applybox.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 892 行 · 第 1/3 页

                if (gblob_bounding_box (blob, polyg).
                overlap (box)) {
                  outline_it.
                    set_to_list (gblob_out_list
                    (blob, polyg));
                  for (outline_it.mark_cycle_pt ();
                    !outline_it.cycled_list ();
                  outline_it.forward ()) {
                    outline = outline_it.data ();
                    if (goutline_bounding_box
                    (outline, polyg).major_overlap (box)) {
                      if ((row_to_process == NULL) ||
                      (row_to_process == row)) {
                        row_to_process = row;
                        row_id_to_process = row_id;
                      }
                      else
                        /* RETURN ERROR Box overlaps blobs in more than one row  */
                        return NULL;
                    }
                  }
                }
              }
            }
          }
        }
      }
    }
  }
  return row_to_process;
}


inT16 resegment_box(  //
                    ROW *row,
                    TBOX box,
                    UNICHAR_ID uch_id,
                    inT16 block_id,
                    inT16 row_id,
                    inT16 boxfile_lineno,
                    inT16 boxfile_charno) {
  WERD_IT word_it;
  WERD *word;
  WERD *new_word = NULL;
  BOOL8 polyg = false;
  PBLOB_IT blob_it;
  PBLOB_IT new_blob_it;
  PBLOB *blob;
  PBLOB *new_blob;
  OUTLINE_IT outline_it;
  OUTLINE_LIST dummy;  // Just to initialize new_outline_it.
  OUTLINE_IT new_outline_it = &dummy;
  OUTLINE *outline;
  TBOX new_word_box;
  float word_x_centre;
  float baseline;
  inT16 error_count = 0;         //number of chars lost

  word_it.set_to_list (row->word_list ());
  for (word_it.mark_cycle_pt (); !word_it.cycled_list (); word_it.forward ()) {
    word = word_it.data ();
    polyg = word->flag (W_POLYGON);
    if (word->bounding_box ().overlap (box)) {
      blob_it.set_to_list (word->gblob_list ());
      for (blob_it.mark_cycle_pt ();
      !blob_it.cycled_list (); blob_it.forward ()) {
        blob = blob_it.data ();
        if (gblob_bounding_box (blob, polyg).overlap (box)) {
          outline_it.set_to_list (gblob_out_list (blob, polyg));
          for (outline_it.mark_cycle_pt ();
          !outline_it.cycled_list (); outline_it.forward ()) {
            outline = outline_it.data ();
            if (goutline_bounding_box (outline, polyg).
            major_overlap (box)) {
              if (strlen (word->text ()) > 0) {
                if (error_count == 0) {
                  error_count = 1;
                  if (applybox_debug > 4)
                    report_failed_box (boxfile_lineno,
                      boxfile_charno,
                      box, unicharset_boxes.id_to_unichar(uch_id),
                      "FAILURE! box overlaps blob in labelled word");
                }
                if (applybox_debug > 4)
                  tprintf
                    ("APPLY_BOXES: ALSO ignoring corrupted char blk:%d row:%d \"%s\"\n",
                    block_id, row_id,
                    word_it.data ()->text ());
                word_it.data ()->set_text ("");
                //UN label it
                error_count++;
              }

              if (error_count == 0) {
                if (new_word == NULL) {
                                 /* Make a new word with a single blob */
                  new_word = word->shallow_copy ();
                  new_word->set_text (unicharset_boxes.id_to_unichar(uch_id));
                  if (polyg)
                    new_blob = new PBLOB;
                  else
                    new_blob = (PBLOB *) new C_BLOB;
                  new_blob_it.set_to_list (new_word->
                    gblob_list ());
                  new_blob_it.add_to_end (new_blob);
                  new_outline_it.
                    set_to_list (gblob_out_list
                    (new_blob, polyg));
                }
                new_outline_it.add_to_end (outline_it.
                  extract ());
                //move blob
              }
            }
          }
                                 //no outlines in blob
          if (outline_it.empty ())
                                 //so delete blob
            delete blob_it.extract ();
        }
      }
      if (blob_it.empty ())      //no blobs in word
                                 //so delete word
          delete word_it.extract ();
    }
  }
  if (error_count > 0)
    return error_count;

  if (new_word != NULL) {
    gblob_sort_list (new_word->gblob_list (), polyg);
    word_it.add_to_end (new_word);
    new_word_box = new_word->bounding_box ();
    word_x_centre = (new_word_box.left () + new_word_box.right ()) / 2.0f;
    baseline = row->base_line (word_x_centre);

#if 0
    if (strlen(unicharset_boxes.id_to_unichar(uch_id)) == 1) {
      if (STRING (chs_caps_ht).contains (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          (new_word_box.top () <
           baseline + (1 + applybox_error_band) * row->x_height ())) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
                           "FAILURE! caps-ht char didn't ascend");
        new_word->set_text ("");
        return 1;
      }
      if (STRING (chs_odd_top).contains (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          (new_word_box.top () <
           baseline + (1 - applybox_error_band) * row->x_height ())) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
                           "FAILURE! Odd top char below xht");
        new_word->set_text ("");
        return 1;
      }
      if (STRING (chs_x_ht).contains (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          ((new_word_box.top () >
            baseline + (1 + applybox_error_band) * row->x_height ()) ||
           (new_word_box.top () <
            baseline + (1 - applybox_error_band) * row->x_height ()))) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
                           "FAILURE! x-ht char didn't have top near xht");
        new_word->set_text ("");
        return 1;
      }
      if (STRING (chs_non_ambig_bl).contains
          (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          ((new_word_box.bottom () <
            baseline - applybox_error_band * row->x_height ()) ||
           (new_word_box.bottom () >
            baseline + applybox_error_band * row->x_height ()))) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
                           "FAILURE! non ambig BL char didnt have bottom near baseline");
        new_word->set_text ("");
        return 1;
      }
      if (STRING (chs_odd_bot).contains (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          (new_word_box.bottom () >
           baseline + applybox_error_band * row->x_height ())) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
                           "FAILURE! Odd bottom char above baseline");
        new_word->set_text ("");
        return 1;
      }
      if (STRING (chs_desc).contains (unicharset_boxes.id_to_unichar(uch_id)[0]) &&
          (new_word_box.bottom () >
           baseline - applybox_error_band * row->x_height ())) {
        report_failed_box (boxfile_lineno, boxfile_charno, box,
                           unicharset_boxes.id_to_unichar(uch_id),
        "FAILURE! Descender doesn't descend");
        new_word->set_text ("");
        return 1;
      }
    }
#endif
    return 0;
  }
  else {
    report_failed_box (boxfile_lineno, boxfile_charno, box,
                       unicharset_boxes.id_to_unichar(uch_id),
                       "FAILURE! Couldn't find any blobs");
    return 1;
  }
}


/*************************************************************************
 * tidy_up()
 *   - report >1 block
 *   - sort the words in each row.
 *   - report any rows with no labelled words.
 *   - report any remaining unlabelled words
 *		- report total labelled words
 *
 *************************************************************************/
void tidy_up(                         //
             BLOCK_LIST *block_list,  //real blocks
             inT16 &ok_char_count,
             inT16 &ok_row_count,
             inT16 &unlabelled_words,
             inT16 *tgt_char_counts,
             inT16 &rebalance_count,
             UNICHAR_ID *min_uch_id,
             inT16 &min_samples,
             inT16 &final_labelled_blob_count) {
  BLOCK_IT block_it(block_list);
  ROW_IT row_it;
  ROW *row;
  WERD_IT word_it;
  WERD *word;
  WERD *duplicate_word;
  inT16 block_idx = 0;
  inT16 row_idx;
  inT16 all_row_idx = 0;
  BOOL8 row_ok;
  BOOL8 rebalance_needed = FALSE;
                                 //No. of unique labelled samples
  inT16 labelled_char_counts[MAX_NUM_CLASSES];
  inT16 i;
  UNICHAR_ID uch_id;
  UNICHAR_ID prev_uch_id = -1;
  BOOL8 at_dupe_of_prev_word;
  ROW *prev_row = NULL;
  inT16 left;
  inT16 prev_left = -1;

  for (i = 0; i < MAX_NUM_CLASSES; i++)
    labelled_char_counts[i] = 0;

  ok_char_count = 0;
  ok_row_count = 0;
  unlabelled_words = 0;
  if ((applybox_debug > 4) && (block_it.length () != 1))

    tprintf ("APPLY_BOXES: More than one block??\n");

  for (block_it.mark_cycle_pt ();
  !block_it.cycled_list (); block_it.forward ()) {
    block_idx++;
    row_idx = 0;
    row_ok = FALSE;
    row_it.set_to_list (block_it.data ()->row_list ());
    for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
      row_idx++;
      all_row_idx++;
      row = row_it.data ();
      word_it.set_to_list (row->word_list ());
      word_it.sort (word_comparator);
      for (word_it.mark_cycle_pt ();
      !word_it.cycled_list (); word_it.forward ()) {
        word = word_it.data ();
        if (strlen (word->text ()) == 0) {
          unlabelled_words++;
          if (applybox_debug > 4) {
            tprintf
              ("APPLY_BOXES: Unlabelled word blk:%d row:%d allrows:%d\n",
              block_idx, row_idx, all_row_idx);
          }
        }
        else {
          if (word->gblob_list ()->length () != 1)
            tprintf
              ("APPLY_BOXES: FATALITY - MULTIBLOB Labelled word blk:%d row:%d allrows:%d\n",
              block_idx, row_idx, all_row_idx);

          ok_char_count++;
          labelled_char_counts[unicharset_boxes.unichar_to_id(word->text ())]++;
          row_ok = TRUE;
        }
      }
      if ((applybox_debug > 4) && (!row_ok)) {
        tprintf
          ("APPLY_BOXES: Row with no labelled words blk:%d row:%d allrows:%d\n",
          block_idx, row_idx, all_row_idx);