Tesseract  3.02
tesseract-ocr/textord/pitsync1.cpp File Reference
#include "mfcpch.h"
#include <math.h>
#include "memry.h"
#include "pitsync1.h"
#include "notdll.h"

Go to the source code of this file.

Defines

#define EXTERN

Functions

 ELISTIZE (FPSEGPT) CLISTIZE(FPSEGPT_LIST) EXTERN int pitsync_linear_version
double check_pitch_sync (BLOBNBOX_IT *blob_it, inT16 blob_count, inT16 pitch, inT16 pitch_error, STATS *projection, FPSEGPT_LIST *seg_list)
void make_illegal_segment (FPSEGPT_LIST *prev_list, TBOX blob_box, BLOBNBOX_IT blob_it, inT16 region_index, inT16 pitch, inT16 pitch_error, FPSEGPT_LIST *seg_list)

Variables

EXTERN double pitsync_joined_edge = 0.75
EXTERN double pitsync_offset_freecut_fraction = 0.25
EXTERN int pitsync_fake_depth = 1

Define Documentation

#define EXTERN

Function Documentation

double check_pitch_sync ( BLOBNBOX_IT *  blob_it,
inT16  blob_count,
inT16  pitch,
inT16  pitch_error,
STATS projection,
FPSEGPT_LIST *  seg_list 
)

Definition at line 151 of file pitsync1.cpp.

                         {
  inT16 x;                       //current coord
  inT16 min_index;               //blob number
  inT16 max_index;               //blob number
  inT16 left_edge;               //of word
  inT16 right_edge;              //of word
  inT16 right_max;               //max allowed x
  inT16 min_x;                   //in this region
  inT16 max_x;
  inT16 region_index;
  inT16 best_region_index = 0;   //for best result
  inT16 offset;                  //dist to legal area
  inT16 left_best_x;             //edge of good region
  inT16 right_best_x;            //right edge
  TBOX min_box;                   //bounding box
  TBOX max_box;                   //bounding box
  TBOX next_box;                  //box of next blob
  FPSEGPT *segpt;                //segment point
  FPSEGPT_LIST *segpts;          //points in a segment
  double best_cost;              //best path
  double mean_sum;               //computes result
  FPSEGPT *best_end;             //end of best path
  BLOBNBOX_IT min_it;            //copy iterator
  BLOBNBOX_IT max_it;            //copy iterator
  FPSEGPT_IT segpt_it;           //iterator
                                 //output segments
  FPSEGPT_IT outseg_it = seg_list;
  FPSEGPT_LIST_CLIST lattice;    //list of lists
                                 //region iterator
  FPSEGPT_LIST_C_IT lattice_it = &lattice;

  //      tprintf("Computing sync on word of %d blobs with pitch %d\n",
  //              blob_count, pitch);
  //      if (blob_count==8 && pitch==27)
  //              projection->print(stdout,TRUE);
  if (pitch < 3)
    pitch = 3;                   //nothing ludicrous
  if ((pitch - 3) / 2 < pitch_error)
    pitch_error = (pitch - 3) / 2;
  min_it = *blob_it;
  min_box = box_next (&min_it);  //get box
  //      if (blob_count==8 && pitch==27)
  //              tprintf("1st box at (%d,%d)->(%d,%d)\n",
  //                      min_box.left(),min_box.bottom(),
  //                      min_box.right(),min_box.top());
                                 //left of word
  left_edge = min_box.left () + pitch_error;
  for (min_index = 1; min_index < blob_count; min_index++) {
    min_box = box_next (&min_it);
    //              if (blob_count==8 && pitch==27)
    //                      tprintf("Box at (%d,%d)->(%d,%d)\n",
    //                              min_box.left(),min_box.bottom(),
    //                              min_box.right(),min_box.top());
  }
  right_edge = min_box.right (); //end of word
  max_x = left_edge;
                                 //min permissible
  min_x = max_x - pitch + pitch_error * 2 + 1;
  right_max = right_edge + pitch - pitch_error - 1;
  segpts = new FPSEGPT_LIST;     //list of points
  segpt_it.set_to_list (segpts);
  for (x = min_x; x <= max_x; x++) {
    segpt = new FPSEGPT (x);     //make a new one
                                 //put in list
    segpt_it.add_after_then_move (segpt);
  }
                                 //first segment
  lattice_it.add_before_then_move (segpts);
  min_index = 0;
  region_index = 1;
  best_cost = MAX_FLOAT32;
  best_end = NULL;
  min_it = *blob_it;
  min_box = box_next (&min_it);  //first box
  do {
    left_best_x = -1;
    right_best_x = -1;
    segpts = new FPSEGPT_LIST;   //list of points
    segpt_it.set_to_list (segpts);
    min_x += pitch - pitch_error;//next limits
    max_x += pitch + pitch_error;
    while (min_box.right () < min_x && min_index < blob_count) {
      min_index++;
      min_box = box_next (&min_it);
    }
    max_it = min_it;
    max_index = min_index;
    max_box = min_box;
    next_box = box_next (&max_it);
    for (x = min_x; x <= max_x && x <= right_max; x++) {
      while (x < right_edge && max_index < blob_count
      && x > max_box.right ()) {
        max_index++;
        max_box = next_box;
        next_box = box_next (&max_it);
      }
      if (x <= max_box.left () + pitch_error
        || x >= max_box.right () - pitch_error || x >= right_edge
        || (max_index < blob_count - 1 && x >= next_box.left ())
        || (x - max_box.left () > pitch * pitsync_joined_edge
      && max_box.right () - x > pitch * pitsync_joined_edge)) {
      //                      || projection->local_min(x))
        if (x - max_box.left () > 0
          && x - max_box.left () <= pitch_error)
                                 //dist to real break
          offset = x - max_box.left ();
        else if (max_box.right () - x > 0
          && max_box.right () - x <= pitch_error
          && (max_index >= blob_count - 1
          || x < next_box.left ()))
          offset = max_box.right () - x;
        else
          offset = 0;
        //                              offset=pitsync_offset_freecut_fraction*projection->pile_count(x);
        segpt = new FPSEGPT (x, FALSE, offset, region_index,
          pitch, pitch_error, lattice_it.data ());
      }
      else {
        offset = projection->pile_count (x);
        segpt = new FPSEGPT (x, TRUE, offset, region_index,
          pitch, pitch_error, lattice_it.data ());
      }
      if (segpt->previous () != NULL) {
        segpt_it.add_after_then_move (segpt);
        if (x >= right_edge - pitch_error) {
          segpt->terminal = TRUE;//no more wanted
          if (segpt->cost_function () < best_cost) {
            best_cost = segpt->cost_function ();
            //find least
            best_end = segpt;
            best_region_index = region_index;
            left_best_x = x;
            right_best_x = x;
          }
          else if (segpt->cost_function () == best_cost
            && right_best_x == x - 1)
            right_best_x = x;
        }
      }
      else {
        delete segpt;            //no good
      }
    }
    if (segpts->empty ()) {
      if (best_end != NULL)
        break;                   //already found one
      make_illegal_segment (lattice_it.data (), min_box, min_it,
        region_index, pitch, pitch_error, segpts);
    }
    else {
      if (right_best_x > left_best_x + 1) {
        left_best_x = (left_best_x + right_best_x + 1) / 2;
        for (segpt_it.mark_cycle_pt (); !segpt_it.cycled_list ()
          && segpt_it.data ()->position () != left_best_x;
          segpt_it.forward ());
        if (segpt_it.data ()->position () == left_best_x)
                                 //middle of region
          best_end = segpt_it.data ();
      }
    }
                                 //new segment
    lattice_it.add_before_then_move (segpts);
    region_index++;
  }
  while (min_x < right_edge);
  ASSERT_HOST (best_end != NULL);//must always find some

  for (lattice_it.mark_cycle_pt (); !lattice_it.cycled_list ();
  lattice_it.forward ()) {
    segpts = lattice_it.data ();
    segpt_it.set_to_list (segpts);
    //              if (blob_count==8 && pitch==27)
    //              {
    //                      for (segpt_it.mark_cycle_pt();!segpt_it.cycled_list();segpt_it.forward())
    //                      {
    //                              segpt=segpt_it.data();
    //                              tprintf("At %d, (%x) cost=%g, m=%g, sq=%g, pred=%x\n",
    //                                      segpt->position(),segpt,segpt->cost_function(),
    //                                      segpt->sum(),segpt->squares(),segpt->previous());
    //                      }
    //                      tprintf("\n");
    //              }
    for (segpt_it.mark_cycle_pt (); !segpt_it.cycled_list ()
      && segpt_it.data () != best_end; segpt_it.forward ());
    if (segpt_it.data () == best_end) {
                                 //save good one
      segpt = segpt_it.extract ();
      outseg_it.add_before_then_move (segpt);
      best_end = segpt->previous ();
    }
  }
  ASSERT_HOST (best_end == NULL);
  ASSERT_HOST (!outseg_it.empty ());
  outseg_it.move_to_last ();
  mean_sum = outseg_it.data ()->sum ();
  mean_sum = mean_sum * mean_sum / best_region_index;
  if (outseg_it.data ()->squares () - mean_sum < 0)
    tprintf ("Impossible sqsum=%g, mean=%g, total=%d\n",
      outseg_it.data ()->squares (), outseg_it.data ()->sum (),
      best_region_index);
  lattice.deep_clear ();         //shift the lot
  return outseg_it.data ()->squares () - mean_sum;
}
ELISTIZE ( FPSEGPT  )

"Use new fast algorithm"

void make_illegal_segment ( FPSEGPT_LIST *  prev_list,
TBOX  blob_box,
BLOBNBOX_IT  blob_it,
inT16  region_index,
inT16  pitch,
inT16  pitch_error,
FPSEGPT_LIST *  seg_list 
)

Definition at line 369 of file pitsync1.cpp.

                           {
  inT16 x;                       //current coord
  inT16 min_x = 0;               //in this region
  inT16 max_x = 0;
  inT16 offset;                  //dist to edge
  FPSEGPT *segpt;                //segment point
  FPSEGPT *prevpt;               //previous point
  float best_cost;               //best path
  FPSEGPT_IT segpt_it = seg_list;//iterator
                                 //previous points
  FPSEGPT_IT prevpt_it = prev_list;

  best_cost = MAX_FLOAT32;
  for (prevpt_it.mark_cycle_pt (); !prevpt_it.cycled_list ();
  prevpt_it.forward ()) {
    prevpt = prevpt_it.data ();
    if (prevpt->cost_function () < best_cost) {
                                 //find least
      best_cost = prevpt->cost_function ();
      min_x = prevpt->position ();
      max_x = min_x;             //limits on coords
    }
    else if (prevpt->cost_function () == best_cost) {
      max_x = prevpt->position ();
    }
  }
  min_x += pitch - pitch_error;
  max_x += pitch + pitch_error;
  for (x = min_x; x <= max_x; x++) {
    while (x > blob_box.right ()) {
      blob_box = box_next (&blob_it);
    }
    offset = x - blob_box.left ();
    if (blob_box.right () - x < offset)
      offset = blob_box.right () - x;
    segpt = new FPSEGPT (x, FALSE, offset,
      region_index, pitch, pitch_error, prev_list);
    if (segpt->previous () != NULL) {
      ASSERT_HOST (offset >= 0);
      fprintf (stderr, "made fake at %d\n", x);
                                 //make one up
      segpt_it.add_after_then_move (segpt);
      segpt->faked = TRUE;
      segpt->fake_count++;
    }
    else
      delete segpt;
  }
}

Variable Documentation

EXTERN int pitsync_fake_depth = 1

"Max advance fake generation"

Definition at line 41 of file pitsync1.cpp.

EXTERN double pitsync_joined_edge = 0.75

"Dist inside big blob for chopping"

Definition at line 36 of file pitsync1.cpp.

EXTERN double pitsync_offset_freecut_fraction = 0.25

"Fraction of cut for free cuts"

Definition at line 39 of file pitsync1.cpp.