tesseract/training/mftraining.cpp

/******************************************************************************
**  Filename:  mftraining.c
**  Purpose:  Separates training pages into files for each character.
**        Strips from files only the features and there parameters of
        the feature type mf.
**  Author:    Dan Johnson
**  Revisment:  Christy Russon
**  Environment: HPUX 6.5
**  Library:     HPUX 6.5
**  History:     Fri Aug 18 08:53:50 1989, DSJ, Created.
**         5/25/90, DSJ, Adapted to multiple feature types.
**        Tuesday, May 17, 1998 Changes made to make feature specific and
**        simplify structures. First step in simplifying training process.
**
 **  (c) Copyright Hewlett-Packard Company, 1988.
 ** Licensed under the Apache License, Version 2.0 (the "License");
 ** you may not use this file except in compliance with the License.
 ** You may obtain a copy of the License at
 ** http://www.apache.org/licenses/LICENSE-2.0
 ** Unless required by applicable law or agreed to in writing, software
 ** distributed under the License is distributed on an "AS IS" BASIS,
 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 ** See the License for the specific language governing permissions and
 ** limitations under the License.
******************************************************************************/
/*----------------------------------------------------------------------------
          Include Files and Type Defines
----------------------------------------------------------------------------*/
#ifdef HAVE_CONFIG_H
#include "config_auto.h"
#endif

#include <string.h>
#include <stdio.h>
#define _USE_MATH_DEFINES
#include <math.h>
#ifdef _WIN32
#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif
#endif

#include "classify.h"
#include "cluster.h"
#include "clusttool.h"
#include "commontraining.h"
#include "danerror.h"
#include "efio.h"
#include "emalloc.h"
#include "featdefs.h"
#include "fontinfo.h"
#include "genericvector.h"
#include "indexmapbidi.h"
#include "intproto.h"
#include "mastertrainer.h"
#include "mergenf.h"
#include "mf.h"
#include "ndminx.h"
#include "ocrfeatures.h"
#include "oldlist.h"
#include "protos.h"
#include "shapetable.h"
#include "tessopt.h"
#include "tprintf.h"
#include "unicity_table.h"

using tesseract::IndexMapBiDi;
using tesseract::MasterTrainer;
using tesseract::Shape;
using tesseract::ShapeTable;

#define PROGRAM_FEATURE_TYPE "mf"

// Max length of a fake shape label.
const int kMaxShapeLabelLength = 10;

DECLARE_STRING_PARAM_FLAG(test_ch);

/*----------------------------------------------------------------------------
          Public Function Prototypes
----------------------------------------------------------------------------*/
int main (
     int  argc,
     char  **argv);


/*----------------------------------------------------------------------------
            Public Code
-----------------------------------------------------------------------------*/
#ifndef GRAPHICS_DISABLED
static void DisplayProtoList(const char* ch, LIST protolist) {
  void* window = c_create_window("Char samples", 50, 200,
                                 520, 520, -130.0, 130.0, -130.0, 130.0);
  LIST proto = protolist;
  iterate(proto) {
    PROTOTYPE* prototype = reinterpret_cast<PROTOTYPE *>(first_node(proto));
    if (prototype->Significant)
      c_line_color_index(window, Green);
    else if (prototype->NumSamples == 0)
      c_line_color_index(window, Blue);
    else if (prototype->Merged)
      c_line_color_index(window, Magenta);
    else
      c_line_color_index(window, Red);
    float x = CenterX(prototype->Mean);
    float y = CenterY(prototype->Mean);
    double angle = OrientationOf(prototype->Mean) * 2 * M_PI;
    float dx = static_cast<float>(LengthOf(prototype->Mean) * cos(angle) / 2);
    float dy = static_cast<float>(LengthOf(prototype->Mean) * sin(angle) / 2);
    c_move(window, (x - dx) * 256, (y - dy) * 256);
    c_draw(window, (x + dx) * 256, (y + dy) * 256);
    if (prototype->Significant)
      tprintf("Green proto at (%g,%g)+(%g,%g) %d samples\n",
              x, y, dx, dy, prototype->NumSamples);
    else if (prototype->NumSamples > 0 && !prototype->Merged)
      tprintf("Red proto at (%g,%g)+(%g,%g) %d samples\n",
              x, y, dx, dy, prototype->NumSamples);
  }
  c_make_current(window);
}
#endif  // GRAPHICS_DISABLED

// Helper to run clustering on a single config.
// Mostly copied from the old mftraining, but with renamed variables.
static LIST ClusterOneConfig(int shape_id, const char* class_label,
                             LIST mf_classes,
                             const ShapeTable& shape_table,
                             MasterTrainer* trainer) {
  int num_samples;
  CLUSTERER  *clusterer = trainer->SetupForClustering(shape_table,
                                                      feature_defs,
                                                      shape_id,
                                                      &num_samples);
  Config.MagicSamples = num_samples;
  LIST proto_list = ClusterSamples(clusterer, &Config);
  CleanUpUnusedData(proto_list);

  // Merge protos where reasonable to make more of them significant by
  // representing almost all samples of the class/font.
  MergeInsignificantProtos(proto_list, class_label, clusterer, &Config);
  #ifndef GRAPHICS_DISABLED
  if (strcmp(FLAGS_test_ch.c_str(), class_label) == 0)
    DisplayProtoList(FLAGS_test_ch.c_str(), proto_list);
  #endif  // GRAPHICS_DISABLED
  // Delete the protos that will not be used in the inttemp output file.
  proto_list = RemoveInsignificantProtos(proto_list, true,
                                         false,
                                         clusterer->SampleSize);
  FreeClusterer(clusterer);
  MERGE_CLASS merge_class = FindClass(mf_classes, class_label);
  if (merge_class == NULL) {
    merge_class = NewLabeledClass(class_label);
    mf_classes = push(mf_classes, merge_class);
  }
  int config_id = AddConfigToClass(merge_class->Class);
  merge_class->Class->font_set.push_back(shape_id);
  LIST proto_it = proto_list;
  iterate(proto_it) {
    PROTOTYPE* prototype = reinterpret_cast<PROTOTYPE*>(first_node(proto_it));
    // See if proto can be approximated by existing proto.
    int p_id = FindClosestExistingProto(merge_class->Class,
                                        merge_class->NumMerged, prototype);
    if (p_id == NO_PROTO) {
      // Need to make a new proto, as it doesn't match anything.
      p_id = AddProtoToClass(merge_class->Class);
      MakeNewFromOld(ProtoIn(merge_class->Class, p_id), prototype);
      merge_class->NumMerged[p_id] = 1;
    } else {
      PROTO_STRUCT dummy_proto;
      MakeNewFromOld(&dummy_proto, prototype);
      // Merge with the similar proto.
      ComputeMergedProto(ProtoIn(merge_class->Class, p_id), &dummy_proto,
                         static_cast<FLOAT32>(merge_class->NumMerged[p_id]),
                         1.0,
                         ProtoIn(merge_class->Class, p_id));
      merge_class->NumMerged[p_id]++;
    }
    AddProtoToConfig(p_id, merge_class->Class->Configurations[config_id]);
  }
  FreeProtoList(&proto_list);
  return mf_classes;
}

// Helper to setup the config map.
// Setup an index mapping from the shapes in the shape table to the classes
// that will be trained. In keeping with the original design, each shape
// with the same list of unichars becomes a different class and the configs
// represent the different combinations of fonts.
static void SetupConfigMap(ShapeTable* shape_table, IndexMapBiDi* config_map) {
  int num_configs = shape_table->NumShapes();
  config_map->Init(num_configs, true);
  config_map->Setup();
  for (int c1 = 0; c1 < num_configs; ++c1) {
    // Only process ids that are not already merged.
    if (config_map->SparseToCompact(c1) == c1) {
      Shape* shape1 = shape_table->MutableShape(c1);
      // Find all the subsequent shapes that are equal.
      for (int c2 = c1 + 1; c2 < num_configs; ++c2) {
        if (shape_table->MutableShape(c2)->IsEqualUnichars(shape1)) {
          config_map->Merge(c1, c2);
        }
      }
    }
  }
  config_map->CompleteMerges();
}

/**
 * This program reads in a text file consisting of feature
 * samples from a training page in the following format:
 * @verbatim
      FontName UTF8-char-str xmin ymin xmax ymax page-number
       NumberOfFeatureTypes(N)
         FeatureTypeName1 NumberOfFeatures(M)
            Feature1
            ...
            FeatureM
         FeatureTypeName2 NumberOfFeatures(M)
            Feature1
            ...
            FeatureM
         ...
         FeatureTypeNameN NumberOfFeatures(M)
            Feature1
            ...
            FeatureM
      FontName CharName ...
    @endverbatim
 * The result of this program is a binary inttemp file used by
 * the OCR engine.
 * @param  argc  number of command line arguments
 * @param  argv  array of command line arguments
 * @return none
 * @note Exceptions: none
 * @note History:  Fri Aug 18 08:56:17 1989, DSJ, Created.
 * @note History: Mon May 18 1998, Christy Russson, Revistion started.
 */
int main (int argc, char **argv) {
  ParseArguments(&argc, &argv);

  ShapeTable* shape_table = NULL;
  STRING file_prefix;
  // Load the training data.
  MasterTrainer* trainer = tesseract::LoadTrainingData(argc, argv,
                                                       false,
                                                       &shape_table,
                                                       &file_prefix);
  if (trainer == NULL)
    return 1;  // Failed.

  // Setup an index mapping from the shapes in the shape table to the classes
  // that will be trained. In keeping with the original design, each shape
  // with the same list of unichars becomes a different class and the configs
  // represent the different combinations of fonts.
  IndexMapBiDi config_map;
  SetupConfigMap(shape_table, &config_map);

  WriteShapeTable(file_prefix, *shape_table);
  // If the shape_table is flat, then either we didn't run shape clustering, or
  // it did nothing, so we just output the trainer's unicharset.
  // Otherwise shape_set will hold a fake unicharset with an entry for each
  // shape in the shape table, and we will output that instead.
  UNICHARSET shape_set;
  const UNICHARSET* unicharset = &trainer->unicharset();
  // If we ran shapeclustering (and it worked) then at least one shape will
  // have multiple unichars, so we have to build a fake unicharset.
  if (shape_table->AnyMultipleUnichars()) {
    unicharset = &shape_set;
    // Now build a fake unicharset for the compact shape space to keep the
    // output modules happy that we are doing things correctly.
    int num_shapes = config_map.CompactSize();
    for (int s = 0; s < num_shapes; ++s) {
      char shape_label[kMaxShapeLabelLength + 1];
      snprintf(shape_label, kMaxShapeLabelLength, "sh%04d", s);
      shape_set.unichar_insert(shape_label);
    }
  }

  // Now train each config separately.
  int num_configs = shape_table->NumShapes();
  LIST mf_classes = NIL_LIST;
  for (int s = 0; s < num_configs; ++s) {
    int unichar_id, font_id;
    if (unicharset == &shape_set) {
      // Using fake unichar_ids from the config_map/shape_set.
      unichar_id = config_map.SparseToCompact(s);
    } else {
      // Get the real unichar_id from the shape table/unicharset.
      shape_table->GetFirstUnicharAndFont(s, &unichar_id, &font_id);
    }
    const char* class_label = unicharset->id_to_unichar(unichar_id);
    mf_classes = ClusterOneConfig(s, class_label, mf_classes, *shape_table,
                                  trainer);
  }
  STRING inttemp_file = file_prefix;
  inttemp_file += "inttemp";
  STRING pffmtable_file = file_prefix;
  pffmtable_file += "pffmtable";
  CLASS_STRUCT* float_classes = SetUpForFloat2Int(*unicharset, mf_classes);
  // Now write the inttemp and pffmtable.
  trainer->WriteInttempAndPFFMTable(trainer->unicharset(), *unicharset,
                                    *shape_table, float_classes,
                                    inttemp_file.string(),
                                    pffmtable_file.string());
  for (int c = 0; c < unicharset->size(); ++c) {
    FreeClassFields(&float_classes[c]);
  }
  delete [] float_classes;
  FreeLabeledClassList(mf_classes);
  delete trainer;
  delete shape_table;
  printf("Done!\n");
  if (!FLAGS_test_ch.empty()) {
    // If we are displaying debug window(s), wait for the user to look at them.
    printf("Hit return to exit...\n");
    while (getchar() != '\n');
  }
  return 0;
}  /* main */