/*!
* \brief pixFindBaselines()
*
* \param[in] pixs 1 bpp, 300 ppi
* \param[out] ppta [optional] pairs of pts corresponding to
* approx. ends of each text line
* \param[in] pixadb for debug output; use NULL to skip
* \return na of baseline y values, or NULL on error
*
* <pre>
* Notes:
* (1) Input binary image must have text lines already aligned
* horizontally. This can be done by either rotating the
* image with pixDeskew(), or, if a projective transform
* is required, by doing pixDeskewLocal() first.
* (2) Input null for &pta if you don't want this returned.
* The pta will come in pairs of points (left and right end
* of each baseline).
* (3) Caution: this will not work properly on text with multiple
* columns, where the lines are not aligned between columns.
* If there are multiple columns, they should be extracted
* separately before finding the baselines.
* (4) This function constructs different types of output
* for baselines; namely, a set of raster line values and
* a set of end points of each baseline.
* (5) This function was designed to handle short and long text lines
* without using dangerous thresholds on the peak heights. It does
* this by combining the differential signal with a morphological
* analysis of the locations of the text lines. One can also
* combine this data to normalize the peak heights, by weighting
* the differential signal in the region of each baseline
* by the inverse of the width of the text line found there.
* </pre>
*/
NUMA *
pixFindBaselines(PIX *pixs,
PTA **ppta,
PIXA *pixadb)
{
l_int32 h, i, j, nbox, val1, val2, ndiff, bx, by, bw, bh;
l_int32 imaxloc, peakthresh, zerothresh, inpeak;
l_int32 mintosearch, max, maxloc, nloc, locval;
l_int32 *array;
l_float32 maxval;
BOXA *boxa1, *boxa2, *boxa3;
GPLOT *gplot;
NUMA *nasum, *nadiff, *naloc, *naval;
PIX *pix1, *pix2;
PTA *pta;
PROCNAME("pixFindBaselines");
if (ppta) *ppta = NULL;
if (!pixs || pixGetDepth(pixs) != 1)
return (NUMA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
/* Close up the text characters, removing noise */
pix1 = pixMorphSequence(pixs, "c25.1 + e15.1", 0);
/* Estimate the resolution */
if (pixadb) pixaAddPix(pixadb, pixScale(pix1, 0.25, 0.25), L_INSERT);
/* Save the difference of adjacent row sums.
* The high positive-going peaks are the baselines */
if ((nasum = pixCountPixelsByRow(pix1, NULL)) == NULL) {
pixDestroy(&pix1);
return (NUMA *)ERROR_PTR("nasum not made", procName, NULL);
}
h = pixGetHeight(pixs);
nadiff = numaCreate(h);
numaGetIValue(nasum, 0, &val2);
for (i = 0; i < h - 1; i++) {
val1 = val2;
numaGetIValue(nasum, i + 1, &val2);
numaAddNumber(nadiff, val1 - val2);
}
numaDestroy(&nasum);
if (pixadb) { /* show the difference signal */
lept_mkdir("lept/baseline");
gplotSimple1(nadiff, GPLOT_PNG, "/tmp/lept/baseline/diff", "Diff Sig");
pix2 = pixRead("/tmp/lept/baseline/diff.png");
pixaAddPix(pixadb, pix2, L_INSERT);
}
/* Use the zeroes of the profile to locate each baseline. */
array = numaGetIArray(nadiff);
ndiff = numaGetCount(nadiff);
numaGetMax(nadiff, &maxval, &imaxloc);
numaDestroy(&nadiff);
/* Use this to begin locating a new peak: */
peakthresh = (l_int32)maxval / PEAK_THRESHOLD_RATIO;
/* Use this to begin a region between peaks: */
zerothresh = (l_int32)maxval / ZERO_THRESHOLD_RATIO;
naloc = numaCreate(0);
naval = numaCreate(0);
inpeak = FALSE;
for (i = 0; i < ndiff; i++) {
if (inpeak == FALSE) {
if (array[i] > peakthresh) { /* transition to in-peak */
inpeak = TRUE;
mintosearch = i + MIN_DIST_IN_PEAK; /* accept no zeros
* between i and mintosearch */
max = array[i];
maxloc = i;
}
} else { /* inpeak == TRUE; look for max */
if (array[i] > max) {
max = array[i];
maxloc = i;
mintosearch = i + MIN_DIST_IN_PEAK;
} else if (i > mintosearch && array[i] <= zerothresh) { /* leave */
inpeak = FALSE;
numaAddNumber(naval, max);
numaAddNumber(naloc, maxloc);
}
}
}
LEPT_FREE(array);
/* If array[ndiff-1] is max, eg. no descenders, baseline at bottom */
if (inpeak) {
numaAddNumber(naval, max);
numaAddNumber(naloc, maxloc);
}
if (pixadb) { /* show the raster locations for the peaks */
gplot = gplotCreate("/tmp/lept/baseline/loc", GPLOT_PNG, "Peak locs",
"rasterline", "height");
gplotAddPlot(gplot, naloc, naval, GPLOT_POINTS, "locs");
gplotMakeOutput(gplot);
gplotDestroy(&gplot);
pix2 = pixRead("/tmp/lept/baseline/loc.png");
pixaAddPix(pixadb, pix2, L_INSERT);
}
numaDestroy(&naval);
/* Generate an approximate profile of text line width.
* First, filter the boxes of text, where there may be
* more than one box for a given textline. */
pix2 = pixMorphSequence(pix1, "r11 + c20.1 + o30.1 +c1.3", 0);
if (pixadb) pixaAddPix(pixadb, pix2, L_COPY);
boxa1 = pixConnComp(pix2, NULL, 4);
pixDestroy(&pix1);
pixDestroy(&pix2);
if (boxaGetCount(boxa1) == 0) {
numaDestroy(&naloc);
boxaDestroy(&boxa1);
L_INFO("no compnents after filtering\n", procName);
return NULL;
}
boxa2 = boxaTransform(boxa1, 0, 0, 4., 4.);
boxa3 = boxaSort(boxa2, L_SORT_BY_Y, L_SORT_INCREASING, NULL);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
/* Optionally, find the baseline segments */
pta = NULL;
if (ppta) {
pta = ptaCreate(0);
*ppta = pta;
}
if (pta) {
nloc = numaGetCount(naloc);
nbox = boxaGetCount(boxa3);
for (i = 0; i < nbox; i++) {
boxaGetBoxGeometry(boxa3, i, &bx, &by, &bw, &bh);
for (j = 0; j < nloc; j++) {
numaGetIValue(naloc, j, &locval);
if (L_ABS(locval - (by + bh)) > 25)
continue;
ptaAddPt(pta, bx, locval);
ptaAddPt(pta, bx + bw, locval);
break;
}
}
}
boxaDestroy(&boxa3);
if (pixadb && pta) { /* display baselines */
l_int32 npts, x1, y1, x2, y2;
pix1 = pixConvertTo32(pixs);
npts = ptaGetCount(pta);
for (i = 0; i < npts; i += 2) {
ptaGetIPt(pta, i, &x1, &y1);
ptaGetIPt(pta, i + 1, &x2, &y2);
pixRenderLineArb(pix1, x1, y1, x2, y2, 2, 255, 0, 0);
}
pixWrite("/tmp/lept/baseline/baselines.png", pix1, IFF_PNG);
pixaAddPix(pixadb, pixScale(pix1, 0.25, 0.25), L_INSERT);
pixDestroy(&pix1);
}
return naloc;
}
/*!
* \brief pmsCreate()
*
* \param[in] minsize of data chunk that can be supplied by pms
* \param[in] smallest bytes of the smallest pre-allocated data chunk.
* \param[in] numalloc array with the number of data chunks for each
* size that are in the memory store
* \param[in] logfile use for debugging; null otherwise
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) This computes the size of the block of memory required
* and allocates it. Each chunk starts on a 32-bit word boundary.
* The chunk sizes are in powers of 2, starting at %smallest,
* and the number of levels and chunks at each level is
* specified by %numalloc.
* (2) This is intended to manage the image data for a small number
* of relatively large pix. The system malloc is expected to
* handle very large numbers of small chunks efficiently.
* (3) Important: set the allocators and call this function
* before any pix have been allocated. Destroy all the pix
* in the normal way before calling pmsDestroy().
* (4) The pms struct is stored in a static global, so this function
* is not thread-safe. When used, there must be only one thread
* per process.
* </pre>
*/
l_ok
pmsCreate(size_t minsize,
size_t smallest,
NUMA *numalloc,
const char *logfile)
{
l_int32 nlevels, i, j, nbytes;
l_int32 *alloca;
l_float32 nchunks;
l_uint32 *baseptr, *data;
l_uint32 **firstptr;
size_t *sizes;
L_PIX_MEM_STORE *pms;
L_PTRA *pa;
L_PTRAA *paa;
PROCNAME("createPMS");
if (!numalloc)
return ERROR_INT("numalloc not defined", procName, 1);
numaGetSum(numalloc, &nchunks);
if (nchunks > 1000.0)
L_WARNING("There are %.0f chunks\n", procName, nchunks);
if ((pms = (L_PIX_MEM_STORE *)LEPT_CALLOC(1, sizeof(L_PIX_MEM_STORE)))
== NULL)
return ERROR_INT("pms not made", procName, 1);
CustomPMS = pms;
/* Make sure that minsize and smallest are multiples of 32 bit words */
if (minsize % 4 != 0)
minsize -= minsize % 4;
pms->minsize = minsize;
nlevels = numaGetCount(numalloc);
pms->nlevels = nlevels;
if ((sizes = (size_t *)LEPT_CALLOC(nlevels, sizeof(size_t))) == NULL)
return ERROR_INT("sizes not made", procName, 1);
pms->sizes = sizes;
if (smallest % 4 != 0)
smallest += 4 - (smallest % 4);
pms->smallest = smallest;
for (i = 0; i < nlevels; i++)
sizes[i] = smallest * (1 << i);
pms->largest = sizes[nlevels - 1];
alloca = numaGetIArray(numalloc);
pms->allocarray = alloca;
if ((paa = ptraaCreate(nlevels)) == NULL)
return ERROR_INT("paa not made", procName, 1);
pms->paa = paa;
for (i = 0, nbytes = 0; i < nlevels; i++)
nbytes += alloca[i] * sizes[i];
pms->nbytes = nbytes;
if ((baseptr = (l_uint32 *)LEPT_CALLOC(nbytes / 4, sizeof(l_uint32)))
== NULL)
return ERROR_INT("calloc fail for baseptr", procName, 1);
pms->baseptr = baseptr;
pms->maxptr = baseptr + nbytes / 4; /* just beyond the memory store */
if ((firstptr = (l_uint32 **)LEPT_CALLOC(nlevels, sizeof(l_uint32 *)))
== NULL)
return ERROR_INT("calloc fail for firstptr", procName, 1);
pms->firstptr = firstptr;
data = baseptr;
for (i = 0; i < nlevels; i++) {
if ((pa = ptraCreate(alloca[i])) == NULL)
return ERROR_INT("pa not made", procName, 1);
ptraaInsertPtra(paa, i, pa);
firstptr[i] = data;
for (j = 0; j < alloca[i]; j++) {
ptraAdd(pa, data);
data += sizes[i] / 4;
}
}
if (logfile) {
pms->memused = (l_int32 *)LEPT_CALLOC(nlevels, sizeof(l_int32));
pms->meminuse = (l_int32 *)LEPT_CALLOC(nlevels, sizeof(l_int32));
pms->memmax = (l_int32 *)LEPT_CALLOC(nlevels, sizeof(l_int32));
pms->memempty = (l_int32 *)LEPT_CALLOC(nlevels, sizeof(l_int32));
pms->logfile = stringNew(logfile);
}
return 0;
}
/*!
* \brief recogMakeDecodingArray()
*
* \param[in] recog
* \param[in] index of averaged template
* \param[in] debug 1 for debug output; 0 otherwise
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) Generates the bit-and sum array for a character template along pixs.
* (2) The values are saved in the scoring arrays at the left edge
* of the template as it is positioned on pixs.
* </pre>
*/
static l_int32
recogMakeDecodingArray(L_RECOG *recog,
l_int32 index,
l_int32 debug)
{
l_int32 i, j, w1, h1, w2, h2, nx, ycent2, count, maxcount, maxdely;
l_int32 sum, moment, dely, shifty;
l_int32 *counta, *delya, *ycent1, *arraysum, *arraymoment, *sumtab;
NUMA *nasum, *namoment;
PIX *pix1, *pix2, *pix3;
L_RDID *did;
PROCNAME("recogMakeDecodingArray");
if (!recog)
return ERROR_INT("recog not defined", procName, 1);
if ((did = recogGetDid(recog)) == NULL)
return ERROR_INT("did not defined", procName, 1);
if (index < 0 || index >= did->narray)
return ERROR_INT("invalid index", procName, 1);
/* Check that pix1 is large enough for this template. */
pix1 = did->pixs; /* owned by did; do not destroy */
pixGetDimensions(pix1, &w1, &h1, NULL);
pix2 = pixaGetPix(recog->pixa_u, index, L_CLONE);
pixGetDimensions(pix2, &w2, &h2, NULL);
if (w1 < w2) {
L_INFO("w1 = %d < w2 = %d for index %d\n", procName, w1, w2, index);
pixDestroy(&pix2);
return 0;
}
nasum = did->nasum;
namoment = did->namoment;
ptaGetIPt(recog->pta_u, index, NULL, &ycent2);
sumtab = recog->sumtab;
counta = did->counta[index];
delya = did->delya[index];
/* Set up the array for ycent1. This gives the y-centroid location
* for a window of width w2, starting at location i. */
nx = w1 - w2 + 1; /* number of positions w2 can be placed in w1 */
ycent1 = (l_int32 *)LEPT_CALLOC(nx, sizeof(l_int32));
arraysum = numaGetIArray(nasum);
arraymoment = numaGetIArray(namoment);
for (i = 0, sum = 0, moment = 0; i < w2; i++) {
sum += arraysum[i];
moment += arraymoment[i];
}
for (i = 0; i < nx - 1; i++) {
ycent1[i] = (sum == 0) ? ycent2 : (l_float32)moment / (l_float32)sum;
sum += arraysum[w2 + i] - arraysum[i];
moment += arraymoment[w2 + i] - arraymoment[i];
}
ycent1[nx - 1] = (sum == 0) ? ycent2 : (l_float32)moment / (l_float32)sum;
/* Compute the bit-and sum between the template pix2 and pix1, at
* locations where the left side of pix2 goes from 0 to nx - 1
* in pix1. Do this around the vertical alignment of the pix2
* centroid and the windowed pix1 centroid.
* (1) Start with pix3 cleared and approximately equal in size to pix1.
* (2) Blit the y-shifted pix2 onto pix3. Then all ON pixels
* are within the intersection of pix1 and the shifted pix2.
* (3) AND pix1 with pix3. */
pix3 = pixCreate(w2, h1, 1);
for (i = 0; i < nx; i++) {
shifty = (l_int32)(ycent1[i] - ycent2 + 0.5);
maxcount = 0;
for (j = -MaxYShift; j <= MaxYShift; j++) {
pixClearAll(pix3);
dely = shifty + j; /* amount pix2 is shifted relative to pix1 */
pixRasterop(pix3, 0, dely, w2, h2, PIX_SRC, pix2, 0, 0);
pixRasterop(pix3, 0, 0, w2, h1, PIX_SRC & PIX_DST, pix1, i, 0);
pixCountPixels(pix3, &count, sumtab);
if (count > maxcount) {
maxcount = count;
maxdely = dely;
}
}
counta[i] = maxcount;
delya[i] = maxdely;
}
did->fullarrays = TRUE;
pixDestroy(&pix2);
pixDestroy(&pix3);
LEPT_FREE(ycent1);
LEPT_FREE(arraysum);
LEPT_FREE(arraymoment);
return 0;
}
/*!
* \brief recogRunViterbi()
*
* \param[in] recog with LUT's pre-computed
* \param[out] ppixdb [optional] debug result; can be null
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) This is recursive, in that
* (a) we compute the score successively at all pixel positions x,
* (b) to compute the score at x in the trellis, for each
* template we look backwards to (x - setwidth) to get the
* score if that template were to be printed with its
* setwidth location at x. We save at x the template and
* score that maximizes the sum of the score at (x - setwidth)
* and the log-likelihood for the template to be printed with
* its LHS there.
* </pre>
*/
l_int32
recogRunViterbi(L_RECOG *recog,
PIX **ppixdb)
{
l_int32 i, w1, x, narray, minsetw, first, templ, xloc, dely, counts, area1;
l_int32 besttempl, spacetempl;
l_int32 *setw, *didtempl;
l_int32 *area2; /* must be freed */
l_float32 prevscore, matchscore, maxscore, correl;
l_float32 *didscore;
PIX *pixt;
L_RDID *did;
PROCNAME("recogRunViterbi");
if (ppixdb) *ppixdb = NULL;
if (!recog)
return ERROR_INT("recog not defined", procName, 1);
if ((did = recogGetDid(recog)) == NULL)
return ERROR_INT("did not defined", procName, 1);
if (did->fullarrays == 0)
return ERROR_INT("did full arrays not made", procName, 1);
/* The score array is initialized to 0.0. As we proceed to
* the left, the log likelihood for the partial paths goes
* negative, and we prune for the max (least negative) path.
* No matches will be computed until we reach x = min(setwidth);
* until then first == TRUE after looping over templates. */
w1 = did->size;
narray = did->narray;
spacetempl = narray;
setw = did->setwidth;
minsetw = 100000;
for (i = 0; i < narray; i++) {
if (setw[i] < minsetw)
minsetw = setw[i];
}
if (minsetw <= 0) {
L_ERROR("minsetw <= 0; shouldn't happen\n", procName);
minsetw = 1;
}
didscore = did->trellisscore;
didtempl = did->trellistempl;
area2 = numaGetIArray(recog->nasum_u);
for (x = minsetw; x < w1; x++) { /* will always get a score */
first = TRUE;
for (i = 0; i < narray; i++) {
if (x - setw[i] < 0) continue;
matchscore = didscore[x - setw[i]] +
did->gamma[1] * did->counta[i][x - setw[i]] +
did->beta[1] * area2[i];
if (first) {
maxscore = matchscore;
besttempl = i;
first = FALSE;
} else {
if (matchscore > maxscore) {
maxscore = matchscore;
besttempl = i;
}
}
}
/* We can also put down a single pixel space, with no cost
* because all pixels are bg. */
prevscore = didscore[x - 1];
if (prevscore > maxscore) { /* 1 pixel space is best */
maxscore = prevscore;
besttempl = spacetempl;
}
didscore[x] = maxscore;
didtempl[x] = besttempl;
}
/* Backtrack to get the best path.
* Skip over (i.e., ignore) all single pixel spaces. */
for (x = w1 - 1; x >= 0; x--) {
if (didtempl[x] != spacetempl) break;
}
while (x > 0) {
if (didtempl[x] == spacetempl) { /* skip over spaces */
x--;
continue;
}
templ = didtempl[x];
xloc = x - setw[templ];
if (xloc < 0) break;
counts = did->counta[templ][xloc]; /* bit-and counts */
recogGetWindowedArea(recog, templ, xloc, &dely, &area1);
correl = (counts * counts) / (l_float32)(area2[templ] * area1);
pixt = pixaGetPix(recog->pixa_u, templ, L_CLONE);
numaAddNumber(did->natempl, templ);
numaAddNumber(did->naxloc, xloc);
numaAddNumber(did->nadely, dely);
numaAddNumber(did->nawidth, pixGetWidth(pixt));
numaAddNumber(did->nascore, correl);
pixDestroy(&pixt);
x = xloc;
}
if (ppixdb) {
numaWriteStream(stderr, did->natempl);
numaWriteStream(stderr, did->naxloc);
numaWriteStream(stderr, did->nadely);
numaWriteStream(stderr, did->nawidth);
numaWriteStream(stderr, did->nascore);
*ppixdb = recogShowPath(recog, 0);
}
LEPT_FREE(area2);
return 0;
}
/*!
* pixColorGrayRegionsCmap()
*
* Input: pixs (8 bpp, with colormap)
* boxa (of regions in which to apply color)
* type (L_PAINT_LIGHT, L_PAINT_DARK)
* rval, gval, bval (target color)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This is an in-place operation.
* (2) If type == L_PAINT_LIGHT, it colorizes non-black pixels,
* preserving antialiasing.
* If type == L_PAINT_DARK, it colorizes non-white pixels,
* preserving antialiasing. See pixColorGrayCmap() for details.
* (3) This can also be called through pixColorGrayRegions().
* (4) This increases the colormap size by the number of
* different gray (non-black or non-white) colors in the
* selected regions of pixs. If there is not enough room in
* the colormap for this expansion, it returns 1 (error),
* and the caller should check the return value.
* (5) Because two boxes in the boxa can overlap, pixels that
* are colorized in the first box must be excluded in the
* second because their value exceeds the size of the map.
*/
l_int32
pixColorGrayRegionsCmap(PIX *pixs,
BOXA *boxa,
l_int32 type,
l_int32 rval,
l_int32 gval,
l_int32 bval)
{
l_int32 i, j, k, w, h, n, nc, x1, y1, x2, y2, bw, bh, wpl;
l_int32 val, nval;
l_int32 *map;
l_uint32 *line, *data;
BOX *box;
NUMA *na;
PIXCMAP *cmap;
PROCNAME("pixColorGrayRegionsCmap");
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if (!boxa)
return ERROR_INT("boxa not defined", procName, 1);
if ((cmap = pixGetColormap(pixs)) == NULL)
return ERROR_INT("no colormap", procName, 1);
if (pixGetDepth(pixs) != 8)
return ERROR_INT("depth not 8 bpp", procName, 1);
if (type != L_PAINT_DARK && type != L_PAINT_LIGHT)
return ERROR_INT("invalid type", procName, 1);
nc = pixcmapGetCount(cmap);
if (addColorizedGrayToCmap(cmap, type, rval, gval, bval, &na))
return ERROR_INT("no room; cmap full", procName, 1);
map = numaGetIArray(na);
numaDestroy(&na);
if (!map)
return ERROR_INT("map not made", procName, 1);
pixGetDimensions(pixs, &w, &h, NULL);
data = pixGetData(pixs);
wpl = pixGetWpl(pixs);
n = boxaGetCount(boxa);
for (k = 0; k < n; k++) {
box = boxaGetBox(boxa, k, L_CLONE);
boxGetGeometry(box, &x1, &y1, &bw, &bh);
x2 = x1 + bw - 1;
y2 = y1 + bh - 1;
/* Remap gray pixels in the region */
for (i = y1; i <= y2; i++) {
if (i < 0 || i >= h) /* clip */
continue;
line = data + i * wpl;
for (j = x1; j <= x2; j++) {
if (j < 0 || j >= w) /* clip */
continue;
val = GET_DATA_BYTE(line, j);
if (val >= nc) continue; /* from overlapping b.b. */
nval = map[val];
if (nval != 256)
SET_DATA_BYTE(line, j, nval);
}
}
boxDestroy(&box);
}
FREE(map);
return 0;
}
/*!
* pixColorGrayMaskedCmap()
*
* Input: pixs (8 bpp, with colormap)
* pixm (1 bpp mask, through which to apply color)
* type (L_PAINT_LIGHT, L_PAINT_DARK)
* rval, gval, bval (target color)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This is an in-place operation.
* (2) If type == L_PAINT_LIGHT, it colorizes non-black pixels,
* preserving antialiasing.
* If type == L_PAINT_DARK, it colorizes non-white pixels,
* preserving antialiasing. See pixColorGrayCmap() for details.
* (3) This increases the colormap size by the number of
* different gray (non-black or non-white) colors in the
* input colormap. If there is not enough room in the colormap
* for this expansion, it returns 1 (error).
*/
l_int32
pixColorGrayMaskedCmap(PIX *pixs,
PIX *pixm,
l_int32 type,
l_int32 rval,
l_int32 gval,
l_int32 bval)
{
l_int32 i, j, w, h, wm, hm, wmin, hmin, wpl, wplm;
l_int32 val, nval;
l_int32 *map;
l_uint32 *line, *data, *linem, *datam;
NUMA *na;
PIXCMAP *cmap;
PROCNAME("pixColorGrayMaskedCmap");
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if (!pixm || pixGetDepth(pixm) != 1)
return ERROR_INT("pixm undefined or not 1 bpp", procName, 1);
if ((cmap = pixGetColormap(pixs)) == NULL)
return ERROR_INT("no colormap", procName, 1);
if (pixGetDepth(pixs) != 8)
return ERROR_INT("depth not 8 bpp", procName, 1);
if (type != L_PAINT_DARK && type != L_PAINT_LIGHT)
return ERROR_INT("invalid type", procName, 1);
if (addColorizedGrayToCmap(cmap, type, rval, gval, bval, &na))
return ERROR_INT("no room; cmap full", procName, 1);
map = numaGetIArray(na);
numaDestroy(&na);
if (!map)
return ERROR_INT("map not made", procName, 1);
pixGetDimensions(pixs, &w, &h, NULL);
pixGetDimensions(pixm, &wm, &hm, NULL);
if (wm != w)
L_WARNING("wm = %d differs from w = %d\n", procName, wm, w);
if (hm != h)
L_WARNING("hm = %d differs from h = %d\n", procName, hm, h);
wmin = L_MIN(w, wm);
hmin = L_MIN(h, hm);
data = pixGetData(pixs);
wpl = pixGetWpl(pixs);
datam = pixGetData(pixm);
wplm = pixGetWpl(pixm);
/* Remap gray pixels in the region */
for (i = 0; i < hmin; i++) {
line = data + i * wpl;
linem = datam + i * wplm;
for (j = 0; j < wmin; j++) {
if (GET_DATA_BIT(linem, j) == 0)
continue;
val = GET_DATA_BYTE(line, j);
nval = map[val];
if (nval != 256)
SET_DATA_BYTE(line, j, nval);
}
}
FREE(map);
return 0;
}
/*
* countAlignedMatches()
* Input: nai1, nai2 (numas of row pairs for matches)
* nasx, nasy (numas of x and y shifts for the matches)
* n1, n2 (number of rows in images 1 and 2)
* delx, dely (allowed difference in shifts of the match,
* compared to the reference match)
* nreq (number of required aligned matches)
* &same (<return> 1 if %nreq row matches are found; 0 otherwise)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This takes 4 input arrays giving parameters of all the
* line matches. It looks for the maximum set of aligned
* matches (matches with approximately the same overall shifts)
* that do not use rows from either image more than once.
*/
static l_int32
countAlignedMatches(NUMA *nai1,
NUMA *nai2,
NUMA *nasx,
NUMA *nasy,
l_int32 n1,
l_int32 n2,
l_int32 delx,
l_int32 dely,
l_int32 nreq,
l_int32 *psame,
l_int32 debugflag)
{
l_int32 i, j, nm, shiftx, shifty, nmatch, diffx, diffy;
l_int32 *ia1, *ia2, *iasx, *iasy, *index1, *index2;
PROCNAME("countAlignedMatches");
if (!nai1 || !nai2 || !nasx || !nasy)
return ERROR_INT("4 input numas not defined", procName, 1);
if (!psame)
return ERROR_INT("&same not defined", procName, 1);
*psame = 0;
/* Check for sufficient aligned matches, doing a double iteration
* over the set of raw matches. The row index arrays
* are used to verify that the same rows in either image
* are not used in more than one match. Whenever there
* is a match that is properly aligned, those rows are
* marked in the index arrays. */
nm = numaGetCount(nai1); /* number of matches */
if (nm < nreq)
return 0;
ia1 = numaGetIArray(nai1);
ia2 = numaGetIArray(nai2);
iasx = numaGetIArray(nasx);
iasy = numaGetIArray(nasy);
index1 = (l_int32 *)LEPT_CALLOC(n1, sizeof(l_int32)); /* keep track of rows */
index2 = (l_int32 *)LEPT_CALLOC(n2, sizeof(l_int32));
for (i = 0; i < nm; i++) {
if (*psame == 1)
break;
/* Reset row index arrays */
memset(index1, 0, 4 * n1);
memset(index2, 0, 4 * n2);
nmatch = 1;
index1[ia1[i]] = nmatch; /* mark these rows as taken */
index2[ia2[i]] = nmatch;
shiftx = iasx[i]; /* reference shift between two rows */
shifty = iasy[i]; /* ditto */
if (nreq == 1) {
*psame = 1;
break;
}
for (j = 0; j < nm; j++) {
if (j == i) continue;
/* Rows must both be different from any previously seen */
if (index1[ia1[j]] > 0 || index2[ia2[j]] > 0) continue;
/* Check the shift for this match */
diffx = L_ABS(shiftx - iasx[j]);
diffy = L_ABS(shifty - iasy[j]);
if (diffx > delx || diffy > dely) continue;
/* We have a match */
nmatch++;
index1[ia1[j]] = nmatch; /* mark the rows */
index2[ia2[j]] = nmatch;
if (nmatch >= nreq) {
*psame = 1;
if (debugflag)
printRowIndices(index1, n1, index2, n2);
break;
}
}
}
LEPT_FREE(ia1);
LEPT_FREE(ia2);
LEPT_FREE(iasx);
LEPT_FREE(iasy);
LEPT_FREE(index1);
LEPT_FREE(index2);
return 0;
}
/*!
* pixaGenerateFont()
*
* Input: pix (of 95 characters in 3 rows)
* fontsize (4, 6, 8, ... , 20, in pts at 300 ppi)
* &bl1 (<return> baseline of row 1)
* &bl2 (<return> baseline of row 2)
* &bl3 (<return> baseline of row 3)
* Return: pixa of font bitmaps for 95 characters, or null on error
*
* Notes:
* (1) This does all the work. See pixaGenerateFontFromFile()
* for an overview.
* (2) The pix is for one of the 9 fonts. @fontsize is only
* used here for debugging.
*/
PIXA *
pixaGenerateFont(PIX *pixs,
l_int32 fontsize,
l_int32 *pbl0,
l_int32 *pbl1,
l_int32 *pbl2)
{
l_int32 i, j, nrows, nrowchars, nchars, h, yval;
l_int32 width, height;
l_int32 baseline[3];
l_int32 *tab = NULL;
BOX *box, *box1, *box2;
BOXA *boxar, *boxac, *boxacs;
PIX *pix1, *pix2, *pixr, *pixrc, *pixc;
PIXA *pixa;
l_int32 n, w, inrow, top;
l_int32 *ia;
NUMA *na;
PROCNAME("pixaGenerateFont");
if (!pbl0 || !pbl1 || !pbl2)
return (PIXA *)ERROR_PTR("&bl not all defined", procName, NULL);
*pbl0 = *pbl1 = *pbl2 = 0;
if (!pixs)
return (PIXA *)ERROR_PTR("pixs not defined", procName, NULL);
/* Locate the 3 rows of characters */
w = pixGetWidth(pixs);
na = pixCountPixelsByRow(pixs, NULL);
boxar = boxaCreate(0);
n = numaGetCount(na);
ia = numaGetIArray(na);
inrow = 0;
for (i = 0; i < n; i++) {
if (!inrow && ia[i] > 0) {
inrow = 1;
top = i;
} else if (inrow && ia[i] == 0) {
inrow = 0;
box = boxCreate(0, top, w, i - top);
boxaAddBox(boxar, box, L_INSERT);
}
}
FREE(ia);
numaDestroy(&na);
nrows = boxaGetCount(boxar);
#if DEBUG_FONT_GEN
L_INFO("For fontsize %s, have %d rows\n", procName, fontsize, nrows);
#endif /* DEBUG_FONT_GEN */
if (nrows != 3) {
L_INFO("nrows = %d; skipping fontsize %d\n", procName, nrows, fontsize);
return (PIXA *)ERROR_PTR("3 rows not generated", procName, NULL);
}
/* Grab the character images and baseline data */
#if DEBUG_BASELINE
lept_rmdir("baseline");
lept_mkdir("baseline");
#endif /* DEBUG_BASELINE */
tab = makePixelSumTab8();
pixa = pixaCreate(95);
for (i = 0; i < nrows; i++) {
box = boxaGetBox(boxar, i, L_CLONE);
pixr = pixClipRectangle(pixs, box, NULL); /* row of chars */
pixGetTextBaseline(pixr, tab, &yval);
baseline[i] = yval;
#if DEBUG_BASELINE
L_INFO("Baseline info: row %d, yval = %d, h = %d\n", procName,
i, yval, pixGetHeight(pixr));
pix1 = pixCopy(NULL, pixr);
pixRenderLine(pix1, 0, yval, pixGetWidth(pix1), yval, 1,
L_FLIP_PIXELS);
if (i == 0 )
pixWrite("/tmp/baseline/row0.png", pix1, IFF_PNG);
else if (i == 1)
pixWrite("/tmp/baseline/row1.png", pix1, IFF_PNG);
else
pixWrite("/tmp/baseline/row2.png", pix1, IFF_PNG);
pixDestroy(&pix1);
#endif /* DEBUG_BASELINE */
boxDestroy(&box);
pixrc = pixCloseSafeBrick(NULL, pixr, 1, 35);
boxac = pixConnComp(pixrc, NULL, 8);
boxacs = boxaSort(boxac, L_SORT_BY_X, L_SORT_INCREASING, NULL);
if (i == 0) { /* consolidate the two components of '"' */
box1 = boxaGetBox(boxacs, 1, L_CLONE);
box2 = boxaGetBox(boxacs, 2, L_CLONE);
box1->w = box2->x + box2->w - box1->x; /* increase width */
boxDestroy(&box1);
boxDestroy(&box2);
boxaRemoveBox(boxacs, 2);
}
h = pixGetHeight(pixr);
nrowchars = boxaGetCount(boxacs);
for (j = 0; j < nrowchars; j++) {
box = boxaGetBox(boxacs, j, L_COPY);
if (box->w <= 2 && box->h == 1) { /* skip 1x1, 2x1 components */
boxDestroy(&box);
continue;
}
box->y = 0;
box->h = h - 1;
pixc = pixClipRectangle(pixr, box, NULL);
boxDestroy(&box);
if (i == 0 && j == 0) /* add a pix for the space; change later */
pixaAddPix(pixa, pixc, L_COPY);
if (i == 2 && j == 0) /* add a pix for the '\'; change later */
pixaAddPix(pixa, pixc, L_COPY);
pixaAddPix(pixa, pixc, L_INSERT);
}
pixDestroy(&pixr);
pixDestroy(&pixrc);
boxaDestroy(&boxac);
boxaDestroy(&boxacs);
}
FREE(tab);
nchars = pixaGetCount(pixa);
if (nchars != 95)
return (PIXA *)ERROR_PTR("95 chars not generated", procName, NULL);
*pbl0 = baseline[0];
*pbl1 = baseline[1];
*pbl2 = baseline[2];
/* Fix the space character up; it should have no ON pixels,
* and be about twice as wide as the '!' character. */
pix1 = pixaGetPix(pixa, 0, L_CLONE);
width = 2 * pixGetWidth(pix1);
height = pixGetHeight(pix1);
pixDestroy(&pix1);
pix1 = pixCreate(width, height, 1);
pixaReplacePix(pixa, 0, pix1, NULL);
/* Fix up the '\' character; use a LR flip of the '/' char */
pix1 = pixaGetPix(pixa, 15, L_CLONE);
pix2 = pixFlipLR(NULL, pix1);
pixDestroy(&pix1);
pixaReplacePix(pixa, 60, pix2, NULL);
#if DEBUG_CHARS
pix1 = pixaDisplayTiled(pixa, 1500, 0, 10);
pixDisplay(pix1, 100 * i, 200);
pixDestroy(&pix1);
#endif /* DEBUG_CHARS */
boxaDestroy(&boxar);
return pixa;
}
/*!
* pixColorGrayCmap()
*
* Input: pixs (2, 4 or 8 bpp, with colormap)
* box (<optional> region to set color; can be NULL)
* type (L_PAINT_LIGHT, L_PAINT_DARK)
* rval, gval, bval (target color)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This is an in-place operation.
* (2) If type == L_PAINT_LIGHT, it colorizes non-black pixels,
* preserving antialiasing.
* If type == L_PAINT_DARK, it colorizes non-white pixels,
* preserving antialiasing.
* (3) If box is NULL, applies function to the entire image; otherwise,
* clips the operation to the intersection of the box and pix.
* (4) This can also be called through pixColorGray().
* (5) This increases the colormap size by the number of
* different gray (non-black or non-white) colors in the
* input colormap. If there is not enough room in the colormap
* for this expansion, it returns 1 (error), and the caller
* should check the return value. If an error is returned
* and the cmap is only 2 or 4 bpp, the pix can be converted
* to 8 bpp and this function will succeed if run again on
* a larger colormap.
* (6) Using the darkness of each original pixel in the rect,
* it generates a new color (based on the input rgb values).
* If type == L_PAINT_LIGHT, the new color is a (generally)
* darken-to-black version of the input rgb color, where the
* amount of darkening increases with the darkness of the
* original pixel color.
* If type == L_PAINT_DARK, the new color is a (generally)
* faded-to-white version of the input rgb color, where the
* amount of fading increases with the brightness of the
* original pixel color.
*/
l_int32
pixColorGrayCmap(PIX *pixs,
BOX *box,
l_int32 type,
l_int32 rval,
l_int32 gval,
l_int32 bval)
{
l_int32 i, j, w, h, d, x1, y1, x2, y2, bw, bh, wpl;
l_int32 val, nval;
l_int32 *map;
l_uint32 *line, *data;
NUMA *na;
PIX *pixt;
PIXCMAP *cmap, *cmapc;
PROCNAME("pixColorGrayCmap");
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if ((cmap = pixGetColormap(pixs)) == NULL)
return ERROR_INT("no colormap", procName, 1);
d = pixGetDepth(pixs);
if (d != 2 && d != 4 && d != 8)
return ERROR_INT("depth not in {2, 4, 8}", procName, 1);
if (type != L_PAINT_DARK && type != L_PAINT_LIGHT)
return ERROR_INT("invalid type", procName, 1);
/* If 2 bpp or 4 bpp, see if the new colors will fit into
* the existing colormap. If not, convert in-place to 8 bpp. */
if (d == 2 || d == 4) {
cmapc = pixcmapCopy(cmap); /* experiment with a copy */
if (addColorizedGrayToCmap(cmapc, type, rval, gval, bval, NULL)) {
pixt = pixConvertTo8(pixs, 1);
pixTransferAllData(pixs, &pixt, 0, 0);
}
pixcmapDestroy(&cmapc);
}
/* Find gray colors, add the corresponding new colors,
* and set up a mapping table from gray to new.
* That table has the value 256 for all colors that are
* not to be mapped. */
cmap = pixGetColormap(pixs);
if (addColorizedGrayToCmap(cmap, type, rval, gval, bval, &na)) {
numaDestroy(&na);
return ERROR_INT("no room; cmap full", procName, 1);
}
map = numaGetIArray(na);
/* Determine the region of substitution */
pixGetDimensions(pixs, &w, &h, &d); /* d may be different */
data = pixGetData(pixs);
wpl = pixGetWpl(pixs);
if (!box) {
x1 = y1 = 0;
x2 = w;
y2 = h;
}
else {
boxGetGeometry(box, &x1, &y1, &bw, &bh);
x2 = x1 + bw - 1;
y2 = y1 + bh - 1;
}
/* Remap gray pixels in the region */
for (i = y1; i <= y2; i++) {
if (i < 0 || i >= h) /* clip */
continue;
line = data + i * wpl;
for (j = x1; j <= x2; j++) {
if (j < 0 || j >= w) /* clip */
continue;
switch (d)
{
case 2:
val = GET_DATA_DIBIT(line, j);
nval = map[val];
if (nval != 256)
SET_DATA_DIBIT(line, j, nval);
break;
case 4:
val = GET_DATA_QBIT(line, j);
nval = map[val];
if (nval != 256)
SET_DATA_QBIT(line, j, nval);
break;
case 8:
val = GET_DATA_BYTE(line, j);
nval = map[val];
if (nval != 256)
SET_DATA_BYTE(line, j, nval);
break;
}
}
}
FREE(map);
numaDestroy(&na);
return 0;
}
/*!
* pixSplitComponentWithProfile()
*
* Input: pixs (1 bpp, exactly one connected component)
* delta (distance used in extrema finding in a numa; typ. 10)
* mindel (minimum required difference between profile minimum
* and profile values +2 and -2 away; typ. 7)
* &pixdebug (<optional return> debug image of splitting)
* Return: boxa (of c.c. after splitting), or null on error
*
* Notes:
* (1) This will split the most obvious cases of touching characters.
* The split points it is searching for are narrow and deep
* minimima in the vertical pixel projection profile, after a
* large vertical closing has been applied to the component.
*/
BOXA *
pixSplitComponentWithProfile(PIX *pixs,
l_int32 delta,
l_int32 mindel,
PIX **ppixdebug)
{
l_int32 w, h, n2, i, firstmin, xmin, xshift;
l_int32 nmin, nleft, nright, nsplit, isplit, ncomp;
l_int32 *array1, *array2;
BOX *box;
BOXA *boxad;
NUMA *na1, *na2, *nasplit;
PIX *pix1, *pixdb;
PROCNAME("pixSplitComponentsWithProfile");
if (ppixdebug) *ppixdebug = NULL;
if (!pixs || pixGetDepth(pixs) != 1)
return (BOXA *)ERROR_PTR("pixa undefined or not 1 bpp", procName, NULL);
pixGetDimensions(pixs, &w, &h, NULL);
/* Closing to consolidate characters vertically */
pix1 = pixCloseSafeBrick(NULL, pixs, 1, 100);
/* Get extrema of column projections */
boxad = boxaCreate(2);
na1 = pixCountPixelsByColumn(pix1); /* w elements */
pixDestroy(&pix1);
na2 = numaFindExtrema(na1, delta);
n2 = numaGetCount(na2);
if (n2 < 3) { /* no split possible */
box = boxCreate(0, 0, w, h);
boxaAddBox(boxad, box, L_INSERT);
numaDestroy(&na1);
numaDestroy(&na2);
return boxad;
}
/* Look for sufficiently deep and narrow minima.
* All minima of of interest must be surrounded by max on each
* side. firstmin is the index of first possible minimum. */
array1 = numaGetIArray(na1);
array2 = numaGetIArray(na2);
if (ppixdebug) numaWriteStream(stderr, na2);
firstmin = (array1[array2[0]] > array1[array2[1]]) ? 1 : 2;
nasplit = numaCreate(n2); /* will hold split locations */
for (i = firstmin; i < n2 - 1; i+= 2) {
xmin = array2[i];
nmin = array1[xmin];
if (xmin + 2 >= w) break; /* no more splits possible */
nleft = array1[xmin - 2];
nright = array1[xmin + 2];
if (ppixdebug) {
fprintf(stderr,
"Splitting: xmin = %d, w = %d; nl = %d, nmin = %d, nr = %d\n",
xmin, w, nleft, nmin, nright);
}
if (nleft - nmin >= mindel && nright - nmin >= mindel) /* split */
numaAddNumber(nasplit, xmin);
}
nsplit = numaGetCount(nasplit);
#if 0
if (ppixdebug && nsplit > 0)
gplotSimple1(na1, GPLOT_X11, "/tmp/splitroot", NULL);
#endif
numaDestroy(&na1);
numaDestroy(&na2);
FREE(array1);
FREE(array2);
if (nsplit == 0) { /* no splitting */
box = boxCreate(0, 0, w, h);
boxaAddBox(boxad, box, L_INSERT);
return boxad;
}
/* Use split points to generate b.b. after splitting */
for (i = 0, xshift = 0; i < nsplit; i++) {
numaGetIValue(nasplit, i, &isplit);
box = boxCreate(xshift, 0, isplit - xshift, h);
boxaAddBox(boxad, box, L_INSERT);
xshift = isplit + 1;
}
box = boxCreate(xshift, 0, w - xshift, h);
boxaAddBox(boxad, box, L_INSERT);
numaDestroy(&nasplit);
if (ppixdebug) {
pixdb = pixConvertTo32(pixs);
ncomp = boxaGetCount(boxad);
for (i = 0; i < ncomp; i++) {
box = boxaGetBox(boxad, i, L_CLONE);
pixRenderBoxBlend(pixdb, box, 1, 255, 0, 0, 0.5);
boxDestroy(&box);
}
*ppixdebug = pixdb;
}
return boxad;
}
/*!
* wshedApply()
*
* Input: wshed (generated from wshedCreate())
* Return: 0 if OK, 1 on error
*
* Iportant note:
* (1) This is buggy. It seems to locate watersheds that are
* duplicates. The watershed extraction after complete fill
* grabs some regions belonging to existing watersheds.
* See prog/watershedtest.c for testing.
*/
l_int32
wshedApply(L_WSHED *wshed) {
char two_new_watersheds[] = "Two new watersheds";
char seed_absorbed_into_seeded_basin[] = "Seed absorbed into seeded basin";
char one_new_watershed_label[] = "One new watershed (label)";
char one_new_watershed_index[] = "One new watershed (index)";
char minima_absorbed_into_seeded_basin[] =
"Minima absorbed into seeded basin";
char minima_absorbed_by_filler_or_another[] =
"Minima absorbed by filler or another";
l_int32 nseeds, nother, nboth, arraysize;
l_int32 i, j, val, x, y, w, h, index, mindepth;
l_int32 imin, imax, jmin, jmax, cindex, clabel, nindex;
l_int32 hindex, hlabel, hmin, hmax, minhindex, maxhindex;
l_int32 *lut;
l_uint32 ulabel, uval;
void **lines8, **linelab32;
NUMA *nalut, *nalevels, *nash, *namh, *nasi;
NUMA **links;
L_HEAP *lh;
PIX *pixmin, *pixsd;
PIXA *pixad;
L_STACK *rstack;
PTA *ptas, *ptao;
PROCNAME("wshedApply");
if (!wshed)
return ERROR_INT("wshed not defined", procName, 1);
/* ------------------------------------------------------------ *
* Initialize priority queue and pixlab with seeds and minima *
* ------------------------------------------------------------ */
lh = lheapCreate(0, L_SORT_INCREASING); /* remove lowest values first */
rstack = lstackCreate(0); /* for reusing the WSPixels */
pixGetDimensions(wshed->pixs, &w, &h, NULL);
lines8 = wshed->lines8; /* wshed owns this */
linelab32 = wshed->linelab32; /* ditto */
/* Identify seed (marker) pixels, 1 for each c.c. in pixm */
pixSelectMinInConnComp(wshed->pixs, wshed->pixm, &ptas, &nash);
pixsd = pixGenerateFromPta(ptas, w, h);
nseeds = ptaGetCount(ptas);
for (i = 0; i < nseeds; i++) {
ptaGetIPt(ptas, i, &x, &y);
uval = GET_DATA_BYTE(lines8[y], x);
pushWSPixel(lh, rstack, (l_int32) uval, x, y, i);
}
wshed->ptas = ptas;
nasi = numaMakeConstant(1, nseeds); /* indicator array */
wshed->nasi = nasi;
wshed->nash = nash;
wshed->nseeds = nseeds;
/* Identify minima that are not seeds. Use these 4 steps:
* (1) Get the local minima, which can have components
* of arbitrary size. This will be a clipping mask.
* (2) Get the image of the actual seeds (pixsd)
* (3) Remove all elements of the clipping mask that have a seed.
* (4) Shrink each of the remaining elements of the minima mask
* to a single pixel. */
pixLocalExtrema(wshed->pixs, 200, 0, &pixmin, NULL);
pixRemoveSeededComponents(pixmin, pixsd, pixmin, 8, 2);
pixSelectMinInConnComp(wshed->pixs, pixmin, &ptao, &namh);
nother = ptaGetCount(ptao);
for (i = 0; i < nother; i++) {
ptaGetIPt(ptao, i, &x, &y);
uval = GET_DATA_BYTE(lines8[y], x);
pushWSPixel(lh, rstack, (l_int32) uval, x, y, nseeds + i);
}
wshed->namh = namh;
/* ------------------------------------------------------------ *
* Initialize merging lookup tables *
* ------------------------------------------------------------ */
/* nalut should always give the current after-merging index.
* links are effectively backpointers: they are numas associated with
* a dest index of all indices in nalut that point to that index. */
mindepth = wshed->mindepth;
nboth = nseeds + nother;
arraysize = 2 * nboth;
wshed->arraysize = arraysize;
nalut = numaMakeSequence(0, 1, arraysize);
lut = numaGetIArray(nalut);
wshed->lut = lut; /* wshed owns this */
links = (NUMA **) CALLOC(arraysize, sizeof(NUMA * ));
wshed->links = links; /* wshed owns this */
nindex = nseeds + nother; /* the next unused index value */
/* ------------------------------------------------------------ *
* Fill the basins, using the priority queue *
* ------------------------------------------------------------ */
pixad = pixaCreate(nseeds);
wshed->pixad = pixad; /* wshed owns this */
nalevels = numaCreate(nseeds);
wshed->nalevels = nalevels; /* wshed owns this */
L_INFO("nseeds = %d, nother = %d\n", procName, nseeds, nother);
while (lheapGetCount(lh) > 0) {
popWSPixel(lh, rstack, &val, &x, &y, &index);
/* fprintf(stderr, "x = %d, y = %d, index = %d\n", x, y, index); */
ulabel = GET_DATA_FOUR_BYTES(linelab32[y], x);
if (ulabel == MAX_LABEL_VALUE)
clabel = ulabel;
else
clabel = lut[ulabel];
cindex = lut[index];
if (clabel == cindex) continue; /* have already seen this one */
if (clabel == MAX_LABEL_VALUE) { /* new one; assign index and try to
* propagate to all neighbors */
SET_DATA_FOUR_BYTES(linelab32[y], x, cindex);
imin = L_MAX(0, y - 1);
imax = L_MIN(h - 1, y + 1);
jmin = L_MAX(0, x - 1);
jmax = L_MIN(w - 1, x + 1);
for (i = imin; i <= imax; i++) {
for (j = jmin; j <= jmax; j++) {
if (i == y && j == x) continue;
uval = GET_DATA_BYTE(lines8[i], j);
pushWSPixel(lh, rstack, (l_int32) uval, j, i, cindex);
}
}
} else { /* pixel is already labeled (differently); must resolve */
/* If both indices are seeds, check if the min height is
* greater than mindepth. If so, we have two new watersheds;
* locate them and assign to both regions a new index
* for further waterfill. If not, absorb the shallower
* watershed into the deeper one and continue filling it. */
pixGetPixel(pixsd, x, y, &uval);
if (clabel < nseeds && cindex < nseeds) {
wshedGetHeight(wshed, val, clabel, &hlabel);
wshedGetHeight(wshed, val, cindex, &hindex);
hmin = L_MIN(hlabel, hindex);
hmax = L_MAX(hlabel, hindex);
if (hmin == hmax) {
hmin = hlabel;
hmax = hindex;
}
if (wshed->debug) {
fprintf(stderr, "clabel,hlabel = %d,%d\n", clabel, hlabel);
fprintf(stderr, "hmin = %d, hmax = %d\n", hmin, hmax);
fprintf(stderr, "cindex,hindex = %d,%d\n", cindex, hindex);
if (hmin < mindepth)
fprintf(stderr, "Too shallow!\n");
}
if (hmin >= mindepth) {
debugWshedMerge(wshed, two_new_watersheds,
x, y, clabel, cindex);
wshedSaveBasin(wshed, cindex, val - 1);
wshedSaveBasin(wshed, clabel, val - 1);
numaSetValue(nasi, cindex, 0);
numaSetValue(nasi, clabel, 0);
if (wshed->debug) fprintf(stderr, "nindex = %d\n", nindex);
debugPrintLUT(lut, nindex, wshed->debug);
mergeLookup(wshed, clabel, nindex);
debugPrintLUT(lut, nindex, wshed->debug);
mergeLookup(wshed, cindex, nindex);
debugPrintLUT(lut, nindex, wshed->debug);
nindex++;
} else /* extraneous seed within seeded basin; absorb */ {
debugWshedMerge(wshed, seed_absorbed_into_seeded_basin,
x, y, clabel, cindex);
}
maxhindex = clabel; /* TODO: is this part of above 'else'? */
minhindex = cindex;
if (hindex > hlabel) {
maxhindex = cindex;
minhindex = clabel;
}
mergeLookup(wshed, minhindex, maxhindex);
} else if (clabel < nseeds && cindex >= nboth) {
/* If one index is a seed and the other is a merge of
* 2 watersheds, generate a single watershed. */
debugWshedMerge(wshed, one_new_watershed_label,
x, y, clabel, cindex);
wshedSaveBasin(wshed, clabel, val - 1);
numaSetValue(nasi, clabel, 0);
mergeLookup(wshed, clabel, cindex);
} else if (cindex < nseeds && clabel >= nboth) {
debugWshedMerge(wshed, one_new_watershed_index,
x, y, clabel, cindex);
wshedSaveBasin(wshed, cindex, val - 1);
numaSetValue(nasi, cindex, 0);
mergeLookup(wshed, cindex, clabel);
} else if (clabel < nseeds) { /* cindex from minima; absorb */
/* If one index is a seed and the other is from a minimum,
* merge the minimum wshed into the seed wshed. */
debugWshedMerge(wshed, minima_absorbed_into_seeded_basin,
x, y, clabel, cindex);
mergeLookup(wshed, cindex, clabel);
} else if (cindex < nseeds) { /* clabel from minima; absorb */
debugWshedMerge(wshed, minima_absorbed_into_seeded_basin,
x, y, clabel, cindex);
mergeLookup(wshed, clabel, cindex);
} else { /* If neither index is a seed, just merge */
debugWshedMerge(wshed, minima_absorbed_by_filler_or_another,
x, y, clabel, cindex);
mergeLookup(wshed, clabel, cindex);
}
}
}
#if 0
/* Use the indicator array to save any watersheds that fill
* to the maximum value. This seems to screw things up! */
for (i = 0; i < nseeds; i++) {
numaGetIValue(nasi, i, &ival);
if (ival == 1) {
wshedSaveBasin(wshed, lut[i], val - 1);
numaSetValue(nasi, i, 0);
}
}
#endif
numaDestroy(&nalut);
pixDestroy(&pixmin);
pixDestroy(&pixsd);
ptaDestroy(&ptao);
lheapDestroy(&lh, TRUE);
lstackDestroy(&rstack, TRUE);
return 0;
}
