looks up the corresponding
221 // font id.
222 // With shape training, each class_id, config pair represents a single
223 // shape table index, so the fontset_table stores the shape table index,
224 // and the shape_table_ must be consulted to obtain the actual unichar_id/
225 // font combinations that the shape represents.
226 int ClassAndConfigIDToFontOrShapeID(int class_id, int int_result_config) const;
227 // Converts a shape_table_ index to a classifier class_id index (not a
228 // unichar-id!). Uses a search, so not fast.
229 int ShapeIDToClassID(int shape_id) const;
230 UNICHAR_ID *BaselineClassifier(TBLOB *Blob, const std::vector<INT_FEATURE_STRUCT> &int_features,
231 const INT_FX_RESULT_STRUCT &fx_info, ADAPT_TEMPLATES_STRUCT *Templates,
232 ADAPT_RESULTS *Results);
233 int CharNormClassifier(TBLOB *blob, const TrainingSample &sample, ADAPT_RESULTS *adapt_results);
234
235 // As CharNormClassifier, but operates on a TrainingSample and outputs to
236 // a vector of ShapeRating without conversion to classes.
237 int CharNormTrainingSample(bool pruner_only, int keep_this, const TrainingSample &sample,
238 std::vector<UnicharRating> *results);
239 UNICHAR_ID *GetAmbiguities(TBLOB *Blob, CLASS_ID CorrectClass);
240 void DoAdaptiveMatch(TBLOB *Blob, ADAPT_RESULTS *Results);
241 void AdaptToChar(TBLOB *Blob, CLASS_ID ClassId, int FontinfoId, float Threshold,
242 ADAPT_TEMPLATES_STRUCT *adaptive_templates);
243 void DisplayAdaptedChar(TBLOB *blob, INT_CLASS_STRUCT *int_class);
244 bool AdaptableWord(WERD_RES *word);
245 void EndAdaptiveClassifier();
246 void SetupPass1();
247 void SetupPass2();
248 void AdaptiveClassifier(TBLOB *Blob, BLOB_CHOICE_LIST *Choices);
249 void ClassifyAsNoise(ADAPT_RESULTS *Results);
250 void ResetAdaptiveClassifierInternal();
251 void SwitchAdaptiveClassifier();
252 void StartBackupAdaptiveClassifier();
253
254 int GetCharNormFeature(const INT_FX_RESULT_STRUCT &fx_info, INT_TEMPLATES_STRUCT *templates,
255 uint8_t *pruner_norm_array, uint8_t *char_norm_array);
256 // Computes the char_norm_array for the unicharset and, if not nullptr, the
257 // pruner_array as appropriate according to the existence of the shape_table.
258 // The norm_feature is deleted as it is almost certainly no longer needed.
259 void ComputeCharNormArrays(FEATURE_STRUCT *norm_feature, INT_TEMPLATES_STRUCT *templates,
260 uint8_t *char_norm_array, uint8_t *pruner_array);
261
262 bool TempConfigReliable(CLASS_ID class_id, const TEMP_CONFIG_STRUCT *config);
263 void UpdateAmbigsGroup(CLASS_ID class_id, TBLOB *Blob);
264
265 bool AdaptiveClassifierIsFull() const {
266 return NumAdaptationsFailed > 0;
267 }
268 bool AdaptiveClassifierIsEmpty() const {
269 return AdaptedTemplates->NumPermClasses == 0;
270 }
271 bool LooksLikeGarbage(TBLOB *blob);
272 #ifndef GRAPHICS_DISABLED
273 void RefreshDebugWindow(ScrollView **win, const char *msg, int y_offset, const TBOX &wbox);
274 #endif
275 // intfx.cpp
276 // Computes the DENORMS for bl(baseline) and cn(character) normalization
277 // during feature extraction. The input denorm describes the current state
278 // of the blob, which is usually a baseline-normalized word.
279 // The Transforms setup are as follows:
280 // Baseline Normalized (bl) Output:
281 // We center the grapheme by aligning the x-coordinate of its centroid with
282 // x=128 and leaving the already-baseline-normalized y as-is.
283 //
284 // Character Normalized (cn) Output:
285 // We align the grapheme's centroid at the origin and scale it
286 // asymmetrically in x and y so that the 2nd moments are a standard value
287 // (51.2) ie the result is vaguely square.
288 // If classify_nonlinear_norm is true:
289 // A non-linear normalization is setup that attempts to evenly distribute
290 // edges across x and y.
291 //
292 // Some of the fields of fx_info are also setup:
293 // Length: Total length of outline.
294 // Rx: Rounded y second moment. (Reversed by convention.)
295 // Ry: rounded x second moment.
296 // Xmean: Rounded x center of mass of the blob.
297 // Ymean: Rounded y center of mass of the blob.
298 static void SetupBLCNDenorms(const TBLOB &blob, bool nonlinear_norm, DENORM *bl_denorm,
299 DENORM *cn_denorm, INT_FX_RESULT_STRUCT *fx_info);
300
301 // Extracts sets of 3-D features of length kStandardFeatureLength (=12.8), as
302 // (x,y) position and angle as measured counterclockwise from the vector
303 // <-1, 0>, from blob using two normalizations defined by bl_denorm and
304 // cn_denorm. See SetpuBLCNDenorms for definitions.
305 // If outline_cn_counts is not nullptr, on return it contains the cumulative
306 // number of cn features generated for each outline in the blob (in order).
307 // Thus after the first outline, there were (*outline_cn_counts)[0] features,
308 // after the second outline, there were (*outline_cn_counts)[1] features etc.
309 static void ExtractFeatures(const TBLOB &blob, bool nonlinear_norm,
310 std::vector<INT_FEATURE_STRUCT> *bl_features,
311 std::vector<INT_FEATURE_STRUCT> *cn_features,
312 INT_FX_RESULT_STRUCT *results, std::vector<int> *outline_cn_counts);
313 /* float2int.cpp ************************************************************/
314 void ClearCharNormArray(uint8_t *char_norm_array);
315 void ComputeIntCharNormArray(const FEATURE_STRUCT &norm_feature, uint8_t *char_norm_array);
316 void ComputeIntFeatures(FEATURE_SET Features, INT_FEATURE_ARRAY IntFeatures);
317 /* intproto.cpp *************************************************************/
318 INT_TEMPLATES_STRUCT *ReadIntTemplates(TFile *fp);
319 void WriteIntTemplates(FILE *File, INT_TEMPLATES_STRUCT *Templates, const UNICHARSET &target_unicharset);
320 CLASS_ID GetClassToDebug(const char *Prompt, bool *adaptive_on, bool *pretrained_on,
321 int *shape_id);
322 void ShowMatchDisplay();
323 /* font detection ***********************************************************/
324 UnicityTable<FontInfo> &get_fontinfo_table() {
325 return fontinfo_table_;
326 }
327 const UnicityTable<FontInfo> &get_fontinfo_table() const {
328 return fontinfo_table_;
329 }
330 UnicityTable<FontSet> &get_fontset_table() {
331 return fontset_table_;
332 }
333 /* mfoutline.cpp ***********************************************************/
334 void NormalizeOutlines(LIST Outlines, float *XScale, float *YScale);
335 /* outfeat.cpp ***********************************************************/
336 FEATURE_SET ExtractOutlineFeatures(TBLOB *Blob);
337 /* picofeat.cpp ***********************************************************/
338 FEATURE_SET ExtractPicoFeatures(TBLOB *Blob);
339 FEATURE_SET ExtractIntCNFeatures(const TBLOB &blob, const INT_FX_RESULT_STRUCT &fx_info);
340 FEATURE_SET ExtractIntGeoFeatures(const TBLOB &blob, const INT_FX_RESULT_STRUCT &fx_info);
341 /* blobclass.cpp ***********************************************************/
342 // Extracts features from the given blob and saves them in the tr_file_data_
343 // member variable.
344 // fontname: Name of font that this blob was printed in.
345 // cn_denorm: Character normalization transformation to apply to the blob.
346 // fx_info: Character normalization parameters computed with cn_denorm.
347 // blob_text: Ground truth text for the blob.
348 void LearnBlob(const std::string &fontname, TBLOB *Blob, const DENORM &cn_denorm,
349 const INT_FX_RESULT_STRUCT &fx_info, const char *blob_text);
350 // Writes stored training data to a .tr file based on the given filename.
351 // Returns false on error.
352 bool WriteTRFile(const char *filename);
353
354 // Member variables.
355
356 // Parameters.
357 // Set during training (in lang.config) to indicate whether the divisible
358 // blobs chopper should be used (true for latin script.)
359 BOOL_VAR_H(allow_blob_division);
360 // Set during training (in lang.config) to indicate whether the divisible
361 // blobs chopper should be used in preference to chopping. Set to true for
362 // southern Indic scripts.
363 BOOL_VAR_H(prioritize_division);
364 BOOL_VAR_H(classify_enable_learning);
365 INT_VAR_H(classify_debug_level);
366
367 /* mfoutline.cpp ***********************************************************/
368 /* control knobs used to control normalization of outlines */
369 INT_VAR_H(classify_norm_method);
370 double_VAR_H(classify_char_norm_range);
371 double_VAR_H(classify_max_rating_ratio);
372 double_VAR_H(classify_max_certainty_margin);
373
374 /* adaptmatch.cpp ***********************************************************/
375 BOOL_VAR_H(tess_cn_matching);
376 BOOL_VAR_H(tess_bn_matching);
377 BOOL_VAR_H(classify_enable_adaptive_matcher);
378 BOOL_VAR_H(classify_use_pre_adapted_templates);
379 BOOL_VAR_H(classify_save_adapted_templates);
380 BOOL_VAR_H(classify_enable_adaptive_debugger);
381 BOOL_VAR_H(classify_nonlinear_norm);
382 INT_VAR_H(matcher_debug_level);
383 INT_VAR_H(matcher_debug_flags);
384 INT_VAR_H(classify_learning_debug_level);
385 double_VAR_H(matcher_good_threshold);
386 double_VAR_H(matcher_reliable_adaptive_result);
387 double_VAR_H(matcher_perfect_threshold);
388 double_VAR_H(matcher_bad_match_pad);
389 double_VAR_H(matcher_rating_margin);
390 double_VAR_H(matcher_avg_noise_size);
391 INT_VAR_H(matcher_permanent_classes_min);
392 INT_VAR_H(matcher_min_examples_for_prototyping);
393 INT_VAR_H(matcher_sufficient_examples_for_prototyping);
394 double_VAR_H(matcher_clustering_max_angle_delta);
395 double_VAR_H(classify_misfit_junk_penalty);
396 double_VAR_H(rating_scale);
397 double_VAR_H(tessedit_class_miss_scale);
398 double_VAR_H(classify_adapted_pruning_factor);
399 double_VAR_H(classify_adapted_pruning_threshold);
400 INT_VAR_H(classify_adapt_proto_threshold);
401 INT_VAR_H(classify_adapt_feature_threshold);
402 BOOL_VAR_H(disable_character_fragments);
403 double_VAR_H(classify_character_fragments_garbage_certainty_threshold);
404 BOOL_VAR_H(classify_debug_character_fragments);
405 BOOL_VAR_H(matcher_debug_separate_windows);
406 STRING_VAR_H(classify_learn_debug_str);
407
408 /* intmatcher.cpp **********************************************************/
409 INT_VAR_H(classify_class_pruner_threshold);
410 INT_VAR_H(classify_class_pruner_multiplier);
411 INT_VAR_H(classify_cp_cutoff_strength);
412 INT_VAR_H(classify_integer_matcher_multiplier);
413
414 BOOL_VAR_H(classify_bln_numeric_mode);
415 double_VAR_H(speckle_large_max_size);
416 double_VAR_H(speckle_rating_penalty);
417
418 // Use class variables to hold onto built-in templates and adapted templates.
419 INT_TEMPLATES_STRUCT *PreTrainedTemplates = nullptr;
420 ADAPT_TEMPLATES_STRUCT *AdaptedTemplates = nullptr;
421 // The backup adapted templates are created from the previous page (only)
422 // so they are always ready and reasonably well trained if the primary
423 // adapted templates become full.
424 ADAPT_TEMPLATES_STRUCT *BackupAdaptedTemplates = nullptr;
425
426 // Create dummy proto and config masks for use with the built-in templates.
427 BIT_VECTOR AllProtosOn = nullptr;
428 BIT_VECTOR AllConfigsOn = nullptr;
429 BIT_VECTOR AllConfigsOff = nullptr;
430 BIT_VECTOR TempProtoMask = nullptr;
431 /* normmatch.cpp */
432 NORM_PROTOS *NormProtos = nullptr;
433 /* font detection ***********************************************************/
434 UnicityTable<FontInfo> fontinfo_table_;
435 // Without shape training, each class_id, config pair represents a single
436 // unichar id/font combination, so each fontset_table_ entry holds font ids
437 // for each config in the class.
438 // With shape training, each class_id, config pair represents a single
439 // shape_table_ index, so the fontset_table_ stores the shape_table_ index,
440 // and the shape_table_ must be consulted to obtain the actual unichar_id/
441 // font combinations that the shape represents.
442 UnicityTable<FontSet> fontset_table_;
443
444 protected:
445 IntegerMatcher im_;
446 FEATURE_DEFS_STRUCT feature_defs_;
447 // If a shape_table_ is present, it is used to remap classifier output in
448 // ExpandShapesAndApplyCorrections. font_ids referenced by configs actually
449 // mean an index to the shape_table_ and the choices returned are *all* the
450 // shape_table_ entries at that index.
451 ShapeTable *shape_table_ = nullptr;
452
453 private:
454 // The currently active static classifier.
455 ShapeClassifier *static_classifier_ = nullptr;
456 #ifndef GRAPHICS_DISABLED
457 ScrollView *learn_debug_win_ = nullptr;
458 ScrollView *learn_fragmented_word_debug_win_ = nullptr;
459 ScrollView *learn_fragments_debug_win_ = nullptr;
460 #endif
461
462 // Training data gathered here for all the images in a document.
463 std::string tr_file_data_;
464
465 Dict dict_;
466
467 std::vector<uint16_t> shapetable_cutoffs_;
468
469 /* variables used to hold performance statistics */
470 int NumAdaptationsFailed = 0;
471
472 // Expected number of features in the class pruner, used to penalize
473 // unknowns that have too few features (like a c being classified as e) so
474 // it doesn't recognize everything as '@' or '#'.
475 // CharNormCutoffs is for the static classifier (with no shapetable).
476 // BaselineCutoffs gets a copy of CharNormCutoffs as an estimate of the real
477 // value in the adaptive classifier. Both are indexed by unichar_id.
478 // shapetable_cutoffs_ provides a similar value for each shape in the
479 // shape_table_
480 uint16_t CharNormCutoffs[MAX_NUM_CLASSES];
481 uint16_t BaselineCutoffs[MAX_NUM_CLASSES];
482
483 public:
484 bool EnableLearning = true;
485 };
486
487 } // namespace tesseract
488
489 #endif // DISABLED_LEGACY_ENGINE
490
491 #endif // TESSERACT_CLASSIFY_CLASSIFY_H_