l_int32
DoPageSegmentation(PIX *pixs, /* should be at least 300 ppi */
l_int32 which) /* 1, 2, 3, 4 */
{
char buf[256];
l_int32 zero;
BOXA *boxatm, *boxahm;
PIX *pixr; /* image reduced to 150 ppi */
PIX *pixhs; /* image of halftone seed, 150 ppi */
PIX *pixm; /* image of mask of components, 150 ppi */
PIX *pixhm1; /* image of halftone mask, 150 ppi */
PIX *pixhm2; /* image of halftone mask, 300 ppi */
PIX *pixht; /* image of halftone components, 150 ppi */
PIX *pixnht; /* image without halftone components, 150 ppi */
PIX *pixi; /* inverted image, 150 ppi */
PIX *pixvws; /* image of vertical whitespace, 150 ppi */
PIX *pixm1; /* image of closed textlines, 150 ppi */
PIX *pixm2; /* image of refined text line mask, 150 ppi */
PIX *pixm3; /* image of refined text line mask, 300 ppi */
PIX *pixb1; /* image of text block mask, 150 ppi */
PIX *pixb2; /* image of text block mask, 300 ppi */
PIX *pixnon; /* image of non-text or halftone, 150 ppi */
PIX *pix1, *pix2, *pix3, *pix4;
PIXA *pixa;
PIXCMAP *cmap;
PTAA *ptaa;
l_int32 ht_flag = 0;
l_int32 ws_flag = 0;
l_int32 text_flag = 0;
l_int32 block_flag = 0;
PROCNAME("DoPageSegmentation");
if (which == 1)
ht_flag = 1;
else if (which == 2)
ws_flag = 1;
else if (which == 3)
text_flag = 1;
else if (which == 4)
block_flag = 1;
else
return ERROR_INT("invalid parameter: not in [1...4]", procName, 1);
pixa = pixaCreate(0);
lept_mkdir("lept/livre");
/* Reduce to 150 ppi */
pix1 = pixScaleToGray2(pixs);
if (ws_flag || ht_flag || block_flag) pixaAddPix(pixa, pix1, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/orig.gray.150.png", pix1, IFF_PNG);
pixDestroy(&pix1);
pixr = pixReduceRankBinaryCascade(pixs, 1, 0, 0, 0);
/* Get seed for halftone parts */
pix1 = pixReduceRankBinaryCascade(pixr, 4, 4, 3, 0);
pix2 = pixOpenBrick(NULL, pix1, 5, 5);
pixhs = pixExpandBinaryPower2(pix2, 8);
if (ht_flag) pixaAddPix(pixa, pixhs, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/htseed.150.png", pixhs, IFF_PNG);
pixDestroy(&pix1);
pixDestroy(&pix2);
/* Get mask for connected regions */
pixm = pixCloseSafeBrick(NULL, pixr, 4, 4);
if (ht_flag) pixaAddPix(pixa, pixm, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/ccmask.150.png", pixm, IFF_PNG);
/* Fill seed into mask to get halftone mask */
pixhm1 = pixSeedfillBinary(NULL, pixhs, pixm, 4);
if (ht_flag) pixaAddPix(pixa, pixhm1, L_COPY);
if (which == 1) pixWrite("/tmp/lept/livre/htmask.150.png", pixhm1, IFF_PNG);
pixhm2 = pixExpandBinaryPower2(pixhm1, 2);
/* Extract halftone stuff */
pixht = pixAnd(NULL, pixhm1, pixr);
if (which == 1) pixWrite("/tmp/lept/livre/ht.150.png", pixht, IFF_PNG);
/* Extract non-halftone stuff */
pixnht = pixXor(NULL, pixht, pixr);
if (text_flag) pixaAddPix(pixa, pixnht, L_COPY);
if (which == 1) pixWrite("/tmp/lept/livre/text.150.png", pixnht, IFF_PNG);
pixZero(pixht, &zero);
if (zero)
fprintf(stderr, "No halftone parts found\n");
else
fprintf(stderr, "Halftone parts found\n");
/* Get bit-inverted image */
pixi = pixInvert(NULL, pixnht);
if (ws_flag) pixaAddPix(pixa, pixi, L_COPY);
if (which == 1) pixWrite("/tmp/lept/livre/invert.150.png", pixi, IFF_PNG);
/* The whitespace mask will break textlines where there
* is a large amount of white space below or above.
* We can prevent this by identifying regions of the
* inverted image that have large horizontal (bigger than
* the separation between columns) and significant
* vertical extent (bigger than the separation between
* textlines), and subtracting this from the whitespace mask. */
pix1 = pixMorphCompSequence(pixi, "o80.60", 0);
pix2 = pixSubtract(NULL, pixi, pix1);
if (ws_flag) pixaAddPix(pixa, pix2, L_COPY);
pixDestroy(&pix1);
/* Identify vertical whitespace by opening inverted image */
pix3 = pixOpenBrick(NULL, pix2, 5, 1); /* removes thin vertical lines */
pixvws = pixOpenBrick(NULL, pix3, 1, 200); /* gets long vertical lines */
if (text_flag || ws_flag) pixaAddPix(pixa, pixvws, L_COPY);
if (which == 1) pixWrite("/tmp/lept/livre/vertws.150.png", pixvws, IFF_PNG);
pixDestroy(&pix2);
pixDestroy(&pix3);
/* Get proto (early processed) text line mask. */
/* First close the characters and words in the textlines */
pixm1 = pixCloseSafeBrick(NULL, pixnht, 30, 1);
if (text_flag) pixaAddPix(pixa, pixm1, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/textmask1.150.png", pixm1, IFF_PNG);
/* Next open back up the vertical whitespace corridors */
pixm2 = pixSubtract(NULL, pixm1, pixvws);
if (which == 1)
pixWrite("/tmp/lept/livre/textmask2.150.png", pixm2, IFF_PNG);
/* Do a small opening to remove noise */
pixOpenBrick(pixm2, pixm2, 3, 3);
if (text_flag) pixaAddPix(pixa, pixm2, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/textmask3.150.png", pixm2, IFF_PNG);
pixm3 = pixExpandBinaryPower2(pixm2, 2);
/* Join pixels vertically to make text block mask */
pixb1 = pixMorphSequence(pixm2, "c1.10 + o4.1", 0);
if (block_flag) pixaAddPix(pixa, pixb1, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/textblock1.150.png", pixb1, IFF_PNG);
/* Solidify the textblock mask and remove noise:
* (1) For each c.c., close the blocks and dilate slightly
* to form a solid mask.
* (2) Small horizontal closing between components
* (3) Open the white space between columns, again
* (4) Remove small components */
pix1 = pixMorphSequenceByComponent(pixb1, "c30.30 + d3.3", 8, 0, 0, NULL);
pixCloseSafeBrick(pix1, pix1, 10, 1);
if (block_flag) pixaAddPix(pixa, pix1, L_COPY);
pix2 = pixSubtract(NULL, pix1, pixvws);
pix3 = pixSelectBySize(pix2, 25, 5, 8, L_SELECT_IF_BOTH,
L_SELECT_IF_GTE, NULL);
if (block_flag) pixaAddPix(pixa, pix3, L_COPY);
if (which == 1)
pixWrite("/tmp/lept/livre/textblock2.150.png", pix3, IFF_PNG);
pixb2 = pixExpandBinaryPower2(pix3, 2);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
/* Identify the outlines of each textblock */
ptaa = pixGetOuterBordersPtaa(pixb2);
pix1 = pixRenderRandomCmapPtaa(pixb2, ptaa, 1, 8, 1);
cmap = pixGetColormap(pix1);
pixcmapResetColor(cmap, 0, 130, 130, 130); /* set interior to gray */
if (which == 1)
pixWrite("/tmp/lept/livre/textblock3.300.png", pix1, IFF_PNG);
pixDisplayWithTitle(pix1, 480, 360, "textblock mask with outlines", DFLAG);
ptaaDestroy(&ptaa);
pixDestroy(&pix1);
/* Fill line mask (as seed) into the original */
pix1 = pixSeedfillBinary(NULL, pixm3, pixs, 8);
pixOr(pixm3, pixm3, pix1);
pixDestroy(&pix1);
if (which == 1)
pixWrite("/tmp/lept/livre/textmask.300.png", pixm3, IFF_PNG);
pixDisplayWithTitle(pixm3, 480, 360, "textline mask 4", DFLAG);
/* Fill halftone mask (as seed) into the original */
pix1 = pixSeedfillBinary(NULL, pixhm2, pixs, 8);
pixOr(pixhm2, pixhm2, pix1);
pixDestroy(&pix1);
if (which == 1)
pixWrite("/tmp/lept/livre/htmask.300.png", pixhm2, IFF_PNG);
pixDisplayWithTitle(pixhm2, 520, 390, "halftonemask 2", DFLAG);
/* Find objects that are neither text nor halftones */
pix1 = pixSubtract(NULL, pixs, pixm3); /* remove text pixels */
pixnon = pixSubtract(NULL, pix1, pixhm2); /* remove halftone pixels */
pixDestroy(&pix1);
if (which == 1)
pixWrite("/tmp/lept/livre/other.300.png", pixnon, IFF_PNG);
pixDisplayWithTitle(pixnon, 540, 420, "other stuff", DFLAG);
/* Write out b.b. for text line mask and halftone mask components */
boxatm = pixConnComp(pixm3, NULL, 4);
boxahm = pixConnComp(pixhm2, NULL, 8);
if (which == 1) {
boxaWrite("/tmp/lept/livre/textmask.boxa", boxatm);
boxaWrite("/tmp/lept/livre/htmask.boxa", boxahm);
}
pix1 = pixaDisplayTiledAndScaled(pixa, 8, 250, 4, 0, 25, 2);
pixDisplay(pix1, 0, 375 * (which - 1));
snprintf(buf, sizeof(buf), "/tmp/lept/livre/segout.%d.png", which);
pixWrite(buf, pix1, IFF_PNG);
pixDestroy(&pix1);
pixaDestroy(&pixa);
/* clean up to test with valgrind */
pixDestroy(&pixr);
pixDestroy(&pixhs);
pixDestroy(&pixm);
pixDestroy(&pixhm1);
pixDestroy(&pixhm2);
pixDestroy(&pixht);
pixDestroy(&pixi);
pixDestroy(&pixnht);
pixDestroy(&pixvws);
pixDestroy(&pixm1);
pixDestroy(&pixm2);
pixDestroy(&pixm3);
pixDestroy(&pixb1);
pixDestroy(&pixb2);
pixDestroy(&pixnon);
boxaDestroy(&boxatm);
boxaDestroy(&boxahm);
return 0;
}
/*!
* pixGetRegionsBinary()
*
* Input: pixs (1 bpp, assumed to be 300 to 400 ppi)
* &pixhm (<optional return> halftone mask)
* &pixtm (<optional return> textline mask)
* &pixtb (<optional return> textblock mask)
* debug (flag: set to 1 for debug output)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) It is best to deskew the image before segmenting.
* (2) The debug flag enables a number of outputs. These
* are included to show how to generate and save/display
* these results.
*/
l_int32
pixGetRegionsBinary(PIX *pixs,
PIX **ppixhm,
PIX **ppixtm,
PIX **ppixtb,
l_int32 debug)
{
char *tempname;
l_int32 htfound, tlfound;
PIX *pixr, *pixt1, *pixt2;
PIX *pixtext; /* text pixels only */
PIX *pixhm2; /* halftone mask; 2x reduction */
PIX *pixhm; /* halftone mask; */
PIX *pixtm2; /* textline mask; 2x reduction */
PIX *pixtm; /* textline mask */
PIX *pixvws; /* vertical white space mask */
PIX *pixtb2; /* textblock mask; 2x reduction */
PIX *pixtbf2; /* textblock mask; 2x reduction; small comps filtered */
PIX *pixtb; /* textblock mask */
PROCNAME("pixGetRegionsBinary");
if (ppixhm) *ppixhm = NULL;
if (ppixtm) *ppixtm = NULL;
if (ppixtb) *ppixtb = NULL;
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if (pixGetDepth(pixs) != 1)
return ERROR_INT("pixs not 1 bpp", procName, 1);
/* 2x reduce, to 150 -200 ppi */
pixr = pixReduceRankBinaryCascade(pixs, 1, 0, 0, 0);
pixDisplayWrite(pixr, debug);
/* Get the halftone mask */
pixhm2 = pixGenHalftoneMask(pixr, &pixtext, &htfound, debug);
/* Get the textline mask from the text pixels */
pixtm2 = pixGenTextlineMask(pixtext, &pixvws, &tlfound, debug);
/* Get the textblock mask from the textline mask */
pixtb2 = pixGenTextblockMask(pixtm2, pixvws, debug);
pixDestroy(&pixr);
pixDestroy(&pixtext);
pixDestroy(&pixvws);
/* Remove small components from the mask, where a small
* component is defined as one with both width and height < 60 */
pixtbf2 = pixSelectBySize(pixtb2, 60, 60, 4, L_SELECT_IF_EITHER,
L_SELECT_IF_GTE, NULL);
pixDestroy(&pixtb2);
pixDisplayWriteFormat(pixtbf2, debug, IFF_PNG);
/* Expand all masks to full resolution, and do filling or
* small dilations for better coverage. */
pixhm = pixExpandReplicate(pixhm2, 2);
pixt1 = pixSeedfillBinary(NULL, pixhm, pixs, 8);
pixOr(pixhm, pixhm, pixt1);
pixDestroy(&pixt1);
pixDisplayWriteFormat(pixhm, debug, IFF_PNG);
pixt1 = pixExpandReplicate(pixtm2, 2);
pixtm = pixDilateBrick(NULL, pixt1, 3, 3);
pixDestroy(&pixt1);
pixDisplayWriteFormat(pixtm, debug, IFF_PNG);
pixt1 = pixExpandReplicate(pixtbf2, 2);
pixtb = pixDilateBrick(NULL, pixt1, 3, 3);
pixDestroy(&pixt1);
pixDisplayWriteFormat(pixtb, debug, IFF_PNG);
pixDestroy(&pixhm2);
pixDestroy(&pixtm2);
pixDestroy(&pixtbf2);
/* Debug: identify objects that are neither text nor halftone image */
if (debug) {
pixt1 = pixSubtract(NULL, pixs, pixtm); /* remove text pixels */
pixt2 = pixSubtract(NULL, pixt1, pixhm); /* remove halftone pixels */
pixDisplayWriteFormat(pixt2, 1, IFF_PNG);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
}
/* Debug: display textline components with random colors */
if (debug) {
l_int32 w, h;
BOXA *boxa;
PIXA *pixa;
boxa = pixConnComp(pixtm, &pixa, 8);
pixGetDimensions(pixtm, &w, &h, NULL);
pixt1 = pixaDisplayRandomCmap(pixa, w, h);
pixcmapResetColor(pixGetColormap(pixt1), 0, 255, 255, 255);
pixDisplay(pixt1, 100, 100);
pixDisplayWriteFormat(pixt1, 1, IFF_PNG);
pixaDestroy(&pixa);
boxaDestroy(&boxa);
pixDestroy(&pixt1);
}
/* Debug: identify the outlines of each textblock */
if (debug) {
PIXCMAP *cmap;
PTAA *ptaa;
ptaa = pixGetOuterBordersPtaa(pixtb);
tempname = genTempFilename("/tmp", "tb_outlines.ptaa", 0, 0);
ptaaWrite(tempname, ptaa, 1);
FREE(tempname);
pixt1 = pixRenderRandomCmapPtaa(pixtb, ptaa, 1, 16, 1);
cmap = pixGetColormap(pixt1);
pixcmapResetColor(cmap, 0, 130, 130, 130);
pixDisplay(pixt1, 500, 100);
pixDisplayWriteFormat(pixt1, 1, IFF_PNG);
pixDestroy(&pixt1);
ptaaDestroy(&ptaa);
}
/* Debug: get b.b. for all mask components */
if (debug) {
BOXA *bahm, *batm, *batb;
bahm = pixConnComp(pixhm, NULL, 4);
batm = pixConnComp(pixtm, NULL, 4);
batb = pixConnComp(pixtb, NULL, 4);
tempname = genTempFilename("/tmp", "htmask.boxa", 0, 0);
boxaWrite(tempname, bahm);
FREE(tempname);
tempname = genTempFilename("/tmp", "textmask.boxa", 0, 0);
boxaWrite(tempname, batm);
FREE(tempname);
tempname = genTempFilename("/tmp", "textblock.boxa", 0, 0);
boxaWrite(tempname, batb);
FREE(tempname);
boxaDestroy(&bahm);
boxaDestroy(&batm);
boxaDestroy(&batb);
}
if (ppixhm)
*ppixhm = pixhm;
else
pixDestroy(&pixhm);
if (ppixtm)
*ppixtm = pixtm;
else
pixDestroy(&pixtm);
if (ppixtb)
*ppixtb = pixtb;
else
pixDestroy(&pixtb);
return 0;
}