/*!
* pixGenHalftoneMask()
*
* Input: pixs (1 bpp, assumed to be 150 to 200 ppi)
* &pixtext (<optional return> text part of pixs)
* &htfound (<optional return> 1 if the mask is not empty)
* debug (flag: 1 for debug output)
* Return: pixd (halftone mask), or null on error
*/
PIX *
pixGenHalftoneMask(PIX *pixs,
PIX **ppixtext,
l_int32 *phtfound,
l_int32 debug)
{
l_int32 empty;
PIX *pixt1, *pixt2, *pixhs, *pixhm, *pixd;
PROCNAME("pixGenHalftoneMask");
if (ppixtext) *ppixtext = NULL;
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 1)
return (PIX *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
/* Compute seed for halftone parts at 8x reduction */
pixt1 = pixReduceRankBinaryCascade(pixs, 4, 4, 3, 0);
pixt2 = pixOpenBrick(NULL, pixt1, 5, 5);
pixhs = pixExpandReplicate(pixt2, 8); /* back to 2x reduction */
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDisplayWriteFormat(pixhs, debug, IFF_PNG);
/* Compute mask for connected regions */
pixhm = pixCloseSafeBrick(NULL, pixs, 4, 4);
pixDisplayWriteFormat(pixhm, debug, IFF_PNG);
/* Fill seed into mask to get halftone mask */
pixd = pixSeedfillBinary(NULL, pixhs, pixhm, 4);
#if 0
/* Moderate opening to remove thin lines, etc. */
pixOpenBrick(pixd, pixd, 10, 10);
pixDisplayWrite(pixd, debug);
#endif
/* Check if mask is empty */
pixZero(pixd, &empty);
if (phtfound) {
*phtfound = 0;
if (!empty)
*phtfound = 1;
}
/* Optionally, get all pixels that are not under the halftone mask */
if (ppixtext) {
if (empty)
*ppixtext = pixCopy(NULL, pixs);
else
*ppixtext = pixSubtract(NULL, pixs, pixd);
pixDisplayWriteFormat(*ppixtext, debug, IFF_PNG);
}
pixDestroy(&pixhs);
pixDestroy(&pixhm);
return pixd;
}
void
ColorBlend(PIX *pixs,
PIX *pixb,
l_float32 fract)
{
l_int32 i, j, wb, hb, ws, hs, delx, dely, x, y;
pixGetDimensions(pixs, &ws, &hs, NULL);
pixGetDimensions(pixb, &wb, &hb, NULL);
delx = wb + 30;
dely = hb + 25;
x = 200;
y = 300;
for (i = 0; i < 20; i++) {
y = 20 + i * dely;
if (y >= hs - hb)
continue;
for (j = 0; j < 20; j++) {
x = 30 + j * delx;
if (x >= ws - wb)
continue;
pixBlendColor(pixs, pixs, pixb, x, y, fract, 1, 255);
}
}
pixDisplayWriteFormat(pixs, 1, IFF_PNG);
}
main(int argc,
char **argv)
{
l_int32 i;
PIX *pixs, *pixt1, *pixt2, *pixt3, *pixd;
PIXA *pixa;
static char mainName[] = "livre_seedgen";
pixs = pixRead("pageseg2.tif");
startTimer();
for (i = 0; i < 100; i++) {
pixt1 = pixReduceRankBinaryCascade(pixs, 1, 4, 4, 3);
pixDestroy(&pixt1);
}
fprintf(stderr, "Time: %8.4f sec\n", stopTimer() / 100.);
/* 4 2x rank reductions (levels 1, 4, 4, 3), followed by 5x5 opening */
pixDisplayWrite(NULL, -1);
pixDisplayWriteFormat(pixs, 4, IFF_PNG);
pixt1 = pixReduceRankBinaryCascade(pixs, 1, 4, 0, 0);
pixDisplayWriteFormat(pixt1, 1, IFF_PNG);
pixt2 = pixReduceRankBinaryCascade(pixt1, 4, 3, 0, 0);
pixDisplayWriteFormat(pixt2, 1, IFF_PNG);
pixOpenBrick(pixt2, pixt2, 5, 5);
pixt3 = pixExpandBinaryReplicate(pixt2, 2);
pixDisplayWriteFormat(pixt3, 1, IFF_PNG);
/* Generate the output image */
pixa = pixaReadFiles("/tmp", "junk_write_display");
pixd = pixaDisplayTiledAndScaled(pixa, 8, 250, 4, 0, 25, 2);
pixWrite("/tmp/seedgen.png", pixd, IFF_PNG);
pixDestroy(&pixs);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixt3);
pixDestroy(&pixd);
pixaDestroy(&pixa);
return 0;
}
main(int argc,
char **argv)
{
PIX *pixs, *pixsg, *pixg, *pixd;
PIXA *pixa;
static char mainName[] = "livre_tophat";
if (argc != 1)
return ERROR_INT(" Syntax: livre_tophat", mainName, 1);
/* Read the image in at 150 ppi. */
if ((pixs = pixRead("brothers.150.jpg")) == NULL)
return ERROR_INT("pix not made", mainName, 1);
pixDisplayWrite(NULL, -1);
pixDisplayWriteFormat(pixs, 2, IFF_JFIF_JPEG);
pixsg = pixConvertRGBToLuminance(pixs);
/* Black tophat (closing - original-image) and invert */
pixg = pixTophat(pixsg, 15, 15, L_TOPHAT_BLACK);
pixInvert(pixg, pixg);
pixDisplayWriteFormat(pixg, 2, IFF_JFIF_JPEG);
/* Set black point at 200, white point at 245. */
pixd = pixGammaTRC(NULL, pixg, 1.0, 200, 245);
pixDisplayWriteFormat(pixd, 2, IFF_JFIF_JPEG);
pixDestroy(&pixg);
pixDestroy(&pixd);
/* Generate the output image */
pixa = pixaReadFiles("/tmp", "junk_write_display");
pixd = pixaDisplayTiledAndScaled(pixa, 8, 350, 3, 0, 25, 2);
pixWrite("/tmp/tophat.jpg", pixd, IFF_JFIF_JPEG);
pixDisplay(pixd, 0, 0);
pixDestroy(&pixd);
pixDestroy(&pixs);
pixDestroy(&pixsg);
return 0;
}
/*!
* pixGenTextblockMask()
*
* Input: pixs (1 bpp, textline mask, assumed to be 150 to 200 ppi)
* pixvws (vertical white space mask)
* debug (flag: 1 for debug output)
* Return: pixd (textblock mask), or null on error
*
* Notes:
* (1) Both the input masks (textline and vertical white space) and
* the returned textblock mask are at the same resolution.
* (2) The result is somewhat noisy, in that small "blocks" of
* text may be included. These can be removed by post-processing,
* using, e.g.,
* pixSelectBySize(pix, 60, 60, 4, L_SELECT_IF_EITHER,
* L_SELECT_IF_GTE, NULL);
*/
PIX *
pixGenTextblockMask(PIX *pixs,
PIX *pixvws,
l_int32 debug)
{
PIX *pixt1, *pixt2, *pixt3, *pixd;
PROCNAME("pixGenTextblockMask");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (!pixvws)
return (PIX *)ERROR_PTR("pixvws not defined", procName, NULL);
if (pixGetDepth(pixs) != 1)
return (PIX *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
/* Join pixels vertically to make a textblock mask */
pixt1 = pixMorphSequence(pixs, "c1.10 + o4.1", 0);
pixDisplayWriteFormat(pixt1, debug, IFF_PNG);
/* Solidify the textblock mask and remove noise:
* (1) For each cc, 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. */
pixt2 = pixMorphSequenceByComponent(pixt1, "c30.30 + d3.3", 8, 0, 0, NULL);
pixCloseSafeBrick(pixt2, pixt2, 10, 1);
pixDisplayWriteFormat(pixt2, debug, IFF_PNG);
pixt3 = pixSubtract(NULL, pixt2, pixvws);
pixDisplayWriteFormat(pixt3, debug, IFF_PNG);
pixd = pixSelectBySize(pixt3, 25, 5, 8, L_SELECT_IF_BOTH,
L_SELECT_IF_GTE, NULL);
pixDisplayWriteFormat(pixd, debug, IFF_PNG);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixt3);
return pixd;
}
/*!
* pixDisplayWrite()
*
* Input: pix (1, 2, 4, 8, 16, 32 bpp)
* reduction (-1 to reset/erase; 0 to disable;
* otherwise this is a reduction factor)
* Return: 0 if OK; 1 on error
*
* Notes:
* (1) This defaults to jpeg output for pix that are 32 bpp or
* 8 bpp without a colormap. If you want to write all images
* losslessly, use format == IFF_PNG in pixDisplayWriteFormat().
* (2) See pixDisplayWriteFormat() for usage details.
*/
l_int32
pixDisplayWrite(PIX *pixs,
l_int32 reduction)
{
return pixDisplayWriteFormat(pixs, reduction, IFF_JFIF_JPEG);
}
/*!
* \brief pixDisplayWrite()
*
* \param[in] pix 1, 2, 4, 8, 16, 32 bpp
* \param[in] reduction -1 to reset/erase; 0 to disable;
* otherwise this is a reduction factor
* \return 0 if OK; 1 on error
*
* <pre>
* Notes:
* (1) This is a simple interface for writing a set of files that can
* be either looked at individually or saved in a pdf for viewing.
* (2) This defaults to jpeg output for pix that are 32 bpp or
* 8 bpp without a colormap. If you want to write all images
* losslessly, use format == IFF_PNG in pixDisplayWriteFormat().
* (3) See pixDisplayWriteFormat() for usage details.
* </pre>
*/
l_int32
pixDisplayWrite(PIX *pixs,
l_int32 reduction)
{
return pixDisplayWriteFormat(pixs, reduction, IFF_DEFAULT);
}
main(int argc,
char **argv)
{
l_int32 d;
PIX *pixs, *pixc, *pixr, *pixg, *pixb, *pixsg, *pixsm, *pixd;
PIXA *pixa;
static char mainName[] = "livre_adapt";
if (argc != 1)
exit(ERROR_INT(" Syntax: livre_adapt", mainName, 1));
/* Read the image in at 150 ppi. */
pixDisplayWrite(NULL, -1);
if ((pixs = pixRead("brothers.150.jpg")) == NULL)
exit(ERROR_INT("pix not made", mainName, 1));
pixDisplayWriteFormat(pixs, 2, IFF_JFIF_JPEG);
/* Normalize for uneven illumination on RGB image */
pixBackgroundNormRGBArraysMorph(pixs, NULL, 4, 5, 200,
&pixr, &pixg, &pixb);
pixd = pixApplyInvBackgroundRGBMap(pixs, pixr, pixg, pixb, 4, 4);
pixDisplayWriteFormat(pixd, 2, IFF_JFIF_JPEG);
pixDestroy(&pixr);
pixDestroy(&pixg);
pixDestroy(&pixb);
pixDestroy(&pixd);
/* Convert the RGB image to grayscale. */
pixsg = pixConvertRGBToLuminance(pixs);
pixDisplayWriteFormat(pixsg, 2, IFF_JFIF_JPEG);
/* Remove the text in the fg. */
pixc = pixCloseGray(pixsg, 25, 25);
pixDisplayWriteFormat(pixc, 2, IFF_JFIF_JPEG);
/* Smooth the bg with a convolution. */
pixsm = pixBlockconv(pixc, 15, 15);
pixDisplayWriteFormat(pixsm, 2, IFF_JFIF_JPEG);
pixDestroy(&pixc);
/* Normalize for uneven illumination on gray image. */
pixBackgroundNormGrayArrayMorph(pixsg, NULL, 4, 5, 200, &pixg);
pixc = pixApplyInvBackgroundGrayMap(pixsg, pixg, 4, 4);
pixDisplayWriteFormat(pixc, 2, IFF_JFIF_JPEG);
pixDestroy(&pixg);
/* Increase the dynamic range. */
pixd = pixGammaTRC(NULL, pixc, 1.0, 30, 180);
pixDisplayWriteFormat(pixd, 2, IFF_JFIF_JPEG);
pixDestroy(&pixc);
/* Threshold to 1 bpp. */
pixb = pixThresholdToBinary(pixd, 120);
pixDisplayWriteFormat(pixb, 2, IFF_PNG);
pixDestroy(&pixd);
pixDestroy(&pixb);
/* Generate the output image */
pixa = pixaReadFiles("/tmp", "junk_write_display");
pixd = pixaDisplayTiledAndScaled(pixa, 8, 350, 4, 0, 25, 2);
pixWrite("/tmp/adapt.jpg", pixd, IFF_JFIF_JPEG);
pixDisplayWithTitle(pixd, 100, 100, NULL, 1);
pixDestroy(&pixd);
pixDestroy(&pixs);
pixDestroy(&pixsg);
return 0;
}
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 *pixtm1; /* image of closed textlines, 150 ppi */
PIX *pixtm2; /* image of refined text line mask, 150 ppi */
PIX *pixtm3; /* image of refined text line mask, 300 ppi */
PIX *pixtb1; /* image of text block mask, 150 ppi */
PIX *pixtb2; /* image of text block mask, 300 ppi */
PIX *pixnon; /* image of non-text or halftone, 150 ppi */
PIX *pixt1, *pixt2, *pixt3;
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);
pixDisplayWrite(NULL, -1);
/* Reduce to 150 ppi */
pixt1 = pixScaleToGray2(pixs);
pixDisplayWriteFormat(pixt1, L_MAX(ws_flag, L_MAX(ht_flag, block_flag)),
IFF_PNG);
if (which == 1) pixWrite("/tmp/orig.gray.150.png", pixt1, IFF_PNG);
pixDestroy(&pixt1);
pixr = pixReduceRankBinaryCascade(pixs, 1, 0, 0, 0);
/* Get seed for halftone parts */
pixt1 = pixReduceRankBinaryCascade(pixr, 4, 4, 3, 0);
pixt2 = pixOpenBrick(NULL, pixt1, 5, 5);
pixhs = pixExpandBinaryPower2(pixt2, 8);
pixDisplayWriteFormat(pixhs, ht_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/htseed.150.png", pixhs, IFF_PNG);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
/* Get mask for connected regions */
pixm = pixCloseSafeBrick(NULL, pixr, 4, 4);
pixDisplayWriteFormat(pixm, ht_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/ccmask.150.png", pixm, IFF_PNG);
/* Fill seed into mask to get halftone mask */
pixhm1 = pixSeedfillBinary(NULL, pixhs, pixm, 4);
pixDisplayWriteFormat(pixhm1, ht_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/htmask.150.png", pixhm1, IFF_PNG);
pixhm2 = pixExpandBinaryPower2(pixhm1, 2);
/* Extract halftone stuff */
pixht = pixAnd(NULL, pixhm1, pixr);
if (which == 1) pixWrite("/tmp/ht.150.png", pixht, IFF_PNG);
/* Extract non-halftone stuff */
pixnht = pixXor(NULL, pixht, pixr);
pixDisplayWriteFormat(pixnht, text_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/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 (which == 1) pixWrite("/tmp/invert.150.png", pixi, IFF_PNG);
pixDisplayWriteFormat(pixi, ws_flag, 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. */
pixt1 = pixMorphCompSequence(pixi, "o80.60", 0);
pixt2 = pixSubtract(NULL, pixi, pixt1);
pixDisplayWriteFormat(pixt2, ws_flag, IFF_PNG);
pixDestroy(&pixt1);
/* Identify vertical whitespace by opening inverted image */
pixt3 = pixOpenBrick(NULL, pixt2, 5, 1); /* removes thin vertical lines */
pixvws = pixOpenBrick(NULL, pixt3, 1, 200); /* gets long vertical lines */
pixDisplayWriteFormat(pixvws, L_MAX(text_flag, ws_flag), IFF_PNG);
if (which == 1) pixWrite("/tmp/vertws.150.png", pixvws, IFF_PNG);
pixDestroy(&pixt2);
pixDestroy(&pixt3);
/* Get proto (early processed) text line mask. */
/* First close the characters and words in the textlines */
pixtm1 = pixCloseSafeBrick(NULL, pixnht, 30, 1);
pixDisplayWriteFormat(pixtm1, text_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/textmask1.150.png", pixtm1, IFF_PNG);
/* Next open back up the vertical whitespace corridors */
pixtm2 = pixSubtract(NULL, pixtm1, pixvws);
if (which == 1) pixWrite("/tmp/textmask2.150.png", pixtm2, IFF_PNG);
/* Do a small opening to remove noise */
pixOpenBrick(pixtm2, pixtm2, 3, 3);
pixDisplayWriteFormat(pixtm2, text_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/textmask3.150.png", pixtm2, IFF_PNG);
pixtm3 = pixExpandBinaryPower2(pixtm2, 2);
/* Join pixels vertically to make text block mask */
pixtb1 = pixMorphSequence(pixtm2, "c1.10 + o4.1", 0);
pixDisplayWriteFormat(pixtb1, block_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/textblock1.150.png", pixtb1, 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 */
pixt1 = pixMorphSequenceByComponent(pixtb1, "c30.30 + d3.3", 8, 0, 0, NULL);
pixCloseSafeBrick(pixt1, pixt1, 10, 1);
pixDisplayWriteFormat(pixt1, block_flag, IFF_PNG);
pixt2 = pixSubtract(NULL, pixt1, pixvws);
pixt3 = pixSelectBySize(pixt2, 25, 5, 8, L_SELECT_IF_BOTH,
L_SELECT_IF_GTE, NULL);
pixDisplayWriteFormat(pixt3, block_flag, IFF_PNG);
if (which == 1) pixWrite("/tmp/textblock2.150.png", pixt3, IFF_PNG);
pixtb2 = pixExpandBinaryPower2(pixt3, 2);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixt3);
/* Identify the outlines of each textblock */
ptaa = pixGetOuterBordersPtaa(pixtb2);
pixt1 = pixRenderRandomCmapPtaa(pixtb2, ptaa, 1, 8, 1);
cmap = pixGetColormap(pixt1);
pixcmapResetColor(cmap, 0, 130, 130, 130); /* set interior to gray */
if (which == 1) pixWrite("/tmp/textblock3.300.png", pixt1, IFF_PNG);
pixDisplayWithTitle(pixt1, 480, 360, "textblock mask with outlines", DFLAG);
ptaaDestroy(&ptaa);
pixDestroy(&pixt1);
/* Fill line mask (as seed) into the original */
pixt1 = pixSeedfillBinary(NULL, pixtm3, pixs, 8);
pixOr(pixtm3, pixtm3, pixt1);
pixDestroy(&pixt1);
if (which == 1) pixWrite("/tmp/textmask.300.png", pixtm3, IFF_PNG);
pixDisplayWithTitle(pixtm3, 480, 360, "textline mask 4", DFLAG);
/* Fill halftone mask (as seed) into the original */
pixt1 = pixSeedfillBinary(NULL, pixhm2, pixs, 8);
pixOr(pixhm2, pixhm2, pixt1);
pixDestroy(&pixt1);
if (which == 1) pixWrite("/tmp/htmask.300.png", pixhm2, IFF_PNG);
pixDisplayWithTitle(pixhm2, 520, 390, "halftonemask 2", DFLAG);
/* Find objects that are neither text nor halftones */
pixt1 = pixSubtract(NULL, pixs, pixtm3); /* remove text pixels */
pixnon = pixSubtract(NULL, pixt1, pixhm2); /* remove halftone pixels */
if (which == 1) pixWrite("/tmp/other.300.png", pixnon, IFF_PNG);
pixDisplayWithTitle(pixnon, 540, 420, "other stuff", DFLAG);
pixDestroy(&pixt1);
/* Write out b.b. for text line mask and halftone mask components */
boxatm = pixConnComp(pixtm3, NULL, 4);
boxahm = pixConnComp(pixhm2, NULL, 8);
if (which == 1) boxaWrite("/tmp/textmask.boxa", boxatm);
if (which == 1) boxaWrite("/tmp/htmask.boxa", boxahm);
pixa = pixaReadFiles("/tmp/display", "file");
pixt1 = pixaDisplayTiledAndScaled(pixa, 8, 250, 4, 0, 25, 2);
snprintf(buf, sizeof(buf), "/tmp/segout.%d.png", which);
pixWrite(buf, pixt1, IFF_PNG);
pixDestroy(&pixt1);
pixaDestroy(&pixa);
/* clean up to test with valgrind */
pixDestroy(&pixr);
pixDestroy(&pixhs);
pixDestroy(&pixm);
pixDestroy(&pixhm1);
pixDestroy(&pixhm2);
pixDestroy(&pixht);
pixDestroy(&pixnht);
pixDestroy(&pixi);
pixDestroy(&pixvws);
pixDestroy(&pixtm1);
pixDestroy(&pixtm2);
pixDestroy(&pixtm3);
pixDestroy(&pixtb1);
pixDestroy(&pixtb2);
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;
}
/*!
* pixGenTextlineMask()
*
* Input: pixs (1 bpp, assumed to be 150 to 200 ppi)
* &pixvws (<return> vertical whitespace mask)
* &tlfound (<optional return> 1 if the mask is not empty)
* debug (flag: 1 for debug output)
* Return: pixd (textline mask), or null on error
*
* Notes:
* (1) The input pixs should be deskewed.
* (2) pixs should have no halftone pixels.
* (3) Both the input image and the returned textline mask
* are at the same resolution.
*/
PIX *
pixGenTextlineMask(PIX *pixs,
PIX **ppixvws,
l_int32 *ptlfound,
l_int32 debug)
{
l_int32 empty;
PIX *pixt1, *pixt2, *pixvws, *pixd;
PROCNAME("pixGenTextlineMask");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (!ppixvws)
return (PIX *)ERROR_PTR("&pixvws not defined", procName, NULL);
if (pixGetDepth(pixs) != 1)
return (PIX *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
/* First we need a vertical whitespace mask. Invert the image. */
pixt1 = pixInvert(NULL, pixs);
/* The whitespace mask will break textlines where there
* is a large amount of white space below or above.
* This can be prevented by identifying regions of the
* inverted image that have large horizontal extent (bigger than
* the separation between columns) and significant
* vertical extent (bigger than the separation between
* textlines), and subtracting this from the bg. */
pixt2 = pixMorphCompSequence(pixt1, "o80.60", 0);
pixSubtract(pixt1, pixt1, pixt2);
pixDisplayWriteFormat(pixt1, debug, IFF_PNG);
pixDestroy(&pixt2);
/* Identify vertical whitespace by opening the remaining bg.
* o5.1 removes thin vertical bg lines and o1.200 extracts
* long vertical bg lines. */
pixvws = pixMorphCompSequence(pixt1, "o5.1 + o1.200", 0);
*ppixvws = pixvws;
pixDisplayWriteFormat(pixvws, debug, IFF_PNG);
pixDestroy(&pixt1);
/* Three steps to getting text line mask:
* (1) close the characters and words in the textlines
* (2) open the vertical whitespace corridors back up
* (3) small opening to remove noise */
pixt1 = pixCloseSafeBrick(NULL, pixs, 30, 1);
pixDisplayWrite(pixt1, debug);
pixd = pixSubtract(NULL, pixt1, pixvws);
pixOpenBrick(pixd, pixd, 3, 3);
pixDisplayWriteFormat(pixd, debug, IFF_PNG);
pixDestroy(&pixt1);
/* Check if text line mask is empty */
if (ptlfound) {
*ptlfound = 0;
pixZero(pixd, &empty);
if (!empty)
*ptlfound = 1;
}
return pixd;
}
