int main(int argc,
char **argv)
{
char *dirin, *dirout, *infile, *outfile, *tail;
l_int32 i, nfiles, border, x, y, w, h, xb, yb, wb, hb;
BOX *box1, *box2;
BOXA *boxa1, *boxa2;
PIX *pixs, *pixt1, *pixd;
SARRAY *safiles;
static char mainName[] = "croptext";
if (argc != 4)
return ERROR_INT("Syntax: croptext dirin border dirout", mainName, 1);
dirin = argv[1];
border = atoi(argv[2]);
dirout = argv[3];
setLeptDebugOK(1);
safiles = getSortedPathnamesInDirectory(dirin, NULL, 0, 0);
nfiles = sarrayGetCount(safiles);
for (i = 0; i < nfiles; i++) {
infile = sarrayGetString(safiles, i, L_NOCOPY);
splitPathAtDirectory(infile, NULL, &tail);
outfile = genPathname(dirout, tail);
pixs = pixRead(infile);
pixt1 = pixMorphSequence(pixs, "r11 + c10.40 + o5.5 + x4", 0);
boxa1 = pixConnComp(pixt1, NULL, 8);
if (boxaGetCount(boxa1) == 0) {
fprintf(stderr, "Warning: no components on page %s\n", tail);
continue;
}
boxa2 = boxaSort(boxa1, L_SORT_BY_AREA, L_SORT_DECREASING, NULL);
box1 = boxaGetBox(boxa2, 0, L_CLONE);
boxGetGeometry(box1, &x, &y, &w, &h);
xb = L_MAX(0, x - border);
yb = L_MAX(0, y - border);
wb = w + 2 * border;
hb = h + 2 * border;
box2 = boxCreate(xb, yb, wb, hb);
pixd = pixClipRectangle(pixs, box2, NULL);
pixWrite(outfile, pixd, IFF_TIFF_G4);
pixDestroy(&pixs);
pixDestroy(&pixt1);
pixDestroy(&pixd);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
}
return 0;
}
/*!
* \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;
}
main(int argc,
char **argv)
{
char *filein;
l_int32 i, n, ns;
BOX *box;
BOXA *boxa, *boxas;
PIX *pixs, *pixt;
PIXA *pixa, *pixas, *pixas2;
static char mainName[] = "sorttest";
if (argc != 2)
exit(ERROR_INT(" Syntax: sorttest filein", mainName, 1));
filein = argv[1];
if ((pixs = pixRead(filein)) == NULL)
exit(ERROR_INT("pixs not made", mainName, 1));
#if 0
boxa = pixConnComp(pixs, NULL, 8);
n = boxaGetCount(boxa);
boxas = boxaSort(boxa, L_SORT_BY_PERIMETER, L_SORT_DECREASING, NULL);
ns = boxaGetCount(boxas);
fprintf(stderr, "Number of cc: n = %d, ns = %d\n", n, ns);
boxaWrite("/tmp/junkboxa.ba", boxas);
for (i = 0; i < n; i++) {
box = boxaGetBox(boxas, i, L_CLONE);
pixRenderBox(pixs, box, 2, L_FLIP_PIXELS);
boxDestroy(&box);
}
pixWrite("/tmp/junkout.png", pixs, IFF_PNG);
boxaDestroy(&boxa);
boxaDestroy(&boxas);
#endif
#if 1
boxa = pixConnComp(pixs, &pixa, 8);
n = pixaGetCount(pixa);
pixas = pixaSort(pixa, L_SORT_BY_Y, L_SORT_INCREASING, NULL, L_CLONE);
ns = pixaGetCount(pixas);
fprintf(stderr, "Number of cc: n = %d, ns = %d\n", n, ns);
pixaWrite("/tmp/junkpixa.pa", pixas);
pixas2 = pixaRead("/tmp/junkpixa.pa");
pixaWrite("/tmp/junkpixa2.pa", pixas2);
pixt = pixaDisplayOnLattice(pixas, 100, 100);
pixWrite("/tmp/junkpix.png", pixt, IFF_PNG);
boxaWrite("/tmp/junkboxa.ba", pixas->boxa);
pixDestroy(&pixt);
pixaDestroy(&pixa);
pixaDestroy(&pixas);
pixaDestroy(&pixas2);
boxaDestroy(&boxa);
#endif
pixDestroy(&pixs);
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;
}
/*!
* pixFindPageForeground()
*
* Input: pixs (full resolution (any type or depth)
* threshold (for binarization; typically about 128)
* mindist (min distance of text from border to allow
* cleaning near border; at 2x reduction, this
* should be larger than 50; typically about 70)
* erasedist (when conditions are satisfied, erase anything
* within this distance of the edge;
* typically 30 at 2x reduction)
* pagenum (use for debugging when called repeatedly; labels
* debug images that are assembled into pdfdir)
* showmorph (set to a negative integer to show steps in
* generating masks; this is typically used
* for debugging region extraction)
* display (set to 1 to display mask and selected region
* for debugging a single page)
* pdfdir (subdirectory of /tmp where images showing the
* result are placed when called repeatedly; use
* null if no output requested)
* Return: box (region including foreground, with some pixel noise
* removed), or null if not found
*
* Notes:
* (1) This doesn't simply crop to the fg. It attempts to remove
* pixel noise and junk at the edge of the image before cropping.
* The input @threshold is used if pixs is not 1 bpp.
* (2) There are several debugging options, determined by the
* last 4 arguments.
* (3) If you want pdf output of results when called repeatedly,
* the pagenum arg labels the images written, which go into
* /tmp/<pdfdir>/<pagenum>.png. In that case,
* you would clean out the /tmp directory before calling this
* function on each page:
* lept_rmdir(pdfdir);
* lept_mkdir(pdfdir);
*/
BOX *
pixFindPageForeground(PIX *pixs,
l_int32 threshold,
l_int32 mindist,
l_int32 erasedist,
l_int32 pagenum,
l_int32 showmorph,
l_int32 display,
const char *pdfdir)
{
char buf[64];
l_int32 flag, nbox, intersects;
l_int32 w, h, bx, by, bw, bh, left, right, top, bottom;
PIX *pixb, *pixb2, *pixseed, *pixsf, *pixm, *pix1, *pixg2;
BOX *box, *boxfg, *boxin, *boxd;
BOXA *ba1, *ba2;
PROCNAME("pixFindPageForeground");
if (!pixs)
return (BOX *)ERROR_PTR("pixs not defined", procName, NULL);
/* Binarize, downscale by 0.5, remove the noise to generate a seed,
* and do a seedfill back from the seed into those 8-connected
* components of the binarized image for which there was at least
* one seed pixel. Also clear out any components that are within
* 10 pixels of the edge at 2x reduction. */
flag = (showmorph) ? -1 : 0; /* if showmorph == -1, write intermediate
* images to /tmp/seq_output_1.pdf */
pixb = pixConvertTo1(pixs, threshold);
pixb2 = pixScale(pixb, 0.5, 0.5);
pixseed = pixMorphSequence(pixb2, "o1.2 + c9.9 + o3.5", flag);
pixsf = pixSeedfillBinary(NULL, pixseed, pixb2, 8);
pixSetOrClearBorder(pixsf, 10, 10, 10, 10, PIX_SET);
pixm = pixRemoveBorderConnComps(pixsf, 8);
if (display) pixDisplay(pixm, 100, 100);
/* Now, where is the main block of text? We want to remove noise near
* the edge of the image, but to do that, we have to be convinced that
* (1) there is noise and (2) it is far enough from the text block
* and close enough to the edge. For each edge, if the block
* is more than mindist from that edge, then clean 'erasedist'
* pixels from the edge. */
pix1 = pixMorphSequence(pixm, "c50.50", flag - 1);
ba1 = pixConnComp(pix1, NULL, 8);
ba2 = boxaSort(ba1, L_SORT_BY_AREA, L_SORT_DECREASING, NULL);
pixGetDimensions(pix1, &w, &h, NULL);
nbox = boxaGetCount(ba2);
if (nbox > 1) {
box = boxaGetBox(ba2, 0, L_CLONE);
boxGetGeometry(box, &bx, &by, &bw, &bh);
left = (bx > mindist) ? erasedist : 0;
right = (w - bx - bw > mindist) ? erasedist : 0;
top = (by > mindist) ? erasedist : 0;
bottom = (h - by - bh > mindist) ? erasedist : 0;
pixSetOrClearBorder(pixm, left, right, top, bottom, PIX_CLR);
boxDestroy(&box);
}
pixDestroy(&pix1);
boxaDestroy(&ba1);
boxaDestroy(&ba2);
/* Locate the foreground region; don't bother cropping */
pixClipToForeground(pixm, NULL, &boxfg);
/* Sanity check the fg region. Make sure it's not confined
* to a thin boundary on the left and right sides of the image,
* in which case it is likely to be noise. */
if (boxfg) {
boxin = boxCreate(0.1 * w, 0, 0.8 * w, h);
boxIntersects(boxfg, boxin, &intersects);
if (!intersects) {
L_INFO("found only noise on page %d\n", procName, pagenum);
boxDestroy(&boxfg);
}
boxDestroy(&boxin);
}
boxd = NULL;
if (!boxfg) {
L_INFO("no fg region found for page %d\n", procName, pagenum);
} else {
boxAdjustSides(boxfg, boxfg, -2, 2, -2, 2); /* tiny expansion */
boxd = boxTransform(boxfg, 0, 0, 2.0, 2.0);
/* Write image showing box for this page. This is to be
* bundled up into a pdf of all the pages, which can be
* generated by convertFilesToPdf() */
if (pdfdir) {
pixg2 = pixConvert1To4Cmap(pixb);
pixRenderBoxArb(pixg2, boxd, 3, 255, 0, 0);
snprintf(buf, sizeof(buf), "/tmp/%s/%05d.png", pdfdir, pagenum);
if (display) pixDisplay(pixg2, 700, 100);
pixWrite(buf, pixg2, IFF_PNG);
pixDestroy(&pixg2);
}
}
pixDestroy(&pixb);
pixDestroy(&pixb2);
pixDestroy(&pixseed);
pixDestroy(&pixsf);
pixDestroy(&pixm);
boxDestroy(&boxfg);
return boxd;
}
/*!
* pixaGenerateFont()
*
* Input: dir (directory holding image of character set)
* size (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
*
* These font generation functions use 9 sets, each with bitmaps
* of 94 ascii characters, all in Palatino-Roman font.
* Each input bitmap has 3 rows of characters. The range of
* ascii values in each row is as follows:
* row 0: 32-57 (32 is a space)
* row 1: 58-91 (92, '\', is not represented in this font)
* row 2: 93-126
* We LR flip the '/' char to generate a bitmap for the missing
* '\' character, so that we have representations of all 95
* printable chars.
*
* Computation of the bitmaps and baselines for a single
* font takes from 40 to 200 msec on a 2 GHz processor,
* depending on the size. Use pixaGetFont() to read the
* generated character set directly from files that were
* produced in prog/genfonts.c using this function.
*/
PIXA *
pixaGenerateFont(const char *dir,
l_int32 size,
l_int32 *pbl0,
l_int32 *pbl1,
l_int32 *pbl2)
{
char *pathname;
l_int32 fileno;
l_int32 i, j, nrows, nrowchars, nchars, h, yval;
l_int32 width, height;
l_int32 baseline[3];
l_int32 *tab;
BOX *box, *box1, *box2;
BOXA *boxar, *boxac, *boxacs;
PIX *pixs, *pixt1, *pixt2, *pixt3;
PIX *pixr, *pixrc, *pixc;
PIXA *pixa;
PROCNAME("pixaGenerateFont");
if (!pbl0 || !pbl1 || !pbl2)
return (PIXA *)ERROR_PTR("&bl not all defined", procName, NULL);
*pbl0 = *pbl1 = *pbl2 = 0;
fileno = (size / 2) - 2;
if (fileno < 0 || fileno > NFONTS)
return (PIXA *)ERROR_PTR("font size invalid", procName, NULL);
tab = makePixelSumTab8();
pathname = genPathname(dir, inputfonts[fileno]);
if ((pixs = pixRead(pathname)) == NULL)
return (PIXA *)ERROR_PTR("pixs not all defined", procName, NULL);
FREE(pathname);
pixa = pixaCreate(95);
pixt1 = pixMorphSequence(pixs, "c1.35 + c101.1", 0);
boxar = pixConnComp(pixt1, NULL, 8); /* one box for each row */
pixDestroy(&pixt1);
nrows = boxaGetCount(boxar);
#if DEBUG_FONT_GEN
fprintf(stderr, "For font %s, number of rows is %d\n",
inputfonts[fileno], nrows);
#endif /* DEBUG_FONT_GEN */
if (nrows != 3) {
L_INFO_INT2("nrows = %d; skipping font %d", procName, nrows, fileno);
return (PIXA *)ERROR_PTR("3 rows not generated", procName, NULL);
}
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
{ PIX *pixbl;
fprintf(stderr, "row %d, yval = %d, h = %d\n",
i, yval, pixGetHeight(pixr));
pixbl = pixCopy(NULL, pixr);
pixRenderLine(pixbl, 0, yval, pixGetWidth(pixbl), yval, 1,
L_FLIP_PIXELS);
if (i == 0 )
pixWrite("junktl0", pixbl, IFF_PNG);
else if (i == 1)
pixWrite("junktl1", pixbl, IFF_PNG);
else
pixWrite("junktl2", pixbl, IFF_PNG);
pixDestroy(&pixbl);
}
#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);
}
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. */
pixt2 = pixaGetPix(pixa, 0, L_CLONE);
width = 2 * pixGetWidth(pixt2);
height = pixGetHeight(pixt2);
pixDestroy(&pixt2);
pixt2 = pixCreate(width, height, 1);
pixaReplacePix(pixa, 0, pixt2, NULL);
/* Fix up the '\' character; use a LR flip of the '/' char */
pixt2 = pixaGetPix(pixa, 15, L_CLONE);
pixt3 = pixFlipLR(NULL, pixt2);
pixDestroy(&pixt2);
pixaReplacePix(pixa, 60, pixt3, NULL);
#if DEBUG_CHARS
{ PIX *pixd;
pixd = pixaDisplayTiled(pixa, 1500, 0, 10);
pixDisplay(pixd, 100 * i, 200);
pixDestroy(&pixd);
}
#endif /* DEBUG_CHARS */
pixDestroy(&pixs);
boxaDestroy(&boxar);
FREE(tab);
return pixa;
}
/*!
* boxaSort2d()
*
* Input: boxas
* &naa (<optional return> numaa with sorted indices
* whose values are the indices of the input array)
* delta1 (min overlap that permits aggregation of a box
* onto a boxa of horizontally-aligned boxes; pass 1)
* delta2 (min overlap that permits aggregation of a box
* onto a boxa of horizontally-aligned boxes; pass 2)
* minh1 (components less than this height either join an
* existing boxa or are set aside for pass 2)
* Return: boxaa (2d sorted version of boxa), or null on error
*
* Notes:
* (1) The final result is a sort where the 'fast scan' direction is
* left to right, and the 'slow scan' direction is from top
* to bottom. Each boxa in the boxaa represents a sorted set
* of boxes from left to right.
* (2) Two passes are used to aggregate the boxas, which can corresond
* to characters or words in a line of text. In pass 1, only
* taller components, which correspond to xheight or larger,
* are permitted to start a new boxa, whereas in pass 2,
* the remaining vertically-challenged components are allowed
* to join an existing boxa or start a new one.
* (3) If delta1 < 0, the first pass allows aggregation when
* boxes in the same boxa do not overlap vertically.
* The distance by which they can miss and still be aggregated
* is the absolute value |delta1|. Similar for delta2 on
* the second pass.
* (4) On the first pass, any component of height less than minh1
* cannot start a new boxa; it's put aside for later insertion.
* (5) On the second pass, any small component that doesn't align
* with an existing boxa can start a new one.
* (6) This can be used to identify lines of text from
* character or word bounding boxes.
*/
BOXAA *
boxaSort2d(BOXA *boxas,
NUMAA **pnaad,
l_int32 delta1,
l_int32 delta2,
l_int32 minh1)
{
l_int32 i, index, h, nt, ne, n, m, ival;
BOX *box;
BOXA *boxa, *boxae, *boxan, *boxat1, *boxat2, *boxav, *boxavs;
BOXAA *baa, *baad;
NUMA *naindex, *nae, *nan, *nah, *nav, *nat1, *nat2, *nad;
NUMAA *naa, *naad;
PROCNAME("boxaSort2d");
if (pnaad) *pnaad = NULL;
if (!boxas)
return (BOXAA *)ERROR_PTR("boxas not defined", procName, NULL);
/* Sort from left to right */
if ((boxa = boxaSort(boxas, L_SORT_BY_X, L_SORT_INCREASING, &naindex))
== NULL)
return (BOXAA *)ERROR_PTR("boxa not made", procName, NULL);
/* First pass: assign taller boxes to boxa by row */
nt = boxaGetCount(boxa);
baa = boxaaCreate(0);
naa = numaaCreate(0);
boxae = boxaCreate(0); /* save small height boxes here */
nae = numaCreate(0); /* keep track of small height boxes */
for (i = 0; i < nt; i++) {
box = boxaGetBox(boxa, i, L_CLONE);
boxGetGeometry(box, NULL, NULL, NULL, &h);
if (h < minh1) { /* save for 2nd pass */
boxaAddBox(boxae, box, L_INSERT);
numaAddNumber(nae, i);
}
else {
n = boxaaGetCount(baa);
boxaaAlignBox(baa, box, delta1, &index);
if (index < n) { /* append to an existing boxa */
boxaaAddBox(baa, index, box, L_INSERT);
}
else { /* doesn't align, need new boxa */
boxan = boxaCreate(0);
boxaAddBox(boxan, box, L_INSERT);
boxaaAddBoxa(baa, boxan, L_INSERT);
nan = numaCreate(0);
numaaAddNuma(naa, nan, L_INSERT);
}
numaGetIValue(naindex, i, &ival);
numaaAddNumber(naa, index, ival);
}
}
boxaDestroy(&boxa);
numaDestroy(&naindex);
/* Second pass: feed in small height boxes;
* TODO: this correctly, using local y position! */
ne = boxaGetCount(boxae);
for (i = 0; i < ne; i++) {
box = boxaGetBox(boxae, i, L_CLONE);
n = boxaaGetCount(baa);
boxaaAlignBox(baa, box, delta2, &index);
if (index < n) { /* append to an existing boxa */
boxaaAddBox(baa, index, box, L_INSERT);
}
else { /* doesn't align, need new boxa */
boxan = boxaCreate(0);
boxaAddBox(boxan, box, L_INSERT);
boxaaAddBoxa(baa, boxan, L_INSERT);
nan = numaCreate(0);
numaaAddNuma(naa, nan, L_INSERT);
}
numaGetIValue(nae, i, &ival); /* location in original boxas */
numaaAddNumber(naa, index, ival);
}
/* Sort each boxa in the boxaa */
m = boxaaGetCount(baa);
for (i = 0; i < m; i++) {
boxat1 = boxaaGetBoxa(baa, i, L_CLONE);
boxat2 = boxaSort(boxat1, L_SORT_BY_X, L_SORT_INCREASING, &nah);
boxaaReplaceBoxa(baa, i, boxat2);
nat1 = numaaGetNuma(naa, i, L_CLONE);
nat2 = numaSortByIndex(nat1, nah);
numaaReplaceNuma(naa, i, nat2);
boxaDestroy(&boxat1);
numaDestroy(&nat1);
numaDestroy(&nah);
}
/* Sort boxa vertically within boxaa, using the first box
* in each boxa. */
m = boxaaGetCount(baa);
boxav = boxaCreate(m); /* holds first box in each boxa in baa */
naad = numaaCreate(m);
if (pnaad)
*pnaad = naad;
baad = boxaaCreate(m);
for (i = 0; i < m; i++) {
boxat1 = boxaaGetBoxa(baa, i, L_CLONE);
box = boxaGetBox(boxat1, 0, L_CLONE);
boxaAddBox(boxav, box, L_INSERT);
boxaDestroy(&boxat1);
}
boxavs = boxaSort(boxav, L_SORT_BY_Y, L_SORT_INCREASING, &nav);
for (i = 0; i < m; i++) {
numaGetIValue(nav, i, &index);
boxa = boxaaGetBoxa(baa, index, L_CLONE);
boxaaAddBoxa(baad, boxa, L_INSERT);
nad = numaaGetNuma(naa, index, L_CLONE);
numaaAddNuma(naad, nad, L_INSERT);
}
/* fprintf(stderr, "box count = %d, numaa count = %d\n", nt,
numaaGetNumberCount(naad)); */
boxaaDestroy(&baa);
boxaDestroy(&boxav);
boxaDestroy(&boxavs);
boxaDestroy(&boxae);
numaDestroy(&nav);
numaDestroy(&nae);
numaaDestroy(&naa);
if (!pnaad)
numaaDestroy(&naad);
return baad;
}
