Mercurial > hgrepos > Python2 > PyMuPDF

comparison mupdf-source/thirdparty/tesseract/src/wordrec/lm_pain_points.cpp @ 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: pain_points.cpp
3 // Description: Functions that utilize the knowledge about the properties
4 // of the paths explored by the segmentation search in order
5 // to "pain points" - the locations in the ratings matrix
6 // which should be classified next.
7 // Author: Rika Antonova
8 // Created: Mon Jun 20 11:26:43 PST 2012
9 //
10 // (C) Copyright 2012, Google Inc.
11 // Licensed under the Apache License, Version 2.0 (the "License");
12 // you may not use this file except in compliance with the License.
13 // You may obtain a copy of the License at
14 // http://www.apache.org/licenses/LICENSE-2.0
15 // Unless required by applicable law or agreed to in writing, software
16 // distributed under the License is distributed on an "AS IS" BASIS,
17 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
18 // See the License for the specific language governing permissions and
19 // limitations under the License.
20 //
21 ///////////////////////////////////////////////////////////////////////
22
23 #include "lm_pain_points.h"
24
25 #include "associate.h"
26 #include "dict.h"
27 #include "genericheap.h"
28 #include "lm_state.h"
29 #include "matrix.h"
30 #include "pageres.h"
31
32 #include <algorithm>
33
34 namespace tesseract {
35
36 const float LMPainPoints::kDefaultPainPointPriorityAdjustment = 2.0f;
37 const float LMPainPoints::kLooseMaxCharWhRatio = 2.5f;
38
39 LMPainPointsType LMPainPoints::Deque(MATRIX_COORD *pp, float *priority) {
40 for (int h = 0; h < LM_PPTYPE_NUM; ++h) {
41 if (pain_points_heaps_[h].empty()) {
42 continue;
43 }
44 *priority = pain_points_heaps_[h].PeekTop().key();
45 *pp = pain_points_heaps_[h].PeekTop().data();
46 pain_points_heaps_[h].Pop(nullptr);
47 return static_cast<LMPainPointsType>(h);
48 }
49 return LM_PPTYPE_NUM;
50 }
51
52 void LMPainPoints::GenerateInitial(WERD_RES *word_res) {
53 MATRIX *ratings = word_res->ratings;
54 AssociateStats associate_stats;
55 for (int col = 0; col < ratings->dimension(); ++col) {
56 int row_end = std::min(ratings->dimension(), col + ratings->bandwidth() + 1);
57 for (int row = col + 1; row < row_end; ++row) {
58 MATRIX_COORD coord(col, row);
59 if (coord.Valid(*ratings) && ratings->get(col, row) != NOT_CLASSIFIED) {
60 continue;
61 }
62 // Add an initial pain point if needed.
63 if (ratings->Classified(col, row - 1, dict_->WildcardID()) ||
64 (col + 1 < ratings->dimension() &&
65 ratings->Classified(col + 1, row, dict_->WildcardID()))) {
66 GeneratePainPoint(col, row, LM_PPTYPE_SHAPE, 0.0, true, max_char_wh_ratio_, word_res);
67 }
68 }
69 }
70 }
71
72 void LMPainPoints::GenerateFromPath(float rating_cert_scale, ViterbiStateEntry *vse,
73 WERD_RES *word_res) {
74 ViterbiStateEntry *curr_vse = vse;
75 BLOB_CHOICE *curr_b = vse->curr_b;
76 // The following pain point generation and priority calculation approaches
77 // prioritize exploring paths with low average rating of the known part of
78 // the path, while not relying on the ratings of the pieces to be combined.
79 //
80 // A pain point to combine the neighbors is generated for each pair of
81 // neighboring blobs on the path (the path is represented by vse argument
82 // given to GenerateFromPath()). The priority of each pain point is set to
83 // the average rating (per outline length) of the path, not including the
84 // ratings of the blobs to be combined.
85 // The ratings of the blobs to be combined are not used to calculate the
86 // priority, since it is not possible to determine from their magnitude
87 // whether it will be beneficial to combine the blobs. The reason is that
88 // chopped junk blobs (/ | - ') can have very good (low) ratings, however
89 // combining them will be beneficial. Blobs with high ratings might be
90 // over-joined pieces of characters, but also could be blobs from an unseen
91 // font or chopped pieces of complex characters.
92 while (curr_vse->parent_vse != nullptr) {
93 ViterbiStateEntry *parent_vse = curr_vse->parent_vse;
94 const MATRIX_COORD &curr_cell = curr_b->matrix_cell();
95 const MATRIX_COORD &parent_cell = parent_vse->curr_b->matrix_cell();
96 MATRIX_COORD pain_coord(parent_cell.col, curr_cell.row);
97 if (!pain_coord.Valid(*word_res->ratings) ||
98 !word_res->ratings->Classified(parent_cell.col, curr_cell.row, dict_->WildcardID())) {
99 // rat_subtr contains ratings sum of the two adjacent blobs to be merged.
100 // rat_subtr will be subtracted from the ratings sum of the path, since
101 // the blobs will be joined into a new blob, whose rating is yet unknown.
102 float rat_subtr = curr_b->rating() + parent_vse->curr_b->rating();
103 // ol_subtr contains the outline length of the blobs that will be joined.
104 float ol_subtr =
105 AssociateUtils::ComputeOutlineLength(rating_cert_scale, *curr_b) +
106 AssociateUtils::ComputeOutlineLength(rating_cert_scale, *(parent_vse->curr_b));
107 // ol_dif is the outline of the path without the two blobs to be joined.
108 float ol_dif = vse->outline_length - ol_subtr;
109 // priority is set to the average rating of the path per unit of outline,
110 // not counting the ratings of the pieces to be joined.
111 float priority = ol_dif > 0 ? (vse->ratings_sum - rat_subtr) / ol_dif : 0.0;
112 GeneratePainPoint(pain_coord.col, pain_coord.row, LM_PPTYPE_PATH, priority, true,
113 max_char_wh_ratio_, word_res);
114 } else if (debug_level_ > 3) {
115 tprintf("NO pain point (Classified) for col=%d row=%d type=%s\n", pain_coord.col,
116 pain_coord.row, LMPainPointsTypeName[LM_PPTYPE_PATH]);
117 BLOB_CHOICE_IT b_it(word_res->ratings->get(pain_coord.col, pain_coord.row));
118 for (b_it.mark_cycle_pt(); !b_it.cycled_list(); b_it.forward()) {
119 BLOB_CHOICE *choice = b_it.data();
120 choice->print_full();
121 }
122 }
123
124 curr_vse = parent_vse;
125 curr_b = curr_vse->curr_b;
126 }
127 }
128
129 void LMPainPoints::GenerateFromAmbigs(const DANGERR &fixpt, ViterbiStateEntry *vse,
130 WERD_RES *word_res) {
131 // Begins and ends in DANGERR vector now record the blob indices as used
132 // by the ratings matrix.
133 for (auto &&danger : fixpt) {
134 // Only use dangerous ambiguities.
135 if (danger.dangerous) {
136 GeneratePainPoint(danger.begin, danger.end - 1, LM_PPTYPE_AMBIG, vse->cost, true,
137 kLooseMaxCharWhRatio, word_res);
138 }
139 }
140 }
141
142 bool LMPainPoints::GeneratePainPoint(int col, int row, LMPainPointsType pp_type,
143 float special_priority, bool ok_to_extend,
144 float max_char_wh_ratio, WERD_RES *word_res) {
145 MATRIX_COORD coord(col, row);
146 if (coord.Valid(*word_res->ratings) &&
147 word_res->ratings->Classified(col, row, dict_->WildcardID())) {
148 return false;
149 }
150 if (debug_level_ > 3) {
151 tprintf("Generating pain point for col=%d row=%d type=%s\n", col, row,
152 LMPainPointsTypeName[pp_type]);
153 }
154 // Compute associate stats.
155 AssociateStats associate_stats;
156 AssociateUtils::ComputeStats(col, row, nullptr, 0, fixed_pitch_, max_char_wh_ratio, word_res,
157 debug_level_, &associate_stats);
158 // For fixed-pitch fonts/languages: if the current combined blob overlaps
159 // the next blob on the right and it is ok to extend the blob, try extending
160 // the blob until there is no overlap with the next blob on the right or
161 // until the width-to-height ratio becomes too large.
162 if (ok_to_extend) {
163 while (associate_stats.bad_fixed_pitch_right_gap && row + 1 < word_res->ratings->dimension() &&
164 !associate_stats.bad_fixed_pitch_wh_ratio) {
165 AssociateUtils::ComputeStats(col, ++row, nullptr, 0, fixed_pitch_, max_char_wh_ratio,
166 word_res, debug_level_, &associate_stats);
167 }
168 }
169 if (associate_stats.bad_shape) {
170 if (debug_level_ > 3) {
171 tprintf("Discarded pain point with a bad shape\n");
172 }
173 return false;
174 }
175
176 // Insert the new pain point into pain_points_heap_.
177 if (pain_points_heaps_[pp_type].size() < max_heap_size_) {
178 // Compute pain point priority.
179 float priority;
180 if (pp_type == LM_PPTYPE_PATH) {
181 priority = special_priority;
182 } else {
183 priority = associate_stats.gap_sum;
184 }
185 MatrixCoordPair pain_point(priority, MATRIX_COORD(col, row));
186 pain_points_heaps_[pp_type].Push(&pain_point);
187 if (debug_level_) {
188 tprintf("Added pain point with priority %g\n", priority);
189 }
190 return true;
191 } else {
192 if (debug_level_) {
193 tprintf("Pain points heap is full\n");
194 }
195 return false;
196 }
197 }
198
199 /**
200 * Adjusts the pain point coordinates to cope with expansion of the ratings
201 * matrix due to a split of the blob with the given index.
202 */
203 void LMPainPoints::RemapForSplit(int index) {
204 for (auto &pain_points_heap : pain_points_heaps_) {
205 std::vector<MatrixCoordPair> &heap = pain_points_heap.heap();
206 for (auto &&entry : heap) {
207 entry.data().MapForSplit(index);
208 }
209 }
210 }
211
212 } // namespace tesseract