oldbasel.cpp

来自「一个google的OCR源码」· C++ 代码 · 共 1,690 行 · 第 1/5 页

 * find_lesser_parts
 *
 * Average the step from the spline for the other partitions
 * and find the commonest partition which has a descender.
 **********************************************************************/

void
find_lesser_parts (              //get descenders
TO_ROW * row,                    //row to process
TBOX blobcoords[],                //bounding boxes
int blobcount,                   /*no of blobs */
char partids[],                  /*partition of each blob */
int partsizes[],                 /*size of each part */
int partcount,                   /*no of partitions */
int bestpart                     /*biggest partition */
) {
  register int blobindex;        /*index of blob */
  register int partition;        /*current partition */
  int xcentre;                   /*centre of blob */
  int poscount;                  /*count of best up step */
  int negcount;                  /*count of best down step */
  float partsteps[MAXPARTS];     /*average step to part */
  float bestpos;                 /*best up step */
  float bestneg;                 /*best down step */
  int runlength;                 /*length of bad run */
  int biggestrun;                /*biggest bad run */

  biggestrun = 0;
  for (partition = 0; partition < partcount; partition++)
    partsteps[partition] = 0.0;  /*zero accumulators */
  for (runlength = 0, blobindex = 0; blobindex < blobcount; blobindex++) {
    xcentre = (blobcoords[blobindex].left ()
      + blobcoords[blobindex].right ()) >> 1;
                                 /*in other parts */
    if (partids[blobindex] != bestpart) {
      runlength++;               /*run of non bests */
      if (runlength > biggestrun)
        biggestrun = runlength;
      partsteps[partids[blobindex]] += blobcoords[blobindex].bottom ()
        - row->baseline.y (xcentre);
    }
    else
      runlength = 0;
  }
  if (biggestrun > MAXBADRUN)
    row->xheight = -1.0f;        /*failed */
  else
    row->xheight = 1.0f;         /*success */
  poscount = negcount = 0;
  bestpos = bestneg = 0.0;       /*no step yet */
  for (partition = 0; partition < partcount; partition++) {
    if (partition != bestpart) {

	//by jetsoft divide by zero possible
		if (partsizes[partition]==0)
		partsteps[partition]=0;
       else
		partsteps[partition] /= partsizes[partition];
	//


      if (partsteps[partition] >= MINASCRISE
      && partsizes[partition] > poscount) {
                                 /*ascender rise */
        bestpos = partsteps[partition];
                                 /*2nd most popular */
        poscount = partsizes[partition];
      }
      if (partsteps[partition] <= -MINASCRISE
      && partsizes[partition] > negcount) {
                                 /*ascender rise */
        bestneg = partsteps[partition];
                                 /*2nd most popular */
        negcount = partsizes[partition];
      }
    }
  }
                                 /*average x-height */
  partsteps[bestpart] /= blobcount;
  row->descdrop = bestneg;
}


/**********************************************************************
 * old_first_xheight
 *
 * Makes an x-height spline by copying the baseline and shifting it.
 * It estimates the x-height across the line to use as the shift.
 * It also finds the ascender height if it can.
 **********************************************************************/

void
old_first_xheight (              //the wiseowl way
TO_ROW * row,                    /*current row */
TBOX blobcoords[],                /*blob bounding boxes */
int initialheight,               //initial guess
int blobcount,                   /*blobs in blobcoords */
QSPLINE * baseline,              /*established */
float jumplimit                  /*min ascender height */
) {
  register int blobindex;        /*current blob */
                                 /*height statistics */
  STATS heightstat (0, MAXHEIGHT);
  int height;                    /*height of blob */
  int xcentre;                   /*centre of blob */
  int lineheight;                /*approx xheight */
  float ascenders;               /*ascender sum */
  int asccount;                  /*no of ascenders */
  float xsum;                    /*xheight sum */
  int xcount;                    /*xheight count */
  register float diff;           /*height difference */

  if (blobcount > 1) {
    for (blobindex = 0; blobindex < blobcount; blobindex++) {
      xcentre = (blobcoords[blobindex].left ()
        + blobcoords[blobindex].right ()) / 2;
                                 /*height of blob */
      height = (int) (blobcoords[blobindex].top () - baseline->y (xcentre) + 0.5);
      if (height > initialheight * oldbl_xhfract
        && height > textord_min_xheight)
        heightstat.add (height, 1);
    }
    if (heightstat.get_total () > 3) {
      lineheight = (int) heightstat.ile (0.25);
      if (lineheight <= 0)
        lineheight = (int) heightstat.ile (0.5);
    }
    else
      lineheight = initialheight;
  }
  else {
    lineheight = (int) (blobcoords[0].top ()
      - baseline->y ((blobcoords[0].left ()
      + blobcoords[0].right ()) / 2) +
      0.5);
  }

  xsum = 0.0f;
  xcount = 0;
  for (ascenders = 0.0f, asccount = 0, blobindex = 0; blobindex < blobcount;
  blobindex++) {
    xcentre = (blobcoords[blobindex].left ()
      + blobcoords[blobindex].right ()) / 2;
    diff = blobcoords[blobindex].top () - baseline->y (xcentre);
                                 /*is it ascender */
    if (diff > lineheight + jumplimit) {
      ascenders += diff;
      asccount++;                /*count ascenders */
    }
    else if (diff > lineheight - jumplimit) {
      xsum += diff;              /*mean xheight */
      xcount++;
    }
  }
  if (xcount > 0)
    xsum /= xcount;              /*average xheight */
  else
    xsum = (float) lineheight;   /*guess it */
  row->xheight *= xsum;
  if (asccount > 0)
    row->ascrise = ascenders / asccount - xsum;
  else
    row->ascrise = 0.0f;         /*had none */
  if (row->xheight == 0)
    row->xheight = -1.0f;
}


/**********************************************************************
 * make_first_xheight
 *
 * Makes an x-height spline by copying the baseline and shifting it.
 * It estimates the x-height across the line to use as the shift.
 * It also finds the ascender height if it can.
 **********************************************************************/

void
make_first_xheight (             //find xheight
TO_ROW * row,                    /*current row */
TBOX blobcoords[],                /*blob bounding boxes */
int lineheight,                  //initial guess
int init_lineheight,             //block level guess
int blobcount,                   /*blobs in blobcoords */
QSPLINE * baseline,              /*established */
float jumplimit                  /*min ascender height */
) {
  STATS heightstat (0, HEIGHTBUCKETS);
  int lefts[HEIGHTBUCKETS];
  int rights[HEIGHTBUCKETS];
  int modelist[MODENUM];
  int blobindex;
  int mode_count;                //blobs to count in thr
  int sign_bit;
  int mode_threshold;
  const int kBaselineTouch = 2;  // This really should change with resolution.
  const int kGoodStrength = 8;  // Strength of baseline-touching heights.
  const float kMinHeight = 0.25;  // Min fraction of lineheight to use.

  sign_bit = row->xheight > 0 ? 1 : -1;

  memset(lefts, 0, HEIGHTBUCKETS * sizeof(lefts[0]));
  memset(rights, 0, HEIGHTBUCKETS * sizeof(rights[0]));
  mode_count = 0;
  for (blobindex = 0; blobindex < blobcount; blobindex++) {
    int xcenter = (blobcoords[blobindex].left () +
        blobcoords[blobindex].right ()) / 2;
    float base = baseline->y(xcenter);
    float bottomdiff = fabs(base - blobcoords[blobindex].bottom());
    int strength = textord_ocropus_mode &&
                   bottomdiff <= kBaselineTouch ? kGoodStrength : 1;
    int height = static_cast<int>(blobcoords[blobindex].top () - base + 0.5);
    if (blobcoords[blobindex].height () > init_lineheight * kMinHeight) {
      if (height > lineheight * oldbl_xhfract
        && height > textord_min_xheight) {
        heightstat.add (height, strength);
        if (height < HEIGHTBUCKETS) {
          if (xcenter > rights[height])
            rights[height] = xcenter;
          if (xcenter > 0 && (lefts[height] == 0 || xcenter < lefts[height]))
            lefts[height] = xcenter;
        }
      }
      mode_count += strength;
    }
  }

  mode_threshold = (int) (blobcount * 0.1);
  if (oldbl_dot_error_size > 1 || oldbl_xhfix)
    mode_threshold = (int) (mode_count * 0.1);

  if (textord_oldbl_debug) {
    tprintf ("blobcount=%d, mode_count=%d, mode_t=%d\n",
      blobcount, mode_count, mode_threshold);
  }
  find_top_modes(&heightstat, HEIGHTBUCKETS, modelist, MODENUM);
  if (textord_oldbl_debug) {
    for (blobindex = 0; blobindex < MODENUM; blobindex++)
      tprintf ("mode[%d]=%d ", blobindex, modelist[blobindex]);
    tprintf ("\n");
  }
  pick_x_height(row, modelist, lefts, rights, &heightstat, mode_threshold);

  if (textord_oldbl_debug)
    tprintf ("Output xheight=%g\n", row->xheight);
  if (row->xheight < 0 && textord_oldbl_debug)
    tprintf ("warning: Row Line height < 0; %4.2f\n", row->xheight);

  if (sign_bit < 0)
    row->xheight = -row->xheight;
}

/**********************************************************************
 * find_top_modes
 *
 * Fill the input array with the indices of the top ten modes of the
 * input distribution.
 **********************************************************************/

const int kMinModeFactorOcropus = 32;
const int kMinModeFactor = 12;

void
find_top_modes (                 //get modes
STATS * stats,                   //stats to hack
int statnum,                     //no of piles
int modelist[], int modenum      //no of modes to get
) {
  int mode_count;
  int last_i = 0;
  int last_max = MAX_INT32;
  int i;
  int mode;
  int total_max = 0;
  int mode_factor = textord_ocropus_mode ?
                    kMinModeFactorOcropus : kMinModeFactor;

  for (mode_count = 0; mode_count < modenum; mode_count++) {
    mode = 0;
    for (i = 0; i < statnum; i++) {
      if (stats->pile_count (i) > stats->pile_count (mode)) {
        if ((stats->pile_count (i) < last_max) ||
        ((stats->pile_count (i) == last_max) && (i > last_i))) {
          mode = i;
        }
      }
    }
    last_i = mode;
    last_max = stats->pile_count (last_i);
    total_max += last_max;
    if (last_max <= total_max / mode_factor)
      mode = 0;
    modelist[mode_count] = mode;
  }
}


/**********************************************************************
 * pick_x_height
 *
 * Choose based on the height modes the best x height value.
 **********************************************************************/

void pick_x_height(TO_ROW * row,                    //row to do
                   int modelist[],
                   int lefts[], int rights[],
                   STATS * heightstat,
                   int mode_threshold) {
  int x;
  int y;
  int z;
  float ratio;
  int found_one_bigger = FALSE;
  int best_x_height = 0;
  int best_asc = 0;
  int num_in_best;

  for (x = 0; x < MODENUM; x++) {
    for (y = 0; y < MODENUM; y++) {
      /* Check for two modes */
      if (modelist[x] && modelist[y] &&
          heightstat->pile_count (modelist[x]) > mode_threshold &&
          (!textord_ocropus_mode ||
           MIN(rights[modelist[x]], rights[modelist[y]]) >
           MAX(lefts[modelist[x]], lefts[modelist[y]]))) {
        ratio = (float) modelist[y] / (float) modelist[x];
        if (1.2 < ratio && ratio < 1.8) {
          /* Two modes found */
          best_x_height = modelist[x];
          num_in_best = heightstat->pile_count (modelist[x]);

          /* Try to get one higher */
          do {
            found_one_bigger = FALSE;
            for (z = 0; z < MODENUM; z++) {
              if (modelist[z] == best_x_height + 1 &&
                  (!textord_ocropus_mode ||
                    MIN(rights[modelist[x]], rights[modelist[y]]) >
                    MAX(lefts[modelist[x]], lefts[modeli