tospace.cpp

一ＯＣＲ的相关资料。．希望对研究ＯＣＲ的朋友有所帮助．

  float kern_estimate;
  float crude_threshold_estimate;
  INT16 small_gaps_count;
  INT16 total;
                                 //iterator
  BLOBNBOX_IT blob_it = row->blob_list ();
  STATS cert_space_gap_stats (0, MAXSPACING);
  STATS all_space_gap_stats (0, MAXSPACING);
  STATS small_gap_stats (0, MAXSPACING);
  BOX blob_box;
  BOX prev_blob_box;
  INT16 gap_width;
  INT32 end_of_row;
  INT32 row_length;

  kern_estimate = all_gap_stats->median ();
  crude_threshold_estimate = MAX (tosp_init_guess_kn_mult * kern_estimate,
    tosp_init_guess_xht_mult * row->xheight);
  small_gaps_count = stats_count_under (all_gap_stats,
    (INT16)
    ceil (crude_threshold_estimate));
  total = all_gap_stats->get_total ();

  if ((total <= tosp_redo_kern_limit) ||
    ((small_gaps_count / (float) total) < tosp_enough_small_gaps) ||
  (total - small_gaps_count < 1)) {
    if (tosp_debug_level > 5)
      tprintf ("B:%d R:%d -- Cant do isolated row stats.\n",
        block_idx, row_idx);
    return FALSE;
  }
  blob_it.set_to_list (row->blob_list ());
  blob_it.mark_cycle_pt ();
  end_of_row = blob_it.data_relative (-1)->bounding_box ().right ();
  if (tosp_use_pre_chopping)
    blob_box = box_next_pre_chopped (&blob_it);
  else if (tosp_stats_use_xht_gaps)
    blob_box = reduced_box_next (row, &blob_it);
  else
    blob_box = box_next (&blob_it);
  row_length = end_of_row - blob_box.left ();
  prev_blob_box = blob_box;
  while (!blob_it.cycled_list ()) {
    if (tosp_use_pre_chopping)
      blob_box = box_next_pre_chopped (&blob_it);
    else if (tosp_stats_use_xht_gaps)
      blob_box = reduced_box_next (row, &blob_it);
    else
      blob_box = box_next (&blob_it);
    gap_width = blob_box.left () - prev_blob_box.right ();
    if (!ignore_big_gap (row, row_length, gapmap,
      prev_blob_box.right (), blob_box.left ()) &&
    (gap_width > crude_threshold_estimate)) {
      if ((gap_width > tosp_fuzzy_space_factor2 * row->xheight) ||
        ((gap_width > tosp_fuzzy_space_factor1 * row->xheight) &&
        (!tosp_narrow_blobs_not_cert ||
        (!narrow_blob (row, prev_blob_box) &&
        !narrow_blob (row, blob_box)))) ||
        (wide_blob (row, prev_blob_box) && wide_blob (row, blob_box)))
        cert_space_gap_stats.add (gap_width, 1);
      all_space_gap_stats.add (gap_width, 1);
    }
    if (gap_width < crude_threshold_estimate)
      small_gap_stats.add (gap_width, 1);

    prev_blob_box = blob_box;
  }
  if (cert_space_gap_stats.get_total () >=
    tosp_enough_space_samples_for_median)
                                 //median
    row->space_size = cert_space_gap_stats.median ();
  else if (suspected_table && (cert_space_gap_stats.get_total () > 0))
                                 //to avoid spaced
    row->space_size = cert_space_gap_stats.mean ();
  //      1's in tables
  else if (all_space_gap_stats.get_total () >=
    tosp_enough_space_samples_for_median)
                                 //median
    row->space_size = all_space_gap_stats.median ();
  else
    row->space_size = all_space_gap_stats.mean ();

  if (tosp_only_small_gaps_for_kern)
    row->kern_size = small_gap_stats.median ();
  else
    row->kern_size = all_gap_stats->median ();
  row->space_threshold =
    INT32 (floor ((row->space_size + row->kern_size) / 2));
  /* Sanity check */
  if ((row->kern_size >= row->space_threshold) ||
    (row->space_threshold >= row->space_size) ||
  (row->space_threshold <= 0)) {
    if (tosp_debug_level > 0)
      tprintf ("B:%d R:%d -- Isolated row stats SANITY FAILURE: %f %d %f\n",
        block_idx, row_idx,
        row->kern_size, row->space_threshold, row->space_size);
    row->kern_size = 0.0f;
    row->space_threshold = 0;
    row->space_size = 0.0f;
    return FALSE;
  }

  if (tosp_debug_level > 5)
    tprintf ("B:%d R:%d -- Isolated row stats: %f %d %f\n",
      block_idx, row_idx,
      row->kern_size, row->space_threshold, row->space_size);
  return TRUE;
}


INT16 stats_count_under(STATS *stats, INT16 threshold) {
  INT16 index;
  INT16 total = 0;

  for (index = 0; index < threshold; index++)
    total += stats->pile_count (index);
  return total;
}


/*************************************************************************
 * improve_row_threshold()
 *    Try to recognise a "normal line" -
 *           > 25 gaps
 *     &&    space > 3 * kn  && space > 10
 *              (I.e. reasonably large space and kn:sp ratio)
 *     &&    > 3/4 # gaps < kn + (sp - kn)/3
 *              (I.e. most gaps are well away from space estimate)
 *     &&    a gap of max( 3, (sp - kn)/3 ) empty histogram positions is found
 *           somewhere in the histogram between kn and sp
 *     THEN set the threshold and fuzzy limits to this gap - ie NO fuzzies
 *          NO!!!!! the bristol line has "11" with a gap of 12 between the 1's!!!
 *          try moving the default threshold to within this band but leave the
 *          fuzzy limit calculation as at present.
 *************************************************************************/

void improve_row_threshold(TO_ROW *row, STATS *all_gap_stats) {
  float sp = row->space_size;
  float kn = row->kern_size;
  INT16 reqd_zero_width = 0;
  INT16 zero_width = 0;
  INT16 zero_start = 0;
  INT16 index = 0;

  if (tosp_debug_level > 10)
    tprintf ("Improve row threshold 0");
  if ((all_gap_stats->get_total () <= 25) ||
    (sp <= 10) ||
    (sp <= 3 * kn) ||
    (stats_count_under (all_gap_stats,
    (INT16) ceil (kn + (sp - kn) / 3 + 0.5)) <
    (0.75 * all_gap_stats->get_total ())))
    return;
  if (tosp_debug_level > 10)
    tprintf (" 1");
  /*
  Look for the first region of all 0's in the histogram which is wider than
  max( 3, (sp - kn)/3 ) and starts between kn and sp. If found, and current
  threshold is not within it, move the threshold so that is is just inside it.
  */
  reqd_zero_width = (INT16) floor ((sp - kn) / 3 + 0.5);
  if (reqd_zero_width < 3)
    reqd_zero_width = 3;

  for (index = INT16 (ceil (kn)); index < INT16 (floor (sp)); index++) {
    if (all_gap_stats->pile_count (index) == 0) {
      if (zero_width == 0)
        zero_start = index;
      zero_width++;
    }
    else {
      if (zero_width >= reqd_zero_width)
        break;
      else {
        zero_width = 0;
      }
    }
  }
  index--;
  if (tosp_debug_level > 10)
    tprintf (" reqd_z_width: %d found %d 0's, starting %d; thresh: %d/n",
      reqd_zero_width, zero_width, zero_start, row->space_threshold);
  if ((zero_width < reqd_zero_width) ||
    ((row->space_threshold >= zero_start) &&
    (row->space_threshold <= index)))
    return;
  if (tosp_debug_level > 10)
    tprintf (" 2");
  if (row->space_threshold < zero_start) {
    if (tosp_debug_level > 5)
      tprintf
        ("Improve row kn:%5.2f sp:%5.2f 0's: %d -> %d  thresh:%d -> %d\n",
        kn, sp, zero_start, index, row->space_threshold, zero_start);
    row->space_threshold = zero_start;
  }
  if (row->space_threshold > index) {
    if (tosp_debug_level > 5)
      tprintf
        ("Improve row kn:%5.2f sp:%5.2f 0's: %d -> %d  thresh:%d -> %d\n",
        kn, sp, zero_start, index, row->space_threshold, index);
    row->space_threshold = index;
  }
}


/**********************************************************************
 * make_prop_words
 *
 * Convert a TO_BLOCK to a BLOCK.
 **********************************************************************/

ROW *make_prop_words(                 //find lines
                     TO_ROW *row,     //row to make
                     FCOORD rotation  //for drawing
                    ) {
  BOOL8 bol;                     //start of line
  /* prev_ values are for start of word being built. non prev_ values are for
  the gap between the word being built and the next one. */
  BOOL8 prev_fuzzy_sp;           //probably space
  BOOL8 prev_fuzzy_non;          //probably not
  UINT8 prev_blanks;             //in front of word
  BOOL8 fuzzy_sp;                //probably space
  BOOL8 fuzzy_non;               //probably not
  UINT8 blanks;                  //in front of word
  ROW *real_row;                 //output row
  OUTLINE_IT out_it;             //outlines
  C_OUTLINE_IT cout_it;
  PBLOB_LIST blobs;              //blobs in word
  C_BLOB_LIST cblobs;
  PBLOB_IT blob_it = &blobs;     //iterator
  C_BLOB_IT cblob_it = &cblobs;
  WERD_LIST words;
  WERD_IT word_it;               //new words
  WERD *word;                    //new word
  WERD_IT rep_char_it;           //repeated char words
  INT32 next_rep_char_word_right = MAX_INT32;
  float repetition_spacing;      //gap between repetitions
  INT32 xstarts[2];              //row ends
  double coeffs[3];              //quadratic
  INT32 prev_x;                  //end of prev blob
  BLOBNBOX *bblob;               //current blob
  BOX blob_box;                  //bounding box
  BLOBNBOX_IT box_it;            //iterator
  BOX prev_blob_box;
  BOX next_blob_box;
  INT16 prev_gap = MAX_INT16;
  INT16 current_gap = MAX_INT16;
  INT16 next_gap = MAX_INT16;
  INT16 prev_within_xht_gap = MAX_INT16;
  INT16 current_within_xht_gap = MAX_INT16;
  INT16 next_within_xht_gap = MAX_INT16;
  INT16 word_count = 0;
  static INT16 row_count = 0;

  row_count++;
  rep_char_it.set_to_list (&(row->rep_words));
  if (!rep_char_it.empty ()) {
    next_rep_char_word_right =
      rep_char_it.data ()->bounding_box ().right ();
  }

  prev_x = -MAX_INT16;
  blob_it.set_to_list (&blobs);
  cblob_it.set_to_list (&cblobs);
  box_it.set_to_list (row->blob_list ());
  word_it.set_to_list (&words);
  bol = TRUE;
  prev_blanks = 0;
  prev_fuzzy_sp = FALSE;
  prev_fuzzy_non = FALSE;
  if (!box_it.empty ()) {
    xstarts[0] = box_it.data ()->bounding_box ().left ();
    if (xstarts[0] > next_rep_char_word_right) {
      /* We need to insert a repeated char word at the start of the row */
      word = rep_char_it.extract ();
      word_it.add_after_then_move (word);
      /* Set spaces before repeated char word */
      word->set_flag (W_BOL, TRUE);
      bol = FALSE;
      word->set_blanks (0);
                                 //NO uncertainty
      word->set_flag (W_FUZZY_SP, FALSE);
      word->set_flag (W_FUZZY_NON, FALSE);
      xstarts[0] = word->bounding_box ().left ();
      /* Set spaces after repeated char word (and leave current word set) */
      repetition_spacing = find_mean_blob_spacing (word);
      current_gap = box_it.data ()->bounding_box ().left () -
        next_rep_char_word_right;
      current_within_xht_gap = current_gap;
      if (current_gap > tosp_rep_space * repetition_spacing) {
        prev_blanks = (UINT8) floor (current_gap / row->space_size);
        if (prev_blanks < 1)
          prev_blanks = 1;
      }
      else
        prev_blanks = 0;
      if (tosp_debug_level > 5)
        tprintf ("Repch wd at BOL(%d, %d). rep spacing %5.2f;  Rgap:%d  ",
          box_it.data ()->bounding_box ().left (),
          box_it.data ()->bounding_box ().bottom (),
          repetition_spacing, current_gap);
      prev_fuzzy_sp = FALSE;
      prev_fuzzy_non = FALSE;
      if (rep_char_it.empty ()) {
        next_rep_char_word_right = MAX_INT32;
      }
      else {
        rep_char_it.forward ();
        next_rep_char_word_right =
          rep_char_it.data ()->bounding_box ().right ();
      }
    }

    peek_at_next_gap(row,
                     box_it,
                     next_blob_box,
                     next_gap,
                     next_within_xht_gap);
    do {
      bblob = box_it.data ();
      blob_box = bblob->bounding_box ();
      if (bblob->joined_to_prev ()) {
        if (bblob->blob () != NULL) {
          out_it.set_to_list (blob_it.data ()->out_list ());
          out_it.move_to_last ();
          out_it.add_list_after (bblob->blob ()->out_list ());
          delete bblob->blob ();
        }
        else if (bblob->cblob () != NULL) {
          cout_it.set_to_list (cblob_it.data ()->out_list ());
          cout_it.move_to_last ();
          cout_it.add_list_after (bblob->cblob ()->out_list ());
          delete bblob->cblob ();
        }
      }
      else {
        if (bblob->blob () != NULL)
          blob_it.add_after_then_move (bblob->blob ());
        else if (bblob->cblob () != NULL)
          cblob_it.add_after_then_move (bblob->cblob ());
        prev_x = blob_box.right ();
      }
      box_it.forward ();         //next one
      bblob = box_it.data ();
      blob_box = bblob->bounding_box ();

      if (!bblob->joined_to_prev () &&
      (bblob->blob () != NULL || bblob->cblob () != NULL)) {
        /* Real Blob - not multiple outlines or pre-chopped */
        prev_gap = current_gap;
        prev_within_xht_gap = current_within_xht_gap;
        prev_blob_box = next_blob_box;
        current_gap = next_gap;
        current_within_xht_gap = next_within_xht_gap;
        peek_at_next_gap(row,
                         box_it,
                         next_blob_box,
                         next_gap,
                         next_within_xht_gap);

        if ((blob_box.left () > next_rep_char_word_right) ||
          (!tosp_only_use_xht_gaps &&