oldbasel.cpp

一ＯＣＲ的相关资料。．希望对研究ＯＣＲ的朋友有所帮助．

  - MAXOVERLAP * (rightedge - leftedge)) {
    *baseline = *spline;         /*copy it */
    x = (leftedge + rightedge) / 2.0;
    shift = ICOORD (0, (INT16) (gradient * x + c - spline->y (x)));
    baseline->move (shift);
  }
}


/**********************************************************************
 * partition_line
 *
 * Partition a row of blobs into different groups of continuous
 * y position. jumplimit specifies the max allowable limit on a jump
 * before a new partition is started.
 * The return value is the biggest partition
 **********************************************************************/

int
partition_line (                 //partition blobs
BOX blobcoords[],                //bounding boxes
int blobcount,                   /*no of blobs on row */
int *numparts,                   /*number of partitions */
char partids[],                  /*partition no of each blob */
int partsizes[],                 /*no in each partition */
QSPLINE * spline,                /*curve to fit to */
float jumplimit,                 /*allowed delta change */
float ydiffs[]                   /*diff from spline */
) {
  register int blobindex;        /*no along text line */
  int bestpart;                  /*best new partition */
  int biggestpart;               /*part with most members */
  float diff;                    /*difference from line */
  int startx;                    /*index of start blob */
  float partdiffs[MAXPARTS];     /*step between parts */

  for (bestpart = 0; bestpart < MAXPARTS; bestpart++)
    partsizes[bestpart] = 0;     /*zero them all */

  startx = get_ydiffs (blobcoords, blobcount, spline, ydiffs);
  *numparts = 1;                 /*1 partition */
  bestpart = -1;                 /*first point */
  for (blobindex = startx; blobindex < blobcount; blobindex++) {
  /*do each blob in row */
    diff = ydiffs[blobindex];    /*diff from line */
    if (textord_oldbl_debug) {
      tprintf ("%d(%d,%d), ", blobindex,
        blobcoords[blobindex].left (),
        blobcoords[blobindex].bottom ());
    }
    bestpart =
      choose_partition(diff, partdiffs, bestpart, jumplimit, numparts);
                                 /*record partition */
    partids[blobindex] = bestpart;
    partsizes[bestpart]++;       /*another in it */
  }

  bestpart = -1;                 /*first point */
  partsizes[0]--;                /*doing 1st pt again */
                                 /*do each blob in row */
  for (blobindex = startx; blobindex >= 0; blobindex--) {
    diff = ydiffs[blobindex];    /*diff from line */
    if (textord_oldbl_debug) {
      tprintf ("%d(%d,%d), ", blobindex,
        blobcoords[blobindex].left (),
        blobcoords[blobindex].bottom ());
    }
    bestpart =
      choose_partition(diff, partdiffs, bestpart, jumplimit, numparts);
                                 /*record partition */
    partids[blobindex] = bestpart;
    partsizes[bestpart]++;       /*another in it */
  }

  for (biggestpart = 0, bestpart = 1; bestpart < *numparts; bestpart++)
    if (partsizes[bestpart] >= partsizes[biggestpart])
      biggestpart = bestpart;    /*new biggest */
  if (textord_oldbl_merge_parts)
    merge_oldbl_parts(blobcoords,
                      blobcount,
                      partids,
                      partsizes,
                      biggestpart,
                      jumplimit);
  return biggestpart;            /*biggest partition */
}


/**********************************************************************
 * merge_oldbl_parts
 *
 * For any adjacent group of blobs in a different part, put them in the
 * main part if they fit closely to neighbours in the main part.
 **********************************************************************/

void
merge_oldbl_parts (              //partition blobs
BOX blobcoords[],                //bounding boxes
int blobcount,                   /*no of blobs on row */
char partids[],                  /*partition no of each blob */
int partsizes[],                 /*no in each partition */
int biggestpart,                 //major partition
float jumplimit                  /*allowed delta change */
) {
  BOOL8 found_one;               //found a bestpart blob
  BOOL8 close_one;               //found was close enough
  register int blobindex;        /*no along text line */
  int prevpart;                  //previous iteration
  int runlength;                 //no in this part
  float diff;                    /*difference from line */
  int startx;                    /*index of start blob */
  int test_blob;                 //another index
  FCOORD coord;                  //blob coordinate
  float m, c;                    //fitted line
  QLSQ stats;                    //line stuff

  prevpart = biggestpart;
  runlength = 0;
  startx = 0;
  for (blobindex = 0; blobindex < blobcount; blobindex++) {
    if (partids[blobindex] != prevpart) {
      //                      tprintf("Partition change at (%d,%d) from %d to %d after run of %d\n",
      //                              blobcoords[blobindex].left(),blobcoords[blobindex].bottom(),
      //                              prevpart,partids[blobindex],runlength);
      if (prevpart != biggestpart && runlength > MAXBADRUN) {
        stats.clear ();
        for (test_blob = startx; test_blob < blobindex; test_blob++) {
          coord = FCOORD ((blobcoords[test_blob].left ()
            + blobcoords[test_blob].right ()) / 2.0,
            blobcoords[test_blob].bottom ());
          stats.add (coord.x (), coord.y ());
        }
        stats.fit (1);
        m = stats.get_b ();
        c = stats.get_c ();
        if (textord_oldbl_debug)
          tprintf ("Fitted line y=%g x + %g\n", m, c);
        found_one = FALSE;
        close_one = FALSE;
        for (test_blob = 1; !found_one
          && (startx - test_blob >= 0
        || blobindex + test_blob <= blobcount); test_blob++) {
          if (startx - test_blob >= 0
          && partids[startx - test_blob] == biggestpart) {
            found_one = TRUE;
            coord = FCOORD ((blobcoords[startx - test_blob].left ()
              + blobcoords[startx -
              test_blob].right ()) /
              2.0,
              blobcoords[startx -
              test_blob].bottom ());
            diff = m * coord.x () + c - coord.y ();
            if (textord_oldbl_debug)
              tprintf
                ("Diff of common blob to suspect part=%g at (%g,%g)\n",
                diff, coord.x (), coord.y ());
            if (diff < jumplimit && -diff < jumplimit)
              close_one = TRUE;
          }
          if (blobindex + test_blob <= blobcount
          && partids[blobindex + test_blob - 1] == biggestpart) {
            found_one = TRUE;
            coord =
              FCOORD ((blobcoords[blobindex + test_blob - 1].
              left () + blobcoords[blobindex + test_blob -
              1].right ()) / 2.0,
              blobcoords[blobindex + test_blob -
              1].bottom ());
            diff = m * coord.x () + c - coord.y ();
            if (textord_oldbl_debug)
              tprintf
                ("Diff of common blob to suspect part=%g at (%g,%g)\n",
                diff, coord.x (), coord.y ());
            if (diff < jumplimit && -diff < jumplimit)
              close_one = TRUE;
          }
        }
        if (close_one) {
          if (textord_oldbl_debug)
            tprintf
              ("Merged %d blobs back into part %d from %d starting at (%d,%d)\n",
              runlength, biggestpart, prevpart,
              blobcoords[startx].left (),
              blobcoords[startx].bottom ());
                                 //switch sides
          partsizes[prevpart] -= runlength;
          for (test_blob = startx; test_blob < blobindex; test_blob++)
            partids[test_blob] = biggestpart;
        }
      }
      prevpart = partids[blobindex];
      runlength = 1;
      startx = blobindex;
    }
    else
      runlength++;
  }
}


/**********************************************************************
 * get_ydiffs
 *
 * Get the differences between the blobs and the spline,
 * putting them in ydiffs.  The return value is the index
 * of the blob in the middle of the "best behaved" region
 **********************************************************************/

int
get_ydiffs (                     //evaluate differences
BOX blobcoords[],                //bounding boxes
int blobcount,                   /*no of blobs */
QSPLINE * spline,                /*approximating spline */
float ydiffs[]                   /*output */
) {
  register int blobindex;        /*current blob */
  int xcentre;                   /*xcoord */
  int lastx;                     /*last xcentre */
  float diffsum;                 /*sum of diffs */
  float diff;                    /*current difference */
  float drift;                   /*sum of spline steps */
  float bestsum;                 /*smallest diffsum */
  int bestindex;                 /*index of bestsum */

  diffsum = 0.0f;
  bestindex = 0;
  bestsum = (float) MAX_INT32;
  drift = 0.0f;
  lastx = blobcoords[0].left ();
                                 /*do each blob in row */
  for (blobindex = 0; blobindex < blobcount; blobindex++) {
                                 /*centre of blob */
    xcentre = (blobcoords[blobindex].left () + blobcoords[blobindex].right ()) >> 1;
                                 //step functions in spline
    drift += spline->step (lastx, xcentre);
    lastx = xcentre;
    diff = blobcoords[blobindex].bottom ();
    diff -= spline->y (xcentre);
    diff += drift;
    ydiffs[blobindex] = diff;    /*store difference */
    if (blobindex > 2)
                                 /*remove old one */
      diffsum -= ABS (ydiffs[blobindex - 3]);
    diffsum += ABS (diff);       /*add new one */
    if (blobindex >= 2 && diffsum < bestsum) {
      bestsum = diffsum;         /*find min sum */
      bestindex = blobindex - 1; /*middle of set */
    }
  }
  return bestindex;
}


/**********************************************************************
 * choose_partition
 *
 * Choose a partition for the point and return the index.
 **********************************************************************/

int
choose_partition (               //select partition
register float diff,             /*diff from spline */
float partdiffs[],               /*diff on all parts */
int lastpart,                    /*last assigned partition */
float jumplimit,                 /*new part threshold */
int *partcount                   /*no of partitions */
) {
  register int partition;        /*partition no */
  int bestpart;                  /*best new partition */
  float bestdelta;               /*best gap from a part */
  static float drift;            /*drift from spline */
  float delta;                   /*diff from part */
  static float lastdelta;        /*previous delta */

  if (lastpart < 0) {
    partdiffs[0] = diff;
    lastpart = 0;                /*first point */
    drift = 0.0f;
    lastdelta = 0.0f;
  }
                                 /*adjusted diff from part */
  delta = diff - partdiffs[lastpart] - drift;
  if (textord_oldbl_debug) {
    tprintf ("Diff=%.2f, Delta=%.3f, Drift=%.3f, ", diff, delta, drift);
  }
  if (ABS (delta) > jumplimit / 2) {
                                 /*delta on part 0 */
    bestdelta = diff - partdiffs[0] - drift;
    bestpart = 0;                /*0 best so far */
    for (partition = 1; partition < *partcount; partition++) {
      delta = diff - partdiffs[partition] - drift;
      if (ABS (delta) < ABS (bestdelta)) {
        bestdelta = delta;
        bestpart = partition;    /*part with nearest jump */
      }
    }
    delta = bestdelta;
                                 /*too far away */
    if (ABS (bestdelta) > jumplimit
    && *partcount < MAXPARTS) {  /*and spare part left */
      bestpart = (*partcount)++; /*best was new one */
                                 /*start new one */
      partdiffs[bestpart] = diff - drift;
      delta = 0.0f;
    }
  }
  else {
    bestpart = lastpart;         /*best was last one */
  }

  if (bestpart == lastpart
    && (ABS (delta - lastdelta) < jumplimit / 2
    || ABS (delta) < jumplimit / 2))
                                 /*smooth the drift */
    drift = (3 * drift + delta) / 3;
  lastdelta = delta;

  if (textord_oldbl_debug) {
    tprintf ("P=%d\n", bestpart);
  }

  return bestpart;
}


///*merge_partitions(partids,partcount,blobcount,bestpart) discards funny looking
//partitions and gives all the rest partid 0*/
//
//merge_partitions(partids,partcount,blobcount,bestpart)
//register char              *partids;                     /*partition numbers*/
//int                        partcount;                    /*no of partitions*/
//int                        blobcount;                    /*no of blobs*/
//int                        bestpart;                     /*best partition*/
//{
//   register int            blobindex;                    /*no along text line*/
//   int                     runlength;                    /*run of same partition*/
//   int                     bestrun;                      /*biggest runlength*/
//
//   bestrun=0;                                            /*no runs yet*/
//   runlength=1;
//   for (blobindex=1;blobindex<blobcount;blobindex++)
//   {  if (partids[blobindex]!=partids[blobindex-1])