topitch.cpp

一ＯＣＲ的相关资料。．希望对研究ＯＣＲ的朋友有所帮助．

                       INT32 &corr_prop,
                       INT32 &dunno) {
  TO_ROW *row;                   //current row
  TO_ROW_IT row_it = block->get_rows ();

  for (row_it.mark_cycle_pt (); !row_it.cycled_list (); row_it.forward ()) {
    row = row_it.data ();
    switch (row->pitch_decision) {
      case PITCH_DUNNO:
        dunno++;
        break;
      case PITCH_DEF_PROP:
        def_prop++;
        break;
      case PITCH_MAYBE_PROP:
        maybe_prop++;
        break;
      case PITCH_DEF_FIXED:
        def_fixed++;
        break;
      case PITCH_MAYBE_FIXED:
        maybe_fixed++;
        break;
      case PITCH_CORR_PROP:
        corr_prop++;
        break;
      case PITCH_CORR_FIXED:
        corr_fixed++;
        break;
    }
  }
}


/**********************************************************************
 * row_pitch_stats
 *
 * Decide whether each row is fixed pitch individually.
 **********************************************************************/

BOOL8 row_pitch_stats(                  //find line stats
                      TO_ROW *row,      //current row
                      INT32 maxwidth,   //of spaces
                      BOOL8 testing_on  //correct orientation
                     ) {
  BLOBNBOX *blob;                //current blob
  int gap_index;                 //current gap
  INT32 prev_x;                  //end of prev blob
  INT32 cluster_count;           //no of clusters
  INT32 prev_count;              //of clusters
  INT32 smooth_factor;           //for smoothing stats
  BOX blob_box;                  //bounding box
  float lower, upper;            //cluster thresholds
                                 //gap sizes
  float gaps[BLOCK_STATS_CLUSTERS];
                                 //blobs
  BLOBNBOX_IT blob_it = row->blob_list ();
  STATS gap_stats (0, maxwidth);
  STATS cluster_stats[BLOCK_STATS_CLUSTERS + 1];
  //clusters

  smooth_factor =
    (INT32) (row->xheight * textord_wordstats_smooth_factor + 1.5);
  if (!blob_it.empty ()) {
    prev_x = blob_it.data ()->bounding_box ().right ();
    blob_it.forward ();
    while (!blob_it.at_first ()) {
      blob = blob_it.data ();
      if (!blob->joined_to_prev ()) {
        blob_box = blob->bounding_box ();
        if (blob_box.left () - prev_x < maxwidth)
          gap_stats.add (blob_box.left () - prev_x, 1);
        prev_x = blob_box.right ();
      }
      blob_it.forward ();
    }
  }
  if (gap_stats.get_total () == 0) {
    return FALSE;
  }
  cluster_count = 0;
  lower = row->xheight * words_initial_lower;
  upper = row->xheight * words_initial_upper;
  gap_stats.smooth (smooth_factor);
  do {
    prev_count = cluster_count;
    cluster_count = gap_stats.cluster (lower, upper,
      textord_spacesize_ratioprop,
      BLOCK_STATS_CLUSTERS, cluster_stats);
  }
  while (cluster_count > prev_count && cluster_count < BLOCK_STATS_CLUSTERS);
  if (cluster_count < 1) {
    return FALSE;
  }
  for (gap_index = 0; gap_index < cluster_count; gap_index++)
    gaps[gap_index] = cluster_stats[gap_index + 1].ile (0.5);
  //get medians
  if (testing_on) {
    tprintf ("cluster_count=%d:", cluster_count);
    for (gap_index = 0; gap_index < cluster_count; gap_index++)
      tprintf (" %g(%d)", gaps[gap_index],
        cluster_stats[gap_index + 1].get_total ());
    tprintf ("\n");
  }
  qsort (gaps, cluster_count, sizeof (float), sort_floats2);

  //Try to find proportional non-space and space for row.
  lower = row->xheight * words_default_prop_nonspace;
  upper = row->xheight * textord_words_min_minspace;
  for (gap_index = 0; gap_index < cluster_count
    && gaps[gap_index] < lower; gap_index++);
  if (gap_index == 0) {
    if (testing_on)
      tprintf ("No clusters below nonspace threshold!!\n");
    if (cluster_count > 1) {
      row->pr_nonsp = gaps[0];
      row->pr_space = gaps[1];
    }
    else {
      row->pr_nonsp = lower;
      row->pr_space = gaps[0];
    }
  }
  else {
    row->pr_nonsp = gaps[gap_index - 1];
    while (gap_index < cluster_count && gaps[gap_index] < upper)
      gap_index++;
    if (gap_index == cluster_count) {
      if (testing_on)
        tprintf ("No clusters above nonspace threshold!!\n");
      row->pr_space = lower * textord_spacesize_ratioprop;
    }
    else
      row->pr_space = gaps[gap_index];
  }

  //Now try to find the fixed pitch space and non-space.
  upper = row->xheight * words_default_fixed_space;
  for (gap_index = 0; gap_index < cluster_count
    && gaps[gap_index] < upper; gap_index++);
  if (gap_index == 0) {
    if (testing_on)
      tprintf ("No clusters below space threshold!!\n");
    row->fp_nonsp = upper;
    row->fp_space = gaps[0];
  }
  else {
    row->fp_nonsp = gaps[gap_index - 1];
    if (gap_index == cluster_count) {
      if (testing_on)
        tprintf ("No clusters above space threshold!!\n");
      row->fp_space = row->xheight;
    }
    else
      row->fp_space = gaps[gap_index];
  }
  if (testing_on) {
    tprintf
      ("Initial estimates:pr_nonsp=%g, pr_space=%g, fp_nonsp=%g, fp_space=%g\n",
      row->pr_nonsp, row->pr_space, row->fp_nonsp, row->fp_space);
  }
  return TRUE;                   //computed some stats
}


/**********************************************************************
 * find_row_pitch
 *
 * Check to see if this row could be fixed pitch using the given spacings.
 * Blobs with gaps smaller than the lower threshold are assumed to be one.
 * The larger threshold is the word gap threshold.
 **********************************************************************/

BOOL8 find_row_pitch(                    //find lines
                     TO_ROW *row,        //row to do
                     INT32 maxwidth,     //max permitted space
                     INT32 dm_gap,       //ignorable gaps
                     TO_BLOCK *block,    //block of row
                     INT32 block_index,  //block_number
                     INT32 row_index,    //number of row
                     BOOL8 testing_on    //correct orientation
                    ) {
  BOOL8 used_dm_model;           //looks lik dot matrix
  float min_space;               //estimate threshold
  float non_space;               //gap size
  float gap_iqr;                 //interquartile range
  float pitch_iqr;
  float dm_gap_iqr;              //interquartile range
  float dm_pitch_iqr;
  float dm_pitch;                //pitch with dm on
  float pitch;                   //revised estimate
  float initial_pitch;           //guess at pitch
  STATS gap_stats (0, maxwidth);
                                 //centre-centre
  STATS pitch_stats (0, maxwidth);

  row->fixed_pitch = 0.0f;
  initial_pitch = row->fp_space;
  if (initial_pitch > row->xheight * (1 + words_default_fixed_limit))
    initial_pitch = row->xheight;//keep pitch decent
  non_space = row->fp_nonsp;
  if (non_space > initial_pitch)
    non_space = initial_pitch;
  min_space = (initial_pitch + non_space) / 2;

  if (!count_pitch_stats (row, &gap_stats, &pitch_stats,
  initial_pitch, min_space, TRUE, FALSE, dm_gap)) {
    dm_gap_iqr = 0.0001;
    dm_pitch_iqr = maxwidth * 2.0f;
    dm_pitch = initial_pitch;
  }
  else {
    dm_gap_iqr = gap_stats.ile (0.75) - gap_stats.ile (0.25);
    dm_pitch_iqr = pitch_stats.ile (0.75) - pitch_stats.ile (0.25);
    dm_pitch = pitch_stats.ile (0.5);
  }
  gap_stats.clear ();
  pitch_stats.clear ();
  if (!count_pitch_stats (row, &gap_stats, &pitch_stats,
  initial_pitch, min_space, TRUE, FALSE, 0)) {
    gap_iqr = 0.0001;
    pitch_iqr = maxwidth * 3.0f;
  }
  else {
    gap_iqr = gap_stats.ile (0.75) - gap_stats.ile (0.25);
    pitch_iqr = pitch_stats.ile (0.75) - pitch_stats.ile (0.25);
    if (testing_on)
      tprintf
        ("First fp iteration:initial_pitch=%g, gap_iqr=%g, pitch_iqr=%g, pitch=%g\n",
        initial_pitch, gap_iqr, pitch_iqr, pitch_stats.ile (0.5));
    initial_pitch = pitch_stats.ile (0.5);
    if (min_space > initial_pitch
      && count_pitch_stats (row, &gap_stats, &pitch_stats,
    initial_pitch, initial_pitch, TRUE, FALSE, 0)) {
      min_space = initial_pitch;
      gap_iqr = gap_stats.ile (0.75) - gap_stats.ile (0.25);
      pitch_iqr = pitch_stats.ile (0.75) - pitch_stats.ile (0.25);
      if (testing_on)
        tprintf
          ("Revised fp iteration:initial_pitch=%g, gap_iqr=%g, pitch_iqr=%g, pitch=%g\n",
          initial_pitch, gap_iqr, pitch_iqr, pitch_stats.ile (0.5));
      initial_pitch = pitch_stats.ile (0.5);
    }
  }
  if (textord_debug_pitch_metric)
    tprintf ("Blk=%d:Row=%d:%c:p_iqr=%g:g_iqr=%g:dm_p_iqr=%g:dm_g_iqr=%g:%c:",
      block_index, row_index,
      block->block->text_region () != NULL ?
      (block->block->text_region ()->is_prop ()? 'P' : 'F') : 'X',
    pitch_iqr, gap_iqr, dm_pitch_iqr, dm_gap_iqr,
    pitch_iqr > maxwidth && dm_pitch_iqr > maxwidth ? 'D'
    : (pitch_iqr * dm_gap_iqr <=
    dm_pitch_iqr * gap_iqr ? 'S' : 'M'));
  if (pitch_iqr > maxwidth && dm_pitch_iqr > maxwidth) {
    row->pitch_decision = PITCH_DUNNO;
    if (textord_debug_pitch_metric)
      tprintf ("\n");
    return FALSE;                //insufficient data
  }
  if (pitch_iqr * dm_gap_iqr <= dm_pitch_iqr * gap_iqr) {
    if (testing_on)
      tprintf
        ("Choosing non dm version:pitch_iqr=%g, gap_iqr=%g, dm_pitch_iqr=%g, dm_gap_iqr=%g\n",
        pitch_iqr, gap_iqr, dm_pitch_iqr, dm_gap_iqr);
    gap_iqr = gap_stats.ile (0.75) - gap_stats.ile (0.25);
    pitch_iqr = pitch_stats.ile (0.75) - pitch_stats.ile (0.25);
    pitch = pitch_stats.ile (0.5);
    used_dm_model = FALSE;
  }
  else {
    if (testing_on)
      tprintf
        ("Choosing dm version:pitch_iqr=%g, gap_iqr=%g, dm_pitch_iqr=%g, dm_gap_iqr=%g\n",
        pitch_iqr, gap_iqr, dm_pitch_iqr, dm_gap_iqr);
    gap_iqr = dm_gap_iqr;
    pitch_iqr = dm_pitch_iqr;
    pitch = dm_pitch;
    used_dm_model = TRUE;
  }
  if (textord_debug_pitch_metric) {
    tprintf ("rev_p_iqr=%g:rev_g_iqr=%g:pitch=%g:",
      pitch_iqr, gap_iqr, pitch);
    tprintf ("p_iqr/g=%g:p_iqr/x=%g:iqr_res=%c:",
      pitch_iqr / gap_iqr, pitch_iqr / block->xheight,
      pitch_iqr < gap_iqr * textord_fpiqr_ratio
      && pitch_iqr < block->xheight * textord_max_pitch_iqr
      && pitch < block->xheight * textord_words_default_maxspace
      ? 'F' : 'P');
  }
  if (pitch_iqr < gap_iqr * textord_fpiqr_ratio
    && pitch_iqr < block->xheight * textord_max_pitch_iqr
    && pitch < block->xheight * textord_words_default_maxspace)
    row->pitch_decision = PITCH_MAYBE_FIXED;
  else
    row->pitch_decision = PITCH_MAYBE_PROP;
  row->fixed_pitch = pitch;
  row->kern_size = gap_stats.ile (0.5);
  row->min_space = (INT32) (row->fixed_pitch + non_space) / 2;
  if (row->min_space > row->fixed_pitch)
    row->min_space = (INT32) row->fixed_pitch;
  row->max_nonspace = row->min_space;
  row->space_size = row->fixed_pitch;
  row->space_threshold = (row->max_nonspace + row->min_space) / 2;
  row->used_dm_model = used_dm_model;
  return TRUE;
}


/**********************************************************************
 * fixed_pitch_row
 *
 * Check to see if this row could be fixed pitch using the given spacings.
 * Blobs with gaps smaller than the lower threshold are assumed to be one.
 * The larger threshold is the word gap threshold.
 **********************************************************************/

BOOL8 fixed_pitch_row(                   //find lines
                      TO_ROW *row,       //row to do
                      INT32 block_index  //block_number
                     ) {
  const char *res_string;        //pitch result
  INT16 mid_cuts;                //no of cheap cuts
  float non_space;               //gap size
  float pitch_sd;                //error on pitch
  float sp_sd;                   //space sd

  non_space = row->fp_nonsp;
  if (non_space > row->fixed_pitch)
    non_space = row->fixed_pitch;
  if (textord_all_prop) {
    // Set the decision to definitely proportional.
    pitch_sd = textord_words_def_prop * row->fixed_pitch;
    row->pitch_decision = PITCH_DEF_PROP;
  } else {
    pitch_sd = tune_row_pitch (row, &row->projection, row->projection_left,
                               row->projection_right,
                               (row->fixed_pitch + non_space * 3) / 4,
                               row->fixed_pitch, sp_sd, mid_cuts,
                               &row->char_cells,
                               block_index == textord_debug_block);
    if (pitch_sd < textord_words_pitchsd_threshold * row->fixed_pitch
      && ((pitsync_linear_version & 3) < 3
      || (pitsync_linear_version & 3) >= 3 && (row->used_dm_model
      || sp_sd > 20
      || pitch_sd == 0
    && sp_sd > 10))) {
      if (pitch_sd < textord_words_def_fixed * row->fixed_pitch
        && !row->all_caps
        && ((pitsync_linear_version & 3) < 3 || sp_sd > 20))
        row->pitch_decision = PITCH_DEF_FIXED;
      else
        row->pitch_decision = PITCH_MAYBE_FIXED;
    }