tesseract v5.3.3.20231005
shapeclassifier.cpp
Go to the documentation of this file.
1// Copyright 2011 Google Inc. All Rights Reserved.
2// Author: rays@google.com (Ray Smith)
4// File: shapeclassifier.cpp
5// Description: Base interface class for classifiers that return a
6// shape index.
7// Author: Ray Smith
8//
9// (C) Copyright 2011, Google Inc.
10// Licensed under the Apache License, Version 2.0 (the "License");
11// you may not use this file except in compliance with the License.
12// You may obtain a copy of the License at
13// http://www.apache.org/licenses/LICENSE-2.0
14// Unless required by applicable law or agreed to in writing, software
15// distributed under the License is distributed on an "AS IS" BASIS,
16// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17// See the License for the specific language governing permissions and
18// limitations under the License.
19//
21
22#ifdef HAVE_CONFIG_H
23# include "config_auto.h"
24#endif
25
26#include "shapeclassifier.h"
27
28#include "scrollview.h"
29#include "shapetable.h"
30#ifndef GRAPHICS_DISABLED
31#include "svmnode.h"
32#endif
33#include "tprintf.h"
34#include "trainingsample.h"
35
36namespace tesseract {
37
38// Classifies the given [training] sample, writing to results.
39// See shapeclassifier.h for a full description.
40// Default implementation calls the ShapeRating version.
41int ShapeClassifier::UnicharClassifySample(const TrainingSample &sample, Image page_pix, int debug,
42 UNICHAR_ID keep_this,
43 std::vector<UnicharRating> *results) {
44 results->clear();
45 std::vector<ShapeRating> shape_results;
46 int num_shape_results = ClassifySample(sample, page_pix, debug, keep_this, &shape_results);
47 const ShapeTable *shapes = GetShapeTable();
48 std::vector<int> unichar_map(shapes->unicharset().size(), -1);
49 for (int r = 0; r < num_shape_results; ++r) {
50 shapes->AddShapeToResults(shape_results[r], &unichar_map, results);
51 }
52 return results->size();
53}
54
55// Classifies the given [training] sample, writing to results.
56// See shapeclassifier.h for a full description.
57// Default implementation aborts.
58int ShapeClassifier::ClassifySample(const TrainingSample &sample, Image page_pix, int debug,
59 int keep_this, std::vector<ShapeRating> *results) {
60 ASSERT_HOST("Must implement ClassifySample!" == nullptr);
61 return 0;
62}
63
64// Returns the shape that contains unichar_id that has the best result.
65// If result is not nullptr, it is set with the shape_id and rating.
66// Does not need to be overridden if ClassifySample respects the keep_this
67// rule.
69 UNICHAR_ID unichar_id, ShapeRating *result) {
70 std::vector<ShapeRating> results;
71 const ShapeTable *shapes = GetShapeTable();
72 int num_results = ClassifySample(sample, page_pix, 0, unichar_id, &results);
73 for (int r = 0; r < num_results; ++r) {
74 if (shapes->GetShape(results[r].shape_id).ContainsUnichar(unichar_id)) {
75 if (result != nullptr) {
76 *result = results[r];
77 }
78 return results[r].shape_id;
79 }
80 }
81 return -1;
82}
83
84// Provides access to the UNICHARSET that this classifier works with.
85// Only needs to be overridden if GetShapeTable() can return nullptr.
87 return GetShapeTable()->unicharset();
88}
89
90#ifndef GRAPHICS_DISABLED
91
92// Visual debugger classifies the given sample, displays the results and
93// solicits user input to display other classifications. Returns when
94// the user has finished with debugging the sample.
95// Probably doesn't need to be overridden if the subclass provides
96// DisplayClassifyAs.
98 UNICHAR_ID unichar_id) {
99 static ScrollView *terminator = nullptr;
100 if (terminator == nullptr) {
101 terminator = new ScrollView("XIT", 0, 0, 50, 50, 50, 50, true);
102 }
103 ScrollView *debug_win = CreateFeatureSpaceWindow("ClassifierDebug", 0, 0);
104 // Provide a right-click menu to choose the class.
105 auto *popup_menu = new SVMenuNode();
106 popup_menu->AddChild("Choose class to debug", 0, "x", "Class to debug");
107 popup_menu->BuildMenu(debug_win, false);
108 // Display the features in green.
109 const INT_FEATURE_STRUCT *features = sample.features();
110 uint32_t num_features = sample.num_features();
111 for (uint32_t f = 0; f < num_features; ++f) {
112 RenderIntFeature(debug_win, &features[f], ScrollView::GREEN);
113 }
114 debug_win->Update();
115 std::vector<UnicharRating> results;
116 // Debug classification until the user quits.
117 const UNICHARSET &unicharset = GetUnicharset();
118 SVEventType ev_type;
119 do {
120 std::vector<ScrollView *> windows;
121 if (unichar_id >= 0) {
122 tprintf("Debugging class %d = %s\n", unichar_id, unicharset.id_to_unichar(unichar_id));
123 UnicharClassifySample(sample, page_pix, 1, unichar_id, &results);
124 DisplayClassifyAs(sample, page_pix, unichar_id, 1, windows);
125 } else {
126 tprintf("Invalid unichar_id: %d\n", unichar_id);
127 UnicharClassifySample(sample, page_pix, 1, -1, &results);
128 }
129 if (unichar_id >= 0) {
130 tprintf("Debugged class %d = %s\n", unichar_id, unicharset.id_to_unichar(unichar_id));
131 }
132 tprintf("Right-click in ClassifierDebug window to choose debug class,");
133 tprintf(" Left-click or close window to quit...\n");
134 UNICHAR_ID old_unichar_id;
135 do {
136 old_unichar_id = unichar_id;
137 auto ev = debug_win->AwaitEvent(SVET_ANY);
138 ev_type = ev->type;
139 if (ev_type == SVET_POPUP) {
140 if (unicharset.contains_unichar(ev->parameter)) {
141 unichar_id = unicharset.unichar_to_id(ev->parameter);
142 } else {
143 tprintf("Char class '%s' not found in unicharset", ev->parameter);
144 }
145 }
146 } while (unichar_id == old_unichar_id && ev_type != SVET_CLICK && ev_type != SVET_DESTROY);
147 for (auto window : windows) {
148 delete window;
149 }
150 } while (ev_type != SVET_CLICK && ev_type != SVET_DESTROY);
151 delete debug_win;
152}
153
154#endif // !GRAPHICS_DISABLED
155
156// Displays classification as the given shape_id. Creates as many windows
157// as it feels fit, using index as a guide for placement. Adds any created
158// windows to the windows output and returns a new index that may be used
159// by any subsequent classifiers. Caller waits for the user to view and
160// then destroys the windows by clearing the vector.
162 UNICHAR_ID unichar_id, int index,
163 std::vector<ScrollView *> &windows) {
164 // Does nothing in the default implementation.
165 return index;
166}
167
168// Prints debug information on the results.
170 const std::vector<UnicharRating> &results) const {
171 tprintf("%s\n", context);
172 for (const auto &result : results) {
173 tprintf("%g: c_id=%d=%s", result.rating, result.unichar_id,
174 GetUnicharset().id_to_unichar(result.unichar_id));
175 if (!result.fonts.empty()) {
176 tprintf(" Font Vector:");
177 for (auto &&font : result.fonts) {
178 tprintf(" %d", font.fontinfo_id);
179 }
180 }
181 tprintf("\n");
182 }
183}
184void ShapeClassifier::PrintResults(const char *context,
185 const std::vector<ShapeRating> &results) const {
186 tprintf("%s\n", context);
187 for (const auto &result : results) {
188 tprintf("%g:", result.rating);
189 if (result.joined) {
190 tprintf("[J]");
191 }
192 if (result.broken) {
193 tprintf("[B]");
194 }
195 tprintf(" %s\n", GetShapeTable()->DebugStr(result.shape_id).c_str());
196 }
197}
198
199// Removes any result that has all its unichars covered by a better choice,
200// regardless of font.
201void ShapeClassifier::FilterDuplicateUnichars(std::vector<ShapeRating> *results) const {
202 std::vector<ShapeRating> filtered_results;
203 // Copy results to filtered results and knock out duplicate unichars.
204 const ShapeTable *shapes = GetShapeTable();
205 for (unsigned r = 0; r < results->size(); ++r) {
206 if (r > 0) {
207 const Shape &shape_r = shapes->GetShape((*results)[r].shape_id);
208 int c;
209 for (c = 0; c < shape_r.size(); ++c) {
210 int unichar_id = shape_r[c].unichar_id;
211 unsigned s;
212 for (s = 0; s < r; ++s) {
213 const Shape &shape_s = shapes->GetShape((*results)[s].shape_id);
214 if (shape_s.ContainsUnichar(unichar_id)) {
215 break; // We found unichar_id.
216 }
217 }
218 if (s == r) {
219 break; // We didn't find unichar_id.
220 }
221 }
222 if (c == shape_r.size()) {
223 continue; // We found all the unichar ids in previous answers.
224 }
225 }
226 filtered_results.push_back((*results)[r]);
227 }
228 *results = filtered_results;
229}
230
231} // namespace tesseract.
#define ASSERT_HOST(x)
Definition: errcode.h:54
void tprintf(const char *format,...)
Definition: tprintf.cpp:41
@ SVET_DESTROY
Definition: scrollview.h:54
@ SVET_POPUP
Definition: scrollview.h:62
@ SVET_CLICK
Definition: scrollview.h:56
ScrollView * CreateFeatureSpaceWindow(const char *name, int xpos, int ypos)
Definition: intproto.cpp:1622
void RenderIntFeature(ScrollView *window, const INT_FEATURE_STRUCT *Feature, ScrollView::Color color)
Definition: intproto.cpp:1500
int UNICHAR_ID
Definition: unichar.h:34
const char * id_to_unichar(UNICHAR_ID id) const
Definition: unicharset.cpp:279
bool contains_unichar(const char *const unichar_repr) const
Definition: unicharset.cpp:695
UNICHAR_ID unichar_to_id(const char *const unichar_repr) const
Definition: unicharset.cpp:186
size_t size() const
Definition: unicharset.h:355
void FilterDuplicateUnichars(std::vector< ShapeRating > *results) const
virtual const ShapeTable * GetShapeTable() const =0
virtual int UnicharClassifySample(const TrainingSample &sample, Image page_pix, int debug, UNICHAR_ID keep_this, std::vector< UnicharRating > *results)
virtual int BestShapeForUnichar(const TrainingSample &sample, Image page_pix, UNICHAR_ID unichar_id, ShapeRating *result)
virtual int ClassifySample(const TrainingSample &sample, Image page_pix, int debug, UNICHAR_ID keep_this, std::vector< ShapeRating > *results)
virtual const UNICHARSET & GetUnicharset() const
void DebugDisplay(const TrainingSample &sample, Image page_pix, UNICHAR_ID unichar_id)
virtual void PrintResults(const char *context, const std::vector< ShapeRating > &results) const
virtual int DisplayClassifyAs(const TrainingSample &sample, Image page_pix, UNICHAR_ID unichar_id, int index, std::vector< ScrollView * > &windows)
virtual void UnicharPrintResults(const char *context, const std::vector< UnicharRating > &results) const
int size() const
Definition: shapetable.h:169
bool ContainsUnichar(int unichar_id) const
Definition: shapetable.cpp:150
const Shape & GetShape(unsigned shape_id) const
Definition: shapetable.h:292
void AddShapeToResults(const ShapeRating &shape_rating, std::vector< int > *unichar_map, std::vector< UnicharRating > *results) const
Definition: shapetable.cpp:716
const UNICHARSET & unicharset() const
Definition: shapetable.h:251
const INT_FEATURE_STRUCT * features() const
uint32_t num_features() const
std::unique_ptr< SVEvent > AwaitEvent(SVEventType type)
Definition: scrollview.cpp:432
static void Update()
Definition: scrollview.cpp:700