void
MakeWordBoxes2(PIX *pixs,
l_int32 reduction,
L_REGPARAMS *rp)
{
l_int32 default_minwidth = 10;
l_int32 default_minheight = 10;
l_int32 default_maxwidth = 400;
l_int32 default_maxheight = 70;
l_int32 minwidth, minheight, maxwidth, maxheight;
BOXA *boxa1, *boxa2;
NUMA *na;
PIX *pixd1, *pixd2;
PIXA *pixa;
minwidth = default_minwidth / reduction;
minheight = default_minheight / reduction;
maxwidth = default_maxwidth / reduction;
maxheight = default_maxheight / reduction;
/* Get the word boxes */
pixGetWordsInTextlines(pixs, reduction, minwidth, minheight,
maxwidth, maxheight, &boxa1, &pixa, &na);
pixaDestroy(&pixa);
numaDestroy(&na);
if (reduction == 1)
boxa2 = boxaCopy(boxa1, L_CLONE);
else
boxa2 = boxaTransform(boxa1, 0, 0, 2.0, 2.0);
pixd1 = pixConvertTo8(pixs, 1);
pixRenderBoxaArb(pixd1, boxa2, 2, 255, 0, 0);
regTestWritePixAndCheck(rp, pixd1, IFF_PNG);
pixDisplayWithTitle(pixd1, 800, 100, NULL, rp->display);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
/* Do it again with this interface. The result should be the same. */
pixGetWordBoxesInTextlines(pixs, reduction, minwidth, minheight,
maxwidth, maxheight, &boxa1, NULL);
if (reduction == 1)
boxa2 = boxaCopy(boxa1, L_CLONE);
else
boxa2 = boxaTransform(boxa1, 0, 0, 2.0, 2.0);
pixd2 = pixConvertTo8(pixs, 1);
pixRenderBoxaArb(pixd2, boxa2, 2, 255, 0, 0);
if (regTestComparePix(rp, pixd1, pixd2)) {
L_ERROR("pix not the same", "MakeWordBoxes2");
pixDisplayWithTitle(pixd2, 800, 100, NULL, rp->display);
}
pixDestroy(&pixd1);
pixDestroy(&pixd2);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
return;
}
void
MakeWordBoxes1(PIX *pixs,
l_int32 maxdil,
L_REGPARAMS *rp) {
BOXA *boxa;
PIX *pix1, *pixd;
pixWordMaskByDilation(pixs, maxdil, &pix1, NULL);
pixd = NULL;
if (pix1) {
boxa = pixConnComp(pix1, NULL, 8);
pixd = pixConvertTo8(pixs, 1);
pixRenderBoxaArb(pixd, boxa, 2, 255, 0, 0);
boxaDestroy(&boxa);
}
regTestWritePixAndCheck(rp, pixd, IFF_PNG);
pixDisplayWithTitle(pixd, 0, 100, NULL, rp->display);
pixDestroy(&pix1);
pixDestroy(&pixd);
return;
}
static PIX *
DisplayBoxa(BOXA *boxa)
{
l_int32 w, h;
BOX *box;
PIX *pix1, *pix2, *pix3;
PIXA *pixa;
pixa = pixaCreate(2);
boxaGetExtent(boxa, &w, &h, &box);
pix1 = pixCreate(w, h, 1);
pixMaskBoxa(pix1, pix1, boxa, L_SET_PIXELS);
pixaAddPix(pixa, pix1, L_INSERT);
pix2 = pixCreate(w, h, 32);
pixSetAll(pix2);
pixRenderBoxaArb(pix2, boxa, 2, 0, 255, 0);
pixRenderBoxArb(pix2, box, 3, 255, 0, 0);
pixaAddPix(pixa, pix2, L_INSERT);
pix3 = pixaDisplayTiledInRows(pixa, 32, 1000, 1.0, 0, 30, 2);
boxDestroy(&box);
pixaDestroy(&pixa);
return pix3;
}
l_int32 main(int argc,
char **argv)
{
char buf[64];
BOXA *boxa1, *boxa2, *boxa3, *boxa4;
L_DEWARP *dew;
L_DEWARPA *dewa;
PIX *pixs, *pixn, *pixg, *pixb, *pix2, *pix3, *pix4, *pix5, *pix6;
lept_mkdir("lept");
snprintf(buf, sizeof(buf), "cat.%03d.jpg", pageno);
pixs = pixRead(buf);
dewa = dewarpaCreate(40, 30, 1, 15, 10);
dewarpaUseBothArrays(dewa, 1);
/* Normalize for varying background and binarize */
pixn = pixBackgroundNormSimple(pixs, NULL, NULL);
pixg = pixConvertRGBToGray(pixn, 0.5, 0.3, 0.2);
pixb = pixThresholdToBinary(pixg, 130);
pixDisplay(pixb, 0, 100);
/* Build the model */
dew = dewarpCreate(pixb, pageno);
dewarpaInsertDewarp(dewa, dew);
if (build_output) {
snprintf(buf, sizeof(buf), "/tmp/lept/dewarp_build_%d.pdf", pageno);
dewarpBuildPageModel(dew, buf);
} else {
dewarpBuildPageModel(dew, NULL);
}
/* Apply the model */
dewarpPopulateFullRes(dew, pixg, 0, 0);
if (apply_output) {
snprintf(buf, sizeof(buf), "/tmp/lept/dewarp_apply_%d.pdf", pageno);
dewarpaApplyDisparity(dewa, pageno, pixb, 200, 0, 0, &pix2, buf);
} else {
dewarpaApplyDisparity(dewa, pageno, pixb, 200, 0, 0, &pix2, NULL);
}
pixDisplay(pix2, 200, 100);
/* Reverse direction: get the word boxes for the dewarped pix ... */
pixGetWordBoxesInTextlines(pix2, 5, 5, 500, 100, &boxa1, NULL);
pix3 = pixConvertTo32(pix2);
pixRenderBoxaArb(pix3, boxa1, 2, 255, 0, 0);
pixDisplay(pix3, 400, 100);
/* ... and map to the word boxes for the input image */
if (map_output) {
snprintf(buf, sizeof(buf), "/tmp/lept/dewarp_map1_%d.pdf", pageno);
dewarpaApplyDisparityBoxa(dewa, pageno, pix2, boxa1, 0, 0, 0, &boxa2,
buf);
} else {
dewarpaApplyDisparityBoxa(dewa, pageno, pix2, boxa1, 0, 0, 0, &boxa2,
NULL);
}
pix4 = pixConvertTo32(pixb);
pixRenderBoxaArb(pix4, boxa2, 2, 0, 255, 0);
pixDisplay(pix4, 600, 100);
/* Forward direction: get the word boxes for the input pix ... */
pixGetWordBoxesInTextlines(pixb, 5, 5, 500, 100, &boxa3, NULL);
pix5 = pixConvertTo32(pixb);
pixRenderBoxaArb(pix5, boxa3, 2, 255, 0, 0);
pixDisplay(pix5, 800, 100);
/* ... and map to the word boxes for the dewarped image */
if (map_output) {
snprintf(buf, sizeof(buf), "/tmp/lept/dewarp_map2_%d.pdf", pageno);
dewarpaApplyDisparityBoxa(dewa, pageno, pixb, boxa3, 1, 0, 0, &boxa4,
buf);
} else {
dewarpaApplyDisparityBoxa(dewa, pageno, pixb, boxa3, 1, 0, 0, &boxa4,
NULL);
}
pix6 = pixConvertTo32(pix2);
pixRenderBoxaArb(pix6, boxa4, 2, 0, 255, 0);
pixDisplay(pix6, 1000, 100);
dewarpaDestroy(&dewa);
pixDestroy(&pixs);
pixDestroy(&pixn);
pixDestroy(&pixg);
pixDestroy(&pixb);
pixDestroy(&pix2);
pixDestroy(&pix3);
pixDestroy(&pix4);
pixDestroy(&pix5);
pixDestroy(&pix6);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
boxaDestroy(&boxa3);
boxaDestroy(&boxa4);
return 0;
}
/*!
* \brief pixItalicWords()
*
* \param[in] pixs 1 bpp
* \param[in] boxaw [optional] word bounding boxes; can be NULL
* \param[in] pixw [optional] word box mask; can be NULL
* \param[out] pboxa boxa of italic words
* \param[in] debugflag 1 for debug output; 0 otherwise
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) You can input the bounding boxes for the words in one of
* two forms: as bounding boxes (%boxaw) or as a word mask with
* the word bounding boxes filled (%pixw). For example,
* to compute %pixw, you can use pixWordMaskByDilation().
* (2) Alternatively, you can set both of these inputs to NULL,
* in which case the word mask is generated here. This is
* done by dilating and closing the input image to connect
* letters within a word, while leaving the words separated.
* The parameters are chosen under the assumption that the
* input is 10 to 12 pt text, scanned at about 300 ppi.
* (3) sel_ital1 and sel_ital2 detect the right edges that are
* nearly vertical, at approximately the angle of italic
* strokes. We use the right edge to avoid getting seeds
* from lower-case 'y'. The typical italic slant has a smaller
* angle with the vertical than the 'W', so in most cases we
* will not trigger on the slanted lines in the 'W'.
* (4) Note that sel_ital2 is shorter than sel_ital1. It is
* more appropriate for a typical font scanned at 200 ppi.
* </pre>
*/
l_int32
pixItalicWords(PIX *pixs,
BOXA *boxaw,
PIX *pixw,
BOXA **pboxa,
l_int32 debugflag)
{
char opstring[32];
l_int32 size;
BOXA *boxa;
PIX *pixsd, *pixm, *pixd;
SEL *sel_ital1, *sel_ital2, *sel_ital3;
PROCNAME("pixItalicWords");
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if (!pboxa)
return ERROR_INT("&boxa not defined", procName, 1);
if (boxaw && pixw)
return ERROR_INT("both boxaw and pixw are defined", procName, 1);
sel_ital1 = selCreateFromString(str_ital1, 13, 6, NULL);
sel_ital2 = selCreateFromString(str_ital2, 10, 6, NULL);
sel_ital3 = selCreateFromString(str_ital3, 4, 2, NULL);
/* Make the italic seed: extract with HMT; remove noise.
* The noise removal close/open is important to exclude
* situations where a small slanted line accidentally
* matches sel_ital1. */
pixsd = pixHMT(NULL, pixs, sel_ital1);
pixClose(pixsd, pixsd, sel_ital3);
pixOpen(pixsd, pixsd, sel_ital3);
/* Make the word mask. Use input boxes or mask if given. */
size = 0; /* init */
if (boxaw) {
pixm = pixCreateTemplate(pixs);
pixMaskBoxa(pixm, pixm, boxaw, L_SET_PIXELS);
} else if (pixw) {
pixm = pixClone(pixw);
} else {
pixWordMaskByDilation(pixs, NULL, &size, NULL);
L_INFO("dilation size = %d\n", procName, size);
snprintf(opstring, sizeof(opstring), "d1.5 + c%d.1", size);
pixm = pixMorphSequence(pixs, opstring, 0);
}
/* Binary reconstruction to fill in those word mask
* components for which there is at least one seed pixel. */
pixd = pixSeedfillBinary(NULL, pixsd, pixm, 8);
boxa = pixConnComp(pixd, NULL, 8);
*pboxa = boxa;
if (debugflag) {
/* Save results at at 2x reduction */
lept_mkdir("lept/ital");
l_int32 res, upper;
BOXA *boxat;
GPLOT *gplot;
NUMA *na;
PIXA *pad;
PIX *pix1, *pix2, *pix3;
pad = pixaCreate(0);
boxat = pixConnComp(pixm, NULL, 8);
boxaWrite("/tmp/lept/ital/ital.ba", boxat);
pixSaveTiledOutline(pixs, pad, 0.5, 1, 20, 2, 32); /* orig */
pixSaveTiledOutline(pixsd, pad, 0.5, 1, 20, 2, 0); /* seed */
pix1 = pixConvertTo32(pixm);
pixRenderBoxaArb(pix1, boxat, 3, 255, 0, 0);
pixSaveTiledOutline(pix1, pad, 0.5, 1, 20, 2, 0); /* mask + outline */
pixDestroy(&pix1);
pixSaveTiledOutline(pixd, pad, 0.5, 1, 20, 2, 0); /* ital mask */
pix1 = pixConvertTo32(pixs);
pixRenderBoxaArb(pix1, boxa, 3, 255, 0, 0);
pixSaveTiledOutline(pix1, pad, 0.5, 1, 20, 2, 0); /* orig + outline */
pixDestroy(&pix1);
pix1 = pixCreateTemplate(pixs);
pix2 = pixSetBlackOrWhiteBoxa(pix1, boxa, L_SET_BLACK);
pixCopy(pix1, pixs);
pix3 = pixDilateBrick(NULL, pixs, 3, 3);
pixCombineMasked(pix1, pix3, pix2);
pixSaveTiledOutline(pix1, pad, 0.5, 1, 20, 2, 0); /* ital bolded */
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
pix2 = pixaDisplay(pad, 0, 0);
pixWrite("/tmp/lept/ital/ital.png", pix2, IFF_PNG);
pixDestroy(&pix2);
/* Assuming the image represents 6 inches of actual page width,
* the pixs resolution is approximately
* (width of pixs in pixels) / 6
* and the images have been saved at half this resolution. */
res = pixGetWidth(pixs) / 12;
L_INFO("resolution = %d\n", procName, res);
l_pdfSetDateAndVersion(0);
pixaConvertToPdf(pad, res, 1.0, L_FLATE_ENCODE, 75, "Italic Finder",
"/tmp/lept/ital/ital.pdf");
l_pdfSetDateAndVersion(1);
pixaDestroy(&pad);
boxaDestroy(&boxat);
/* Plot histogram of horizontal white run sizes. A small
* initial vertical dilation removes most runs that are neither
* inter-character nor inter-word. The larger first peak is
* from inter-character runs, and the smaller second peak is
* from inter-word runs. */
pix1 = pixDilateBrick(NULL, pixs, 1, 15);
upper = L_MAX(30, 3 * size);
na = pixRunHistogramMorph(pix1, L_RUN_OFF, L_HORIZ, upper);
pixDestroy(&pix1);
gplot = gplotCreate("/tmp/lept/ital/runhisto", GPLOT_PNG,
"Histogram of horizontal runs of white pixels, vs length",
"run length", "number of runs");
gplotAddPlot(gplot, NULL, na, GPLOT_LINES, "plot1");
gplotMakeOutput(gplot);
gplotDestroy(&gplot);
numaDestroy(&na);
}
selDestroy(&sel_ital1);
selDestroy(&sel_ital2);
selDestroy(&sel_ital3);
pixDestroy(&pixsd);
pixDestroy(&pixm);
pixDestroy(&pixd);
return 0;
}
main(int argc,
char **argv)
{
l_int32 i, j, w, h;
l_int32 minsum[5] = { 2, 40, 50, 50, 70};
l_int32 skipdist[5] = { 5, 5, 10, 10, 30};
l_int32 delta[5] = { 2, 10, 10, 25, 40};
l_int32 maxbg[5] = {10, 15, 10, 20, 40};
BOX *box1, *box2, *box3, *box4;
BOXA *boxa;
PIX *pixs, *pixc, *pixt, *pixd, *pix32;
PIXA *pixas, *pixad;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
/* Generate and save 1 bpp masks */
pixas = pixaCreate(0);
pixs = pixCreate(300, 250, 1);
pixSetAll(pixs);
box1 = boxCreate(50, 0, 140, 25);
box2 = boxCreate(120, 100, 100, 25);
box3 = boxCreate(75, 170, 80, 20);
box4 = boxCreate(150, 80, 25, 70);
pixClearInRect(pixs, box1);
pixaAddPix(pixas, pixs, L_COPY);
pixt = pixRotateOrth(pixs, 1);
pixaAddPix(pixas, pixt, L_INSERT);
pixClearInRect(pixs, box2);
pixaAddPix(pixas, pixs, L_COPY);
pixt = pixRotateOrth(pixs, 1);
pixaAddPix(pixas, pixt, L_INSERT);
pixClearInRect(pixs, box3);
pixaAddPix(pixas, pixs, L_COPY);
pixt = pixRotateOrth(pixs, 1);
pixaAddPix(pixas, pixt, L_INSERT);
pixClearInRect(pixs, box4);
pixaAddPix(pixas, pixs, L_COPY);
pixt = pixRotateOrth(pixs, 1);
pixaAddPix(pixas, pixt, L_INSERT);
boxDestroy(&box1);
boxDestroy(&box2);
boxDestroy(&box3);
boxDestroy(&box4);
pixDestroy(&pixs);
/* Do 5 splittings on each of the 8 masks */
pixad = pixaCreate(0);
for (j = 0; j < 8; j++) {
pixt = pixaGetPix(pixas, j, L_CLONE);
pixGetDimensions(pixt, &w, &h, NULL);
pix32 = pixCreate(w, h, 32);
pixSetAll(pix32);
pixPaintThroughMask(pix32, pixt, 0, 0, 0xc0c0c000);
pixSaveTiled(pix32, pixad, 1, 1, 30, 32);
for (i = 0; i < 5; i++) {
pixc = pixCopy(NULL, pix32);
boxa = pixSplitComponentIntoBoxa(pixt, NULL, minsum[i], skipdist[i],
delta[i], maxbg[i], 0, 1);
/* boxaWriteStream(stderr, boxa); */
pixd = pixBlendBoxaRandom(pixc, boxa, 0.4);
pixRenderBoxaArb(pixd, boxa, 2, 255, 0, 0);
pixSaveTiled(pixd, pixad, 1, 0, 30, 32);
pixDestroy(&pixd);
pixDestroy(&pixc);
boxaDestroy(&boxa);
}
pixDestroy(&pixt);
pixDestroy(&pix32);
}
/* Display results */
pixd = pixaDisplay(pixad, 0, 0);
regTestWritePixAndCheck(rp, pixd, IFF_PNG); /* 0 */
pixDisplayWithTitle(pixd, 100, 100, NULL, rp->display);
pixDestroy(&pixd);
pixaDestroy(&pixad);
/* Put the 8 masks all together, and split 5 ways */
pixad = pixaCreate(0);
pixs = pixaDisplayOnLattice(pixas, 325, 325);
pixGetDimensions(pixs, &w, &h, NULL);
pix32 = pixCreate(w, h, 32);
pixSetAll(pix32);
pixPaintThroughMask(pix32, pixs, 0, 0, 0xc0c0c000);
pixSaveTiled(pix32, pixad, 1, 1, 30, 32);
for (i = 0; i < 5; i++) {
pixc = pixCopy(NULL, pix32);
boxa = pixSplitIntoBoxa(pixs, minsum[i], skipdist[i],
delta[i], maxbg[i], 0, 1);
/* boxaWriteStream(stderr, boxa); */
pixd = pixBlendBoxaRandom(pixc, boxa, 0.4);
pixRenderBoxaArb(pixd, boxa, 2, 255, 0, 0);
pixSaveTiled(pixd, pixad, 1, 0, 30, 32);
pixDestroy(&pixd);
pixDestroy(&pixc);
boxaDestroy(&boxa);
}
pixDestroy(&pix32);
pixDestroy(&pixs);
/* Display results */
pixd = pixaDisplay(pixad, 0, 0);
regTestWritePixAndCheck(rp, pixd, IFF_PNG); /* 1 */
pixDisplayWithTitle(pixd, 600, 100, NULL, rp->display);
pixDestroy(&pixd);
pixaDestroy(&pixad);
pixaDestroy(&pixas);
regTestCleanup(rp);
return 0;
}
/*!
* dewarpaApplyDisparityBoxa()
*
* Input: dewa
* pageno (of page model to be used; may be a ref model)
* pixs (initial pix reference; for alignment and debugging)
* boxas (boxa to be mapped)
* mapdir (1 if mapping forward from original to dewarped;
* 0 if backward)
* x, y (origin for generation of disparity arrays with
* respect to the source region)
* &boxad (<return> disparity corrected boxa)
* debugfile (use null to skip writing this)
* Return: 0 if OK, 1 on error (no models or ref models available)
*
* Notes:
* (1) This applies the disparity arrays in one of two mapping directions
* to the specified boxa. It can be used in the backward direction
* to locate a box in the original coordinates that would have
* been dewarped to to the specified image.
* (2) If there is no model for @pageno, this will use the model for
* 'refpage' and put the result in the dew for @pageno.
* (3) This works with both stripped and full resolution page models.
* If the full res disparity array(s) are missing, they are remade.
* (4) If an error occurs, a copy of the input boxa is returned.
*/
l_int32
dewarpaApplyDisparityBoxa(L_DEWARPA *dewa,
l_int32 pageno,
PIX *pixs,
BOXA *boxas,
l_int32 mapdir,
l_int32 x,
l_int32 y,
BOXA **pboxad,
const char *debugfile)
{
l_int32 debug_out;
L_DEWARP *dew1, *dew;
BOXA *boxav, *boxah;
PIX *pixv, *pixh;
PROCNAME("dewarpaApplyDisparityBoxa");
/* Initialize the output with the input, so we'll have that
* in case we can't apply the page model. */
if (!pboxad)
return ERROR_INT("&boxad not defined", procName, 1);
*pboxad = boxaCopy(boxas, L_CLONE);
/* Find the appropriate dew to use and fully populate its array(s) */
if (dewarpaApplyInit(dewa, pageno, pixs, x, y, &dew, debugfile))
return ERROR_INT("no model available", procName, 1);
/* Correct for vertical disparity and save the result */
if ((boxav = boxaApplyDisparity(dew, boxas, L_VERT, mapdir)) == NULL) {
dewarpMinimize(dew);
return ERROR_INT("boxa1 not made", procName, 1);
}
boxaDestroy(pboxad);
*pboxad = boxav;
pixv = NULL;
pixh = NULL;
if (debugfile && mapdir != 1)
L_INFO("Reverse map direction; no debug output\n", procName);
debug_out = debugfile && (mapdir == 1);
if (debug_out) {
PIX *pix1;
lept_rmdir("lept/dewboxa"); /* remove previous images */
lept_mkdir("lept/dewboxa");
pix1 = pixConvertTo32(pixs);
pixRenderBoxaArb(pix1, boxas, 2, 255, 0, 0);
pixWrite("/tmp/lept/dewboxa/01.png", pix1, IFF_PNG);
pixDestroy(&pix1);
pixv = pixApplyVertDisparity(dew, pixs, 255);
pix1 = pixConvertTo32(pixv);
pixRenderBoxaArb(pix1, boxav, 2, 0, 255, 0);
pixWrite("/tmp/lept/dewboxa/02.png", pix1, IFF_PNG);
pixDestroy(&pix1);
}
/* Optionally, correct for horizontal disparity */
if (dewa->useboth && dew->hsuccess) {
if (dew->hvalid == FALSE) {
L_INFO("invalid horiz model for page %d\n", procName, pageno);
} else {
boxah = boxaApplyDisparity(dew, boxav, L_HORIZ, mapdir);
if (!boxah) {
L_ERROR("horiz disparity fails on page %d\n", procName, pageno);
} else {
boxaDestroy(pboxad);
*pboxad = boxah;
if (debug_out) {
PIX *pix1;
pixh = pixApplyHorizDisparity(dew, pixv, 255);
pix1 = pixConvertTo32(pixh);
pixRenderBoxaArb(pix1, boxah, 2, 0, 0, 255);
pixWrite("/tmp/lept/dewboxa/03.png", pix1, IFF_PNG);
pixDestroy(&pixh);
pixDestroy(&pix1);
}
}
}
}
if (debug_out) {
pixDestroy(&pixv);
dew1 = dewarpaGetDewarp(dewa, pageno);
dewarpDebug(dew1, "lept/dewapply", 0);
convertFilesToPdf("/tmp/lept/dewboxa", NULL, 135, 1.0, 0, 0,
"Dewarp Apply Disparity Boxa", debugfile);
fprintf(stderr, "Dewarp Apply Disparity Boxa pdf file: %s\n",
debugfile);
}
/* Get rid of the large full res disparity arrays */
dewarpMinimize(dew);
return 0;
}