/*!
* pixTranslate()
*
* Input: pixd (<optional> destination: this can be null,
* equal to pixs, or different from pixs)
* pixs
* hshift (horizontal shift; hshift > 0 is to right)
* vshift (vertical shift; vshift > 0 is down)
* incolor (L_BRING_IN_WHITE, L_BRING_IN_BLACK)
* Return: pixd, or null on error.
*
* Notes:
* (1) The general pattern is:
* pixd = pixTranslate(pixd, pixs, ...);
* For clarity, when you know the case, use one of these:
* pixd = pixTranslate(NULL, pixs, ...); // new
* pixTranslate(pixs, pixs, ...); // in-place
* pixTranslate(pixd, pixs, ...); // to existing pixd
* (2) If an existing pixd is not the same size as pixs, the
* image data will be reallocated.
*/
PIX *
pixTranslate(PIX *pixd,
PIX *pixs,
l_int32 hshift,
l_int32 vshift,
l_int32 incolor)
{
PROCNAME("pixTranslate");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
/* Prepare pixd for in-place operation */
if ((pixd = pixCopy(pixd, pixs)) == NULL)
return (PIX *)ERROR_PTR("pixd not made", procName, NULL);
pixRasteropIP(pixd, hshift, vshift, incolor);
return pixd;
}
/*!
* pixRotateOrth()
*
* Input: pixs (all depths)
* quads (0-3; number of 90 degree cw rotations)
* Return: pixd, or null on error
*/
PIX *
pixRotateOrth(PIX *pixs,
l_int32 quads)
{
PROCNAME("pixRotateOrth");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (quads < 0 || quads > 4)
return (PIX *)ERROR_PTR("quads not in {0,1,2,3,4}", procName, NULL);
if (quads == 0 || quads == 4)
return pixCopy(NULL, pixs);
else if (quads == 1)
return pixRotate90(pixs, 1);
else if (quads == 2)
return pixRotate180(NULL, pixs);
else /* quads == 3 */
return pixRotate90(pixs, -1);
}
main(int argc,
char **argv)
{
l_int32 i, j;
PIX *pixs, *pixt, *pixd;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
pixs = pixRead("test8.jpg");
pixt = pixCopy(NULL, pixs);
/* Copy, in-place and one COLUMN at a time, from the right
side to the left side. */
for (j = 0; j < 200; j++)
pixRasterop(pixs, 20 + j, 20, 1, 250, PIX_SRC, pixs, 250 + j, 20);
pixDisplayWithTitle(pixs, 50, 50, "in-place copy", rp->display);
/* Copy, in-place and one ROW at a time, from the right
side to the left side. */
for (i = 0; i < 250; i++)
pixRasterop(pixt, 20, 20 + i, 200, 1, PIX_SRC, pixt, 250, 20 + i);
/* Test */
regTestComparePix(rp, pixs, pixt); /* 0 */
pixDestroy(&pixs);
pixDestroy(&pixt);
/* Show the mirrored border, which uses the general
pixRasterop() on an image in-place. */
pixs = pixRead("test8.jpg");
pixt = pixRemoveBorder(pixs, 40);
pixd = pixAddMirroredBorder(pixt, 40, 40, 40, 40);
regTestWritePixAndCheck(rp, pixd, IFF_PNG); /* 1 */
pixDisplayWithTitle(pixd, 650, 50, "mirrored border", rp->display);
pixDestroy(&pixs);
pixDestroy(&pixt);
pixDestroy(&pixd);
return regTestCleanup(rp);
}
static void
CopyPtras(L_PTRA *papixs,
L_PTRA *paboxs,
L_PTRA **ppapixd,
L_PTRA **ppaboxd)
{
l_int32 i, imax;
BOX *box;
PIX *pix;
ptraGetMaxIndex(papixs, &imax);
*ppapixd = ptraCreate(imax + 1);
*ppaboxd = ptraCreate(imax + 1);
for (i = 0; i <= imax; i++) {
pix = pixCopy(NULL, (PIX *)ptraGetPtrToItem(papixs, i));
box = boxCopy((BOX *)ptraGetPtrToItem(paboxs, i));
ptraAdd(*ppapixd, pix);
ptraAdd(*ppaboxd, box);
}
return;
}
/*!
* \brief pixSetStrokeWidth()
*
* \param[in] pixs 1 bpp pix
* \param[in] width set stroke width to this value, in [1 ... 100].
* \param[in] thinfirst 1 to thin all pix to a skeleton first; 0 to skip
* \param[in] connectivity 4 or 8, to be used if %thinfirst == 1
* \return pixd with stroke width set to %width, or NULL on error
*
* <pre>
* Notes:
* (1) See notes in pixaSetStrokeWidth().
* (2) A white border of sufficient width to avoid boundary
* artifacts in the thickening step is added before thinning.
* (3) %connectivity == 8 usually gives a slightly smoother result.
* </pre>
*/
PIX *
pixSetStrokeWidth(PIX *pixs,
l_int32 width,
l_int32 thinfirst,
l_int32 connectivity)
{
char buf[16];
l_int32 border;
PIX *pix1, *pix2, *pixd;
PROCNAME("pixSetStrokeWidth");
if (!pixs || (pixGetDepth(pixs) != 1))
return (PIX *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
if (width < 1 || width > 100)
return (PIX *)ERROR_PTR("width not in [1 ... 100]", procName, NULL);
if (connectivity != 4 && connectivity != 8)
return (PIX *)ERROR_PTR("connectivity not 4 or 8", procName, NULL);
if (!thinfirst && width == 1) /* nothing to do */
return pixCopy(NULL, pixs);
/* Add a white border */
border = width / 2;
pix1 = pixAddBorder(pixs, border, 0);
/* Thin to a skeleton */
if (thinfirst)
pix2 = pixThinConnected(pix1, L_THIN_FG, connectivity, 0);
else
pix2 = pixClone(pix1);
pixDestroy(&pix1);
/* Dilate */
snprintf(buf, sizeof(buf), "D%d.%d", width, width);
pixd = pixMorphSequence(pix2, buf, 0);
pixCopyText(pixd, pixs);
pixDestroy(&pix2);
return pixd;
}
/*!
* pixThresholdToValue()
*
* Input: pixd (<optional>; if not null, must be equal to pixs)
* pixs (8, 16, 32 bpp)
* threshval
* setval
* Return: pixd always
*
* Notes:
* - operation can be in-place (pixs == pixd) or to a new pixd
* - if setval > threshval, sets pixels with a value >= threshval to setval
* - if setval < threshval, sets pixels with a value <= threshval to setval
* - if setval == threshval, no-op
*/
PIX *
pixThresholdToValue(PIX *pixd,
PIX *pixs,
l_int32 threshval,
l_int32 setval)
{
l_int32 w, h, d, wpld;
l_uint32 *datad;
PROCNAME("pixThresholdToValue");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, pixd);
d = pixGetDepth(pixs);
if (d != 8 && d != 16 && d != 32)
return (PIX *)ERROR_PTR("pixs not 8, 16 or 32 bpp", procName, pixd);
if (pixd && (pixs != pixd))
return (PIX *)ERROR_PTR("pixd exists and is not pixs", procName, pixd);
if (threshval < 0 || setval < 0)
return (PIX *)ERROR_PTR("threshval & setval not < 0", procName, pixd);
if (d == 8 && setval > 255)
return (PIX *)ERROR_PTR("setval > 255 for 8 bpp", procName, pixd);
if (d == 16 && setval > 0xffff)
return (PIX *)ERROR_PTR("setval > 0xffff for 16 bpp", procName, pixd);
if (!pixd)
pixd = pixCopy(NULL, pixs);
if (setval == threshval) {
L_WARNING("setval == threshval; no operation", procName);
return pixd;
}
datad = pixGetData(pixd);
pixGetDimensions(pixd, &w, &h, NULL);
wpld = pixGetWpl(pixd);
thresholdToValueLow(datad, w, h, d, wpld, threshval, setval);
return pixd;
}
/*!
* pixDilateGray3()
*
* Input: pixs (8 bpp, not cmapped)
* hsize (1 or 3)
* vsize (1 or 3)
* Return: pixd, or null on error
*
* Notes:
* (1) Special case for 1x3, 3x1 or 3x3 brick sel (all hits)
* (2) If hsize = vsize = 1, just returns a copy.
*/
PIX *
pixDilateGray3(PIX *pixs,
l_int32 hsize,
l_int32 vsize)
{
PIX *pixt, *pixb, *pixbd, *pixd;
PROCNAME("pixDilateGray3");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 8)
return (PIX *)ERROR_PTR("pixs not 8 bpp", procName, NULL);
if (pixGetColormap(pixs))
return (PIX *)ERROR_PTR("pix has colormap", procName, NULL);
if ((hsize != 1 && hsize != 3) ||
(vsize != 1 && vsize != 3))
return (PIX *)ERROR_PTR("invalid size: must be 1 or 3", procName, NULL);
if (hsize == 1 && vsize == 1)
return pixCopy(NULL, pixs);
pixb = pixAddBorderGeneral(pixs, 4, 8, 2, 8, 0);
if (vsize == 1)
pixbd = pixDilateGray3h(pixb);
else if (hsize == 1)
pixbd = pixDilateGray3v(pixb);
else { /* vize == hsize == 3 */
pixt = pixDilateGray3h(pixb);
pixbd = pixDilateGray3v(pixt);
pixDestroy(&pixt);
}
pixd = pixRemoveBorderGeneral(pixbd, 4, 8, 2, 8);
pixDestroy(&pixb);
pixDestroy(&pixbd);
return pixd;
}
/*
* Clean dark background action handling
*/
void MainWindow::on_actionCleanDarkBackground_triggered() {
PIX *pixt;
CDBDialog cdb_dialog(this);
cdb_dialog.setValues(blackval, whiteval, thresh);
connect(&cdb_dialog, SIGNAL(cdbParamsChanged(int, int, int)),
this, SLOT(slotCleanDarkBackground(int, int , int)));
if (cdb_dialog.exec() == QDialog::Accepted) {
blackval = cdb_dialog.blackVal->value();
whiteval = cdb_dialog.whiteVal->value();
thresh = cdb_dialog.treshold->value();
pixt = cleanDarkBackground(blackval, whiteval, thresh);
pixs = pixCopy(NULL, pixt);
pixDestroy(&pixt);
setPixToScene();
modified = true;
updateTitle();
this->statusBar()->showMessage(tr("Finished..."), 2000);
} else {
setPixToScene();
}
}
/*!
* pixRankFilterRGB()
*
* Input: pixs (32 bpp)
* wf, hf (width and height of filter; each is >= 1)
* rank (in [0.0 ... 1.0])
* Return: pixd (of rank values), or null on error
*
* Notes:
* (1) This defines, for each pixel in pixs, a neighborhood of
* pixels given by a rectangle "centered" on the pixel.
* This set of wf*hf pixels has a distribution of values.
* For each component, if the values are sorted in increasing
* order, we choose the component such that rank*(wf*hf-1)
* pixels have a lower or equal value and
* (1-rank)*(wf*hf-1) pixels have an equal or greater value.
* (2) Apply gray rank filtering to each component independently.
* (3) See notes in pixRankFilterGray() for further details.
*/
PIX *
pixRankFilterRGB(PIX *pixs,
l_int32 wf,
l_int32 hf,
l_float32 rank)
{
PIX *pixr, *pixg, *pixb, *pixrf, *pixgf, *pixbf, *pixd;
PROCNAME("pixRankFilterRGB");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 32)
return (PIX *)ERROR_PTR("pixs not 32 bpp", procName, NULL);
if (wf < 1 || hf < 1)
return (PIX *)ERROR_PTR("wf < 1 || hf < 1", procName, NULL);
if (rank < 0.0 || rank > 1.0)
return (PIX *)ERROR_PTR("rank must be in [0.0, 1.0]", procName, NULL);
if (wf == 1 && hf == 1) /* no-op */
return pixCopy(NULL, pixs);
pixr = pixGetRGBComponent(pixs, COLOR_RED);
pixg = pixGetRGBComponent(pixs, COLOR_GREEN);
pixb = pixGetRGBComponent(pixs, COLOR_BLUE);
pixrf = pixRankFilterGray(pixr, wf, hf, rank);
pixgf = pixRankFilterGray(pixg, wf, hf, rank);
pixbf = pixRankFilterGray(pixb, wf, hf, rank);
pixd = pixCreateRGBImage(pixrf, pixgf, pixbf);
pixDestroy(&pixr);
pixDestroy(&pixg);
pixDestroy(&pixb);
pixDestroy(&pixrf);
pixDestroy(&pixgf);
pixDestroy(&pixbf);
return pixd;
}
/*!
* pixRotate180()
*
* Input: pixd (<optional>; can be null, equal to pixs,
* or different from pixs)
* pixs (all depths)
* Return: pixd, or null on error
*
* Notes:
* (1) This does a 180 rotation of the image about the center,
* which is equivalent to a left-right flip about a vertical
* line through the image center, followed by a top-bottom
* flip about a horizontal line through the image center.
* (2) There are 3 cases for input:
* (a) pixd == null (creates a new pixd)
* (b) pixd == pixs (in-place operation)
* (c) pixd != pixs (existing pixd)
* (3) For clarity, use these three patterns, respectively:
* (a) pixd = pixRotate180(NULL, pixs);
* (b) pixRotate180(pixs, pixs);
* (c) pixRotate180(pixd, pixs);
*/
PIX *
pixRotate180(PIX *pixd,
PIX *pixs)
{
l_int32 d;
PROCNAME("pixRotate180");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
d = pixGetDepth(pixs);
if (d != 1 && d != 2 && d != 4 && d != 8 && d != 16 && d != 32)
return (PIX *)ERROR_PTR("pixs not in {1,2,4,8,16,32} bpp",
procName, NULL);
/* Prepare pixd for in-place operation */
if ((pixd = pixCopy(pixd, pixs)) == NULL)
return (PIX *)ERROR_PTR("pixd not made", procName, NULL);
pixFlipLR(pixd, pixd);
pixFlipTB(pixd, pixd);
return pixd;
}
// Creates and returns a Pix distorted by various means according to the bool
// flags. If boxes is not nullptr, the boxes are resized/positioned according to
// any spatial distortion and also by the integer reduction factor box_scale
// so they will match what the network will output.
// Returns nullptr on error. The returned Pix must be pixDestroyed.
Pix* PrepareDistortedPix(const Pix* pix, bool perspective, bool invert,
bool white_noise, bool smooth_noise, bool blur,
int box_reduction, TRand* randomizer,
GenericVector<TBOX>* boxes) {
Pix* distorted = pixCopy(nullptr, const_cast<Pix*>(pix));
// Things to do to synthetic training data.
if (invert && randomizer->SignedRand(1.0) < 0)
pixInvert(distorted, distorted);
if ((white_noise || smooth_noise) && randomizer->SignedRand(1.0) > 0.0) {
// TODO(rays) Cook noise in a more thread-safe manner than rand().
// Attempt to make the sequences reproducible.
srand(randomizer->IntRand());
Pix* pixn = pixAddGaussianNoise(distorted, 8.0);
pixDestroy(&distorted);
if (smooth_noise) {
distorted = pixBlockconv(pixn, 1, 1);
pixDestroy(&pixn);
} else {
distorted = pixn;
}
}
if (blur && randomizer->SignedRand(1.0) > 0.0) {
Pix* blurred = pixBlockconv(distorted, 1, 1);
pixDestroy(&distorted);
distorted = blurred;
}
if (perspective)
GeneratePerspectiveDistortion(0, 0, randomizer, &distorted, boxes);
if (boxes != nullptr) {
for (int b = 0; b < boxes->size(); ++b) {
(*boxes)[b].scale(1.0f / box_reduction);
if ((*boxes)[b].width() <= 0)
(*boxes)[b].set_right((*boxes)[b].left() + 1);
}
}
return distorted;
}
main(int argc,
char **argv)
{
l_int32 error;
l_uint32 *data;
PIX *pix1, *pix2, *pix3, *pix1c, *pix2c, *pix1t, *pix2t, *pixd;
PIXA *pixa;
static char mainName[] = "pixmem_reg";
error = 0;
pixa = pixaCreate(0);
/* Copy with internal resizing: onto a cmapped image */
pix1 = pixRead("weasel4.16c.png");
pix2 = pixRead("feyn-fract.tif");
pix3 = pixRead("lucasta.150.jpg");
fprintf(stderr, "before copy 2 --> 3\n");
pixCopy(pix3, pix2);
Compare(pix2, pix3, &error);
pixSaveTiled(pix3, pixa, 4, 1, 30, 32);
fprintf(stderr, "before copy 3 --> 1\n");
pixCopy(pix1, pix3);
Compare(pix2, pix1, &error);
pixSaveTiled(pix1, pixa, 4, 0, 30, 32);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
/* Copy with internal resizing: from a cmapped image */
pix1 = pixRead("weasel4.16c.png");
pix2 = pixRead("feyn-fract.tif");
pix3 = pixRead("lucasta.150.jpg");
fprintf(stderr, "before copy 1 --> 2\n");
pixCopy(pix2, pix1);
Compare(pix2, pix1, &error);
pixSaveTiled(pix2, pixa, 1, 1, 30, 32);
fprintf(stderr, "before copy 2 --> 3\n");
pixCopy(pix3, pix2);
Compare(pix3, pix2, &error);
pixSaveTiled(pix3, pixa, 1, 0, 30, 32);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
/* Transfer of data pixs --> pixd, when pixs is not cloned.
* pixs is destroyed. */
pix1 = pixRead("weasel4.16c.png");
pix2 = pixRead("feyn-fract.tif");
pix3 = pixRead("lucasta.150.jpg");
pix1c = pixCopy(NULL, pix1);
fprintf(stderr, "before transfer 1 --> 2\n");
pixTransferAllData(pix2, &pix1, 0, 0);
Compare(pix2, pix1c, &error);
pixSaveTiled(pix2, pixa, 1, 1, 30, 32);
fprintf(stderr, "before transfer 2 --> 3\n");
pixTransferAllData(pix3, &pix2, 0, 0);
Compare(pix3, pix1c, &error);
pixSaveTiled(pix3, pixa, 1, 0, 30, 32);
pixDestroy(&pix1c);
pixDestroy(&pix3);
/* Another transfer of data pixs --> pixd, when pixs is not cloned.
* pixs is destroyed. */
pix1 = pixRead("weasel4.16c.png");
pix2 = pixRead("feyn-fract.tif");
pix3 = pixRead("lucasta.150.jpg");
pix1c = pixCopy(NULL, pix1);
pix2c = pixCopy(NULL, pix2);
fprintf(stderr, "before copy transfer 1 --> 2\n");
pixTransferAllData(pix2, &pix1c, 0, 0);
Compare(pix2, pix1, &error);
pixSaveTiled(pix2, pixa, 1, 0, 30, 32);
fprintf(stderr, "before copy transfer 2 --> 3\n");
pixTransferAllData(pix3, &pix2, 0, 0);
Compare(pix3, pix1, &error);
pixSaveTiled(pix3, pixa, 1, 0, 30, 32);
pixDestroy(&pix1);
pixDestroy(&pix2c);
pixDestroy(&pix3);
/* Transfer of data pixs --> pixd, when pixs is cloned.
* pixs has its refcount reduced by 1. */
pix1 = pixRead("weasel4.16c.png");
pix2 = pixRead("feyn-fract.tif");
pix3 = pixRead("lucasta.150.jpg");
pix1c = pixClone(pix1);
pix2c = pixClone(pix2);
fprintf(stderr, "before clone transfer 1 --> 2\n");
pixTransferAllData(pix2, &pix1c, 0, 0);
Compare(pix2, pix1, &error);
pixSaveTiled(pix2, pixa, 1, 0, 30, 32);
fprintf(stderr, "before clone transfer 2 --> 3\n");
pixTransferAllData(pix3, &pix2c, 0, 0);
Compare(pix3, pix1, &error);
pixSaveTiled(pix3, pixa, 1, 0, 30, 32);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
/* Extraction of data when pixs is not cloned, putting
* the data into a new template of pixs. */
pix2 = pixRead("feyn-fract.tif");
fprintf(stderr, "no clone: before extraction and reinsertion of 2\n");
pix2c = pixCopy(NULL, pix2); /* for later reference */
data = pixExtractData(pix2);
pix2t = pixCreateTemplateNoInit(pix2);
pixFreeData(pix2t);
pixSetData(pix2t, data);
Compare(pix2c, pix2t, &error);
pixSaveTiled(pix2t, pixa, 4, 1, 30, 32);
pixDestroy(&pix2);
pixDestroy(&pix2c);
pixDestroy(&pix2t);
/* Extraction of data when pixs is cloned, putting
* a copy of the data into a new template of pixs. */
pix1 = pixRead("weasel4.16c.png");
fprintf(stderr, "clone: before extraction and reinsertion of 1\n");
pix1c = pixClone(pix1); /* bump refcount of pix1 to 2 */
data = pixExtractData(pix1); /* should make a copy of data */
pix1t = pixCreateTemplateNoInit(pix1);
pixFreeData(pix1t);
pixSetData(pix1t, data);
Compare(pix1c, pix1t, &error);
pixSaveTiled(pix1t, pixa, 1, 0, 30, 32);
pixDestroy(&pix1);
pixDestroy(&pix1c);
pixDestroy(&pix1t);
pixd = pixaDisplay(pixa, 0, 0);
pixDisplay(pixd, 100, 100);
pixWrite("/tmp/junkpixmem.png", pixd, IFF_PNG);
pixaDestroy(&pixa);
pixDestroy(&pixd);
if (error)
fprintf(stderr, "Fail: an error occurred\n");
else
fprintf(stderr, "Success: no errors\n");
return 0;
}
l_int32 main(int argc,
char **argv)
{
l_int32 irval, igval, ibval;
l_float32 rval, gval, bval, fract, fgfract;
L_BMF *bmf;
BOX *box;
BOXA *boxa;
FPIX *fpix;
PIX *pixs, *pix1, *pix2, *pix3, *pix4, *pix5, *pix6, *pix7;
PIX *pix8, *pix9, *pix10, *pix11, *pix12, *pix13, *pix14, *pix15;
PIXA *pixa;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
pixa = pixaCreate(0);
pixs = pixRead("breviar38.150.jpg");
/* pixs = pixRead("breviar32.150.jpg"); */
pixaAddPix(pixa, pixs, L_CLONE);
regTestWritePixAndCheck(rp, pixs, IFF_JFIF_JPEG); /* 0 */
pixDisplayWithTitle(pixs, 0, 0, "Input image", rp->display);
/* Extract the blue component, which is small in all the text
* regions, including in the highlight color region */
pix1 = pixGetRGBComponent(pixs, COLOR_BLUE);
pixaAddPix(pixa, pix1, L_CLONE);
regTestWritePixAndCheck(rp, pix1, IFF_JFIF_JPEG); /* 1 */
pixDisplayWithTitle(pix1, 200, 0, "Blue component", rp->display);
/* Do a background normalization, with the background set to
* approximately 200 */
pix2 = pixBackgroundNormSimple(pix1, NULL, NULL);
pixaAddPix(pixa, pix2, L_COPY);
regTestWritePixAndCheck(rp, pix2, IFF_JFIF_JPEG); /* 2 */
pixDisplayWithTitle(pix2, 400, 0, "BG normalized to 200", rp->display);
/* Do a linear transform on the gray pixels, with 50 going to
* black and 160 going to white. 50 is sufficiently low to
* make both the red and black print quite dark. Quantize
* to a few equally spaced gray levels. This is the image
* to which highlight color will be applied. */
pixGammaTRC(pix2, pix2, 1.0, 50, 160);
pix3 = pixThresholdOn8bpp(pix2, 7, 1);
pixaAddPix(pixa, pix3, L_CLONE);
regTestWritePixAndCheck(rp, pix3, IFF_JFIF_JPEG); /* 3 */
pixDisplayWithTitle(pix3, 600, 0, "Basic quantized with white bg",
rp->display);
/* Identify the regions of red text. First, make a mask
* consisting of all pixels such that (R-B)/B is larger
* than 2.0. This will have all the red, plus a lot of
* the dark pixels. */
fpix = pixComponentFunction(pixs, 1.0, 0.0, -1.0, 0.0, 0.0, 1.0);
pix4 = fpixThresholdToPix(fpix, 2.0);
pixInvert(pix4, pix4); /* red plus some dark text */
pixaAddPix(pixa, pix4, L_CLONE);
regTestWritePixAndCheck(rp, pix4, IFF_PNG); /* 4 */
pixDisplayWithTitle(pix4, 800, 0, "Red plus dark pixels", rp->display);
/* Make a mask consisting of all the red and background pixels */
pix5 = pixGetRGBComponent(pixs, COLOR_RED);
pix6 = pixThresholdToBinary(pix5, 128);
pixInvert(pix6, pix6); /* red plus background (white) */
/* Intersect the two masks to get a mask consisting of pixels
* that are almost certainly red. This is the seed. */
pix7 = pixAnd(NULL, pix4, pix6); /* red only (seed) */
pixaAddPix(pixa, pix7, L_COPY);
regTestWritePixAndCheck(rp, pix7, IFF_PNG); /* 5 */
pixDisplayWithTitle(pix7, 0, 600, "Seed for red color", rp->display);
/* Make the clipping mask by thresholding the image with
* the background cleaned to white. */
pix8 = pixThresholdToBinary(pix2, 230); /* mask */
pixaAddPix(pixa, pix8, L_CLONE);
regTestWritePixAndCheck(rp, pix8, IFF_PNG); /* 6 */
pixDisplayWithTitle(pix8, 200, 600, "Clipping mask for red components",
rp->display);
/* Fill into the mask from the seed */
pixSeedfillBinary(pix7, pix7, pix8, 8); /* filled: red plus touching */
regTestWritePixAndCheck(rp, pix7, IFF_PNG); /* 7 */
pixDisplayWithTitle(pix7, 400, 600, "Red component mask filled",
rp->display);
/* Remove long horizontal and vertical lines from the filled result */
pix9 = pixMorphSequence(pix7, "o40.1", 0);
pixSubtract(pix7, pix7, pix9); /* remove long horizontal lines */
pixDestroy(&pix9);
pix9 = pixMorphSequence(pix7, "o1.40", 0);
pixSubtract(pix7, pix7, pix9); /* remove long vertical lines */
/* Close the regions to be colored */
pix10 = pixMorphSequence(pix7, "c5.1", 0);
pixaAddPix(pixa, pix10, L_CLONE);
regTestWritePixAndCheck(rp, pix10, IFF_PNG); /* 8 */
pixDisplayWithTitle(pix10, 600, 600,
"Components defining regions allowing coloring",
rp->display);
/* Sanity check on amount to be colored. Only accept images
* with less than 10% of all the pixels with highlight color */
pixForegroundFraction(pix10, &fgfract);
if (fgfract >= 0.10) {
L_INFO("too much highlighting: fract = %6.3f; removing it\n",
rp->testname, fgfract);
pixClearAll(pix10);
pixSetPixel(pix10, 0, 0, 1);
}
/* Get the bounding boxes of the regions to be colored */
boxa = pixConnCompBB(pix10, 8);
/* Get a color to paint that is representative of the
* actual highlight color in the image. Scale each
* color component up from the average by an amount necessary
* to saturate the red. Then divide the green and
* blue components by 2.0. */
pixGetAverageMaskedRGB(pixs, pix7, 0, 0, 1, L_MEAN_ABSVAL,
&rval, &gval, &bval);
fract = 255.0 / rval;
irval = lept_roundftoi(fract * rval);
igval = lept_roundftoi(fract * gval / 2.0);
ibval = lept_roundftoi(fract * bval / 2.0);
fprintf(stderr, "(r,g,b) = (%d,%d,%d)\n", irval, igval, ibval);
/* Color the quantized gray version in the selected regions */
pix11 = pixColorGrayRegions(pix3, boxa, L_PAINT_DARK, 220, irval,
igval, ibval);
pixaAddPix(pixa, pix11, L_CLONE);
regTestWritePixAndCheck(rp, pix11, IFF_PNG); /* 9 */
pixDisplayWithTitle(pix11, 800, 600, "Final colored result", rp->display);
pixaAddPix(pixa, pixs, L_CLONE);
/* Test colorization on gray and cmapped gray */
pix12 = pixColorGrayRegions(pix2, boxa, L_PAINT_DARK, 220, 0, 255, 0);
pixaAddPix(pixa, pix12, L_CLONE);
regTestWritePixAndCheck(rp, pix12, IFF_PNG); /* 10 */
pixDisplayWithTitle(pix12, 900, 600, "Colorizing boxa gray", rp->display);
box = boxCreate(200, 200, 250, 350);
pix13 = pixCopy(NULL, pix2);
pixColorGray(pix13, box, L_PAINT_DARK, 220, 0, 0, 255);
pixaAddPix(pixa, pix13, L_CLONE);
regTestWritePixAndCheck(rp, pix13, IFF_PNG); /* 11 */
pixDisplayWithTitle(pix13, 1000, 600, "Colorizing box gray", rp->display);
pix14 = pixThresholdTo4bpp(pix2, 6, 1);
pix15 = pixColorGrayRegions(pix14, boxa, L_PAINT_DARK, 220, 0, 0, 255);
pixaAddPix(pixa, pix15, L_CLONE);
regTestWritePixAndCheck(rp, pix15, IFF_PNG); /* 12 */
pixDisplayWithTitle(pix15, 1100, 600, "Colorizing boxa cmap", rp->display);
pixColorGrayCmap(pix14, box, L_PAINT_DARK, 0, 255, 255);
pixaAddPix(pixa, pix14, L_CLONE);
regTestWritePixAndCheck(rp, pix14, IFF_PNG); /* 13 */
pixDisplayWithTitle(pix14, 1200, 600, "Colorizing box cmap", rp->display);
boxDestroy(&box);
/* Generate a pdf of the intermediate results */
lept_mkdir("lept");
L_INFO("Writing to /tmp/lept/colorize.pdf\n", rp->testname);
pixaConvertToPdf(pixa, 90, 1.0, 0, 0, "Colorizing highlighted text",
"/tmp/lept/colorize.pdf");
pixaDestroy(&pixa);
fpixDestroy(&fpix);
boxDestroy(&box);
boxaDestroy(&boxa);
pixDestroy(&pixs);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
pixDestroy(&pix4);
pixDestroy(&pix5);
pixDestroy(&pix6);
pixDestroy(&pix7);
pixDestroy(&pix8);
pixDestroy(&pix9);
pixDestroy(&pix10);
pixDestroy(&pix11);
pixDestroy(&pix12);
pixDestroy(&pix13);
pixDestroy(&pix14);
pixDestroy(&pix15);
/* Test the color detector */
pixa = pixaCreate(7);
bmf = bmfCreate("./fonts", 4);
pix1 = TestForRedColor(rp, "brev06.75.jpg", 1, bmf); /* 14 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev10.75.jpg", 0, bmf); /* 15 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev14.75.jpg", 1, bmf); /* 16 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev20.75.jpg", 1, bmf); /* 17 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev36.75.jpg", 0, bmf); /* 18 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev53.75.jpg", 1, bmf); /* 19 */
pixaAddPix(pixa, pix1, L_INSERT);
pix1 = TestForRedColor(rp, "brev56.75.jpg", 1, bmf); /* 20 */
pixaAddPix(pixa, pix1, L_INSERT);
/* Generate a pdf of the color detector results */
L_INFO("Writing to /tmp/lept/colordetect.pdf\n", rp->testname);
pixaConvertToPdf(pixa, 45, 1.0, 0, 0, "Color detection",
"/tmp/lept/colordetect.pdf");
pixaDestroy(&pixa);
bmfDestroy(&bmf);
return regTestCleanup(rp);
}
/*!
* pixGrayMorphSequence()
*
* Input: pixs
* sequence (string specifying sequence)
* dispsep (horizontal separation in pixels between
* successive displays; use zero to suppress display)
* dispy (if dispsep != 0, this gives the y-value of the
* UL corner for display; otherwise it is ignored)
* Return: pixd, or null on error
*
* Notes:
* (1) This works on 8 bpp grayscale images.
* (2) This runs a pipeline of operations; no branching is allowed.
* (3) This only uses brick SELs.
* (4) A new image is always produced; the input image is not changed.
* (5) This contains an interpreter, allowing sequences to be
* generated and run.
* (6) The format of the sequence string is defined below.
* (7) In addition to morphological operations, the composite
* morph/subtract tophat can be performed.
* (8) Sel sizes (width, height) must each be odd numbers.
* (9) Intermediate results can optionally be displayed
* (10) The sequence string is formatted as follows:
* - An arbitrary number of operations, each separated
* by a '+' character. White space is ignored.
* - Each operation begins with a case-independent character
* specifying the operation:
* d or D (dilation)
* e or E (erosion)
* o or O (opening)
* c or C (closing)
* t or T (tophat)
* - The args to the morphological operations are bricks of hits,
* and are formatted as a.b, where a and b are horizontal and
* vertical dimensions, rsp. (each must be an odd number)
* - The args to the tophat are w or W (for white tophat)
* or b or B (for black tophat), followed by a.b as for
* the dilation, erosion, opening and closing.
* Example valid sequences are:
* "c5.3 + o7.5"
* "c9.9 + tw9.9"
*/
PIX *
pixGrayMorphSequence(PIX *pixs,
const char *sequence,
l_int32 dispsep,
l_int32 dispy)
{
char *rawop, *op;
l_int32 nops, i, valid, w, h, x;
PIX *pixt1, *pixt2;
SARRAY *sa;
PROCNAME("pixGrayMorphSequence");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (!sequence)
return (PIX *)ERROR_PTR("sequence not defined", procName, NULL);
/* Split sequence into individual operations */
sa = sarrayCreate(0);
sarraySplitString(sa, sequence, "+");
nops = sarrayGetCount(sa);
/* Verify that the operation sequence is valid */
valid = TRUE;
for (i = 0; i < nops; i++) {
rawop = sarrayGetString(sa, i, 0);
op = stringRemoveChars(rawop, " \n\t");
switch (op[0])
{
case 'd':
case 'D':
case 'e':
case 'E':
case 'o':
case 'O':
case 'c':
case 'C':
if (sscanf(&op[1], "%d.%d", &w, &h) != 2) {
fprintf(stderr, "*** op: %s invalid\n", op);
valid = FALSE;
break;
}
if (w < 1 || (w & 1) == 0 || h < 1 || (h & 1) == 0 ) {
fprintf(stderr,
"*** op: %s; w = %d, h = %d; must both be odd\n",
op, w, h);
valid = FALSE;
break;
}
/* fprintf(stderr, "op = %s; w = %d, h = %d\n", op, w, h); */
break;
case 't':
case 'T':
if (op[1] != 'w' && op[1] != 'W' &&
op[1] != 'b' && op[1] != 'B') {
fprintf(stderr,
"*** op = %s; arg %c must be 'w' or 'b'\n", op, op[1]);
valid = FALSE;
break;
}
sscanf(&op[2], "%d.%d", &w, &h);
if (w < 1 || (w & 1) == 0 || h < 1 || (h & 1) == 0 ) {
fprintf(stderr,
"*** op: %s; w = %d, h = %d; must both be odd\n",
op, w, h);
valid = FALSE;
break;
}
/* fprintf(stderr, "op = %s", op); */
break;
default:
fprintf(stderr, "*** nonexistent op = %s\n", op);
valid = FALSE;
}
FREE(op);
}
if (!valid) {
sarrayDestroy(&sa);
return (PIX *)ERROR_PTR("sequence invalid", procName, NULL);
}
/* Parse and operate */
pixt1 = pixCopy(NULL, pixs);
pixt2 = NULL;
x = 0;
for (i = 0; i < nops; i++) {
rawop = sarrayGetString(sa, i, 0);
op = stringRemoveChars(rawop, " \n\t");
switch (op[0])
{
case 'd':
case 'D':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixDilateGray(pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, dispy);
x += dispsep;
}
break;
case 'e':
case 'E':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixErodeGray(pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, dispy);
x += dispsep;
}
break;
case 'o':
case 'O':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixOpenGray(pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, dispy);
x += dispsep;
}
break;
case 'c':
case 'C':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixCloseGray(pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, dispy);
x += dispsep;
}
break;
case 't':
case 'T':
sscanf(&op[2], "%d.%d", &w, &h);
if (op[1] == 'w' || op[1] == 'W')
pixt2 = pixTophat(pixt1, w, h, L_TOPHAT_WHITE);
else /* 'b' or 'B' */
pixt2 = pixTophat(pixt1, w, h, L_TOPHAT_BLACK);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, dispy);
x += dispsep;
}
break;
default:
/* All invalid ops are caught in the first pass */
break;
}
FREE(op);
}
sarrayDestroy(&sa);
return pixt1;
}
int main(int argc,
char **argv)
{
char *filein, *fileout;
l_int32 i, w, h, liney, linex, same;
l_float32 angle, deg2rad;
PIX *pixt1, *pixt2, *pixs, *pixd;
static char mainName[] = "sheartest";
if (argc != 4)
return ERROR_INT(" Syntax: sheartest filein angle fileout",
mainName, 1);
/* Compare in-place H shear with H shear to a new pix */
pixt1 = pixRead("marge.jpg");
pixGetDimensions(pixt1, &w, &h, NULL);
pixt2 = pixHShear(NULL, pixt1, (l_int32)(0.3 * h), 0.17, L_BRING_IN_WHITE);
pixHShearIP(pixt1, (l_int32)(0.3 * h), 0.17, L_BRING_IN_WHITE);
pixEqual(pixt1, pixt2, &same);
if (same)
fprintf(stderr, "Correct for H shear\n");
else
fprintf(stderr, "Error for H shear\n");
pixDestroy(&pixt1);
pixDestroy(&pixt2);
/* Compare in-place V shear with V shear to a new pix */
pixt1 = pixRead("marge.jpg");
pixGetDimensions(pixt1, &w, &h, NULL);
pixt2 = pixVShear(NULL, pixt1, (l_int32)(0.3 * w), 0.17, L_BRING_IN_WHITE);
pixVShearIP(pixt1, (l_int32)(0.3 * w), 0.17, L_BRING_IN_WHITE);
pixEqual(pixt1, pixt2, &same);
if (same)
fprintf(stderr, "Correct for V shear\n");
else
fprintf(stderr, "Error for V shear\n");
pixDestroy(&pixt1);
pixDestroy(&pixt2);
filein = argv[1];
angle = atof(argv[2]);
fileout = argv[3];
deg2rad = 3.1415926535 / 180.;
if ((pixs = pixRead(filein)) == NULL)
return ERROR_INT("pix not made", mainName, 1);
pixGetDimensions(pixs, &w, &h, NULL);
#if 0
/* Select an operation from this list ...
* ------------------------------------------
pixd = pixHShear(NULL, pixs, liney, deg2rad * angle, L_BRING_IN_WHITE);
pixd = pixVShear(NULL, pixs, linex, deg2rad * angle, L_BRING_IN_WHITE);
pixd = pixHShearCorner(NULL, pixs, deg2rad * angle, L_BRING_IN_WHITE);
pixd = pixVShearCorner(NULL, pixs, deg2rad * angle, L_BRING_IN_WHITE);
pixd = pixHShearCenter(NULL, pixs, deg2rad * angle, L_BRING_IN_WHITE);
pixd = pixVShearCenter(NULL, pixs, deg2rad * angle, L_BRING_IN_WHITE);
pixHShearIP(pixs, liney, deg2rad * angle, L_BRING_IN_WHITE); pixd = pixs;
pixVShearIP(pixs, linex, deg2rad * angle, L_BRING_IN_WHITE); pixd = pixs;
pixRasteropHip(pixs, 0, h/3, -50, L_BRING_IN_WHITE); pixd = pixs;
pixRasteropVip(pixs, 0, w/3, -50, L_BRING_IN_WHITE); pixd = pixs;
* ------------------------------------------
* ... and use it in the following: */
pixd = pixHShear(NULL, pixs, liney, deg2rad * angle, L_BRING_IN_WHITE);
pixWrite(fileout, pixd, IFF_PNG);
pixDisplay(pixd, 50, 50);
pixDestroy(&pixd);
#endif
#if 0
/* Do a horizontal shear about a line */
for (i = 0; i < NTIMES; i++) {
liney = i * h / (NTIMES - 1);
if (liney >= h)
liney = h - 1;
pixd = pixHShear(NULL, pixs, liney, deg2rad * angle, L_BRING_IN_WHITE);
pixDisplay(pixd, 50 + 10 * i, 50 + 10 * i);
pixDestroy(&pixd);
}
#endif
#if 0
/* Do a vertical shear about a line */
for (i = 0; i < NTIMES; i++) {
linex = i * w / (NTIMES - 1);
if (linex >= w)
linex = w - 1;
pixd = pixVShear(NULL, pixs, linex, deg2rad * angle, L_BRING_IN_WHITE);
pixDisplay(pixd, 50 + 10 * i, 50 + 10 * i);
pixDestroy(&pixd);
}
#endif
#if 0
/* Do a horizontal in-place shear about a line */
pixSetPadBits(pixs, 0);
for (i = 0; i < NTIMES; i++) {
pixd = pixCopy(NULL, pixs);
liney = i * h / (NTIMES - 1);
if (liney >= h)
liney = h - 1;
pixHShearIP(pixd, liney, deg2rad * angle, L_BRING_IN_WHITE);
pixDisplay(pixd, 50 + 10 * i, 50 + 10 * i);
pixDestroy(&pixd);
}
#endif
#if 0
/* Do a vertical in-place shear about a line */
for (i = 0; i < NTIMES; i++) {
pixd = pixCopy(NULL, pixs);
linex = i * w / (NTIMES - 1);
if (linex >= w)
linex = w - 1;
pixVShearIP(pixd, linex, deg2rad * angle, L_BRING_IN_WHITE);
pixDisplay(pixd, 50 + 10 * i, 50 + 10 * i);
pixDestroy(&pixd);
}
#endif
pixDestroy(&pixs);
return 0;
}
/*!
* pixThinGeneral()
*
* Input: pixs (1 bpp)
* type (L_THIN_FG, L_THIN_BG)
* sela (of Sels for parallel composite HMTs)
* maxiters (max number of iters allowed; use 0 to iterate
* until completion)
* Return: pixd, or null on error
*
* Notes:
* (1) See notes in pixThin(). That function chooses among
* the best of the Sels for thinning.
* (2) This is a general function that takes a Sela of HMTs
* that are used in parallel for thinning from each
* of four directions. One iteration consists of four
* such parallel thins.
*/
PIX *
pixThinGeneral(PIX *pixs,
l_int32 type,
SELA *sela,
l_int32 maxiters)
{
l_int32 i, j, r, nsels, same;
PIXA *pixahmt;
PIX **pixhmt; /* array owned by pixahmt; do not destroy! */
PIX *pixd, *pixt;
SEL *sel, *selr;
PROCNAME("pixThinGeneral");
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);
if (type != L_THIN_FG && type != L_THIN_BG)
return (PIX *)ERROR_PTR("invalid fg/bg type", procName, NULL);
if (!sela)
return (PIX *)ERROR_PTR("sela not defined", procName, NULL);
if (maxiters == 0) maxiters = 10000;
/* Set up array of temp pix to hold hmts */
nsels = selaGetCount(sela);
pixahmt = pixaCreate(nsels);
for (i = 0; i < nsels; i++) {
pixt = pixCreateTemplate(pixs);
pixaAddPix(pixahmt, pixt, L_INSERT);
}
pixhmt = pixaGetPixArray(pixahmt);
if (!pixhmt)
return (PIX *)ERROR_PTR("pixhmt array not made", procName, NULL);
#if DEBUG_SELS
pixt = selaDisplayInPix(sela, 35, 3, 15, 4);
pixDisplayWithTitle(pixt, 100, 100, "allsels", 1);
pixDestroy(&pixt);
#endif /* DEBUG_SELS */
/* Set up initial image for fg thinning */
if (type == L_THIN_FG)
pixd = pixCopy(NULL, pixs);
else /* bg thinning */
pixd = pixInvert(NULL, pixs);
/* Thin the fg, with up to maxiters iterations */
for (i = 0; i < maxiters; i++) {
pixt = pixCopy(NULL, pixd); /* test for completion */
for (r = 0; r < 4; r++) { /* over 90 degree rotations of Sels */
for (j = 0; j < nsels; j++) { /* over individual sels in sela */
sel = selaGetSel(sela, j); /* not a copy */
selr = selRotateOrth(sel, r);
pixHMT(pixhmt[j], pixd, selr);
selDestroy(&selr);
if (j > 0)
pixOr(pixhmt[0], pixhmt[0], pixhmt[j]); /* accum result */
}
pixSubtract(pixd, pixd, pixhmt[0]); /* remove result */
}
pixEqual(pixd, pixt, &same);
pixDestroy(&pixt);
if (same) {
L_INFO("%d iterations to completion\n", procName, i);
break;
}
}
if (type == L_THIN_BG)
pixInvert(pixd, pixd);
pixaDestroy(&pixahmt);
return pixd;
}
/*!
* pixMorphCompSequenceDwa()
*
* Input: pixs
* sequence (string specifying sequence)
* dispsep (horizontal separation in pixels between
* successive displays; use zero to suppress display)
* Return: pixd, or null on error
*
* Notes:
* (1) This does dwa morphology on binary images, using brick Sels.
* (2) This runs a pipeline of operations; no branching is allowed.
* (3) It implements all brick Sels that have dimensions up to 63
* on each side, using a composite (linear + comb) when useful.
* (4) A new image is always produced; the input image is not changed.
* (5) This contains an interpreter, allowing sequences to be
* generated and run.
* (6) See pixMorphSequence() for further information about usage.
*/
PIX *
pixMorphCompSequenceDwa(PIX *pixs,
const char *sequence,
l_int32 dispsep)
{
char *rawop, *op;
l_int32 nops, i, j, nred, fact, w, h, x, y, border;
l_int32 level[4];
PIX *pixt1, *pixt2;
SARRAY *sa;
PROCNAME("pixMorphCompSequenceDwa");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (!sequence)
return (PIX *)ERROR_PTR("sequence not defined", procName, NULL);
/* Split sequence into individual operations */
sa = sarrayCreate(0);
sarraySplitString(sa, sequence, "+");
nops = sarrayGetCount(sa);
if (!morphSequenceVerify(sa)) {
sarrayDestroy(&sa);
return (PIX *)ERROR_PTR("sequence not valid", procName, NULL);
}
/* Parse and operate */
border = 0;
pixt1 = pixCopy(NULL, pixs);
pixt2 = NULL;
x = y = 0;
for (i = 0; i < nops; i++) {
rawop = sarrayGetString(sa, i, 0);
op = stringRemoveChars(rawop, " \n\t");
switch (op[0])
{
case 'd':
case 'D':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixDilateCompBrickDwa(NULL, pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'e':
case 'E':
sscanf(&op[1], "%d.%d", &w, &h);
pixt2 = pixErodeCompBrickDwa(NULL, pixt1, w, h);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'o':
case 'O':
sscanf(&op[1], "%d.%d", &w, &h);
pixOpenCompBrickDwa(pixt1, pixt1, w, h);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'c':
case 'C':
sscanf(&op[1], "%d.%d", &w, &h);
pixCloseCompBrickDwa(pixt1, pixt1, w, h);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'r':
case 'R':
nred = strlen(op) - 1;
for (j = 0; j < nred; j++)
level[j] = op[j + 1] - '0';
for (j = nred; j < 4; j++)
level[j] = 0;
pixt2 = pixReduceRankBinaryCascade(pixt1, level[0], level[1],
level[2], level[3]);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'x':
case 'X':
sscanf(&op[1], "%d", &fact);
pixt2 = pixExpandReplicate(pixt1, fact);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
case 'b':
case 'B':
sscanf(&op[1], "%d", &border);
pixt2 = pixAddBorder(pixt1, border, 0);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
if (dispsep > 0) {
pixDisplay(pixt1, x, y);
x += dispsep;
}
break;
default:
/* All invalid ops are caught in the first pass */
break;
}
FREE(op);
}
if (border > 0) {
pixt2 = pixRemoveBorder(pixt1, border);
pixDestroy(&pixt1);
pixt1 = pixClone(pixt2);
pixDestroy(&pixt2);
}
sarrayDestroy(&sa);
return pixt1;
}
/*!
* \brief pixConnCompPixa()
*
* \param[in] pixs 1 bpp
* \param[out] ppixa pixa of each c.c.
* \param[in] connectivity 4 or 8
* \return boxa, or NULL on error
*
* <pre>
* Notes:
* (1) This finds bounding boxes of 4- or 8-connected components
* in a binary image, and saves images of each c.c
* in a pixa array.
* (2) It sets up 2 temporary pix, and for each c.c. that is
* located in raster order, it erases the c.c. from one pix,
* then uses the b.b. to extract the c.c. from the two pix using
* an XOR, and finally erases the c.c. from the second pix.
* (3) A clone of the returned boxa (where all boxes in the array
* are clones) is inserted into the pixa.
* (4) If the input is valid, this always returns a boxa and a pixa.
* If pixs is empty, the boxa and pixa will be empty.
* </pre>
*/
BOXA *
pixConnCompPixa(PIX *pixs,
PIXA **ppixa,
l_int32 connectivity)
{
l_int32 h, iszero;
l_int32 x, y, xstart, ystart;
PIX *pix1, *pix2, *pix3, *pix4;
PIXA *pixa;
BOX *box;
BOXA *boxa;
L_STACK *stack, *auxstack;
PROCNAME("pixConnCompPixa");
if (!ppixa)
return (BOXA *)ERROR_PTR("&pixa not defined", procName, NULL);
*ppixa = NULL;
if (!pixs || pixGetDepth(pixs) != 1)
return (BOXA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
if (connectivity != 4 && connectivity != 8)
return (BOXA *)ERROR_PTR("connectivity not 4 or 8", procName, NULL);
boxa = NULL;
pix1 = pix2 = pix3 = pix4 = NULL;
stack = NULL;
pixZero(pixs, &iszero);
if (iszero)
return boxaCreate(1); /* return empty boxa */
pix1 = pixCopy(NULL, pixs);
pix2 = pixCopy(NULL, pixs);
if (!pix1 || !pix2) {
L_ERROR("pix1 or pix2 not made\n", procName);
goto cleanup;
}
h = pixGetHeight(pixs);
if ((stack = lstackCreate(h)) == NULL) {
L_ERROR("stack not made\n", procName);
goto cleanup;
}
auxstack = lstackCreate(0);
stack->auxstack = auxstack;
pixa = pixaCreate(0);
boxa = boxaCreate(0);
xstart = 0;
ystart = 0;
while (1) {
if (!nextOnPixelInRaster(pix1, xstart, ystart, &x, &y))
break;
if ((box = pixSeedfillBB(pix1, stack, x, y, connectivity)) == NULL) {
L_ERROR("box not made\n", procName);
pixaDestroy(&pixa);
boxaDestroy(&boxa);
goto cleanup;
}
boxaAddBox(boxa, box, L_INSERT);
/* Save the c.c. and remove from pix2 as well */
pix3 = pixClipRectangle(pix1, box, NULL);
pix4 = pixClipRectangle(pix2, box, NULL);
pixXor(pix3, pix3, pix4);
pixRasterop(pix2, box->x, box->y, box->w, box->h, PIX_SRC ^ PIX_DST,
pix3, 0, 0);
pixaAddPix(pixa, pix3, L_INSERT);
pixDestroy(&pix4);
xstart = x;
ystart = y;
}
#if DEBUG
pixCountPixels(pix1, &iszero, NULL);
fprintf(stderr, "Number of remaining pixels = %d\n", iszero);
pixWrite("junkremain", pix1, IFF_PNG);
#endif /* DEBUG */
/* Remove old boxa of pixa and replace with a clone copy */
boxaDestroy(&pixa->boxa);
pixa->boxa = boxaCopy(boxa, L_CLONE);
*ppixa = pixa;
/* Cleanup, freeing the fillsegs on each stack */
cleanup:
lstackDestroy(&stack, TRUE);
pixDestroy(&pix1);
pixDestroy(&pix2);
return boxa;
}
// Top-level method to perform splitting based on current settings.
// Returns true if a split was actually performed.
// split_for_pageseg should be true if the splitting is being done prior to
// page segmentation. This mode uses the flag
// pageseg_devanagari_split_strategy to determine the splitting strategy.
bool ShiroRekhaSplitter::Split(bool split_for_pageseg) {
SplitStrategy split_strategy = split_for_pageseg ? pageseg_split_strategy_ :
ocr_split_strategy_;
if (split_strategy == NO_SPLIT) {
return false; // Nothing to do.
}
ASSERT_HOST(split_strategy == MINIMAL_SPLIT ||
split_strategy == MAXIMAL_SPLIT);
ASSERT_HOST(orig_pix_);
if (devanagari_split_debuglevel > 0) {
tprintf("Splitting shiro-rekha ...\n");
tprintf("Split strategy = %s\n",
split_strategy == MINIMAL_SPLIT ? "Minimal" : "Maximal");
tprintf("Initial pageseg available = %s\n",
segmentation_block_list_ ? "yes" : "no");
}
// Create a copy of original image to store the splitting output.
pixDestroy(&splitted_image_);
splitted_image_ = pixCopy(NULL, orig_pix_);
// Initialize debug image if required.
if (devanagari_split_debugimage) {
pixDestroy(&debug_image_);
debug_image_ = pixConvertTo32(orig_pix_);
}
// Determine all connected components in the input image. A close operation
// may be required prior to this, depending on the current settings.
Pix* pix_for_ccs = pixClone(orig_pix_);
if (perform_close_ && global_xheight_ != kUnspecifiedXheight &&
!segmentation_block_list_) {
if (devanagari_split_debuglevel > 0) {
tprintf("Performing a global close operation..\n");
}
// A global measure is available for xheight, but no local information
// exists.
pixDestroy(&pix_for_ccs);
pix_for_ccs = pixCopy(NULL, orig_pix_);
PerformClose(pix_for_ccs, global_xheight_);
}
Pixa* ccs;
Boxa* tmp_boxa = pixConnComp(pix_for_ccs, &ccs, 8);
boxaDestroy(&tmp_boxa);
pixDestroy(&pix_for_ccs);
// Iterate over all connected components. Get their bounding boxes and clip
// out the image regions corresponding to these boxes from the original image.
// Conditionally run splitting on each of them.
Boxa* regions_to_clear = boxaCreate(0);
for (int i = 0; i < pixaGetCount(ccs); ++i) {
Box* box = ccs->boxa->box[i];
Pix* word_pix = pixClipRectangle(orig_pix_, box, NULL);
ASSERT_HOST(word_pix);
int xheight = GetXheightForCC(box);
if (xheight == kUnspecifiedXheight && segmentation_block_list_ &&
devanagari_split_debugimage) {
pixRenderBoxArb(debug_image_, box, 1, 255, 0, 0);
}
// If some xheight measure is available, attempt to pre-eliminate small
// blobs from the shiro-rekha process. This is primarily to save the CCs
// corresponding to punctuation marks/small dots etc which are part of
// larger graphemes.
if (xheight == kUnspecifiedXheight ||
(box->w > xheight / 3 && box->h > xheight / 2)) {
SplitWordShiroRekha(split_strategy, word_pix, xheight,
box->x, box->y, regions_to_clear);
} else if (devanagari_split_debuglevel > 0) {
tprintf("CC dropped from splitting: %d,%d (%d, %d)\n",
box->x, box->y, box->w, box->h);
}
pixDestroy(&word_pix);
}
// Actually clear the boxes now.
for (int i = 0; i < boxaGetCount(regions_to_clear); ++i) {
Box* box = boxaGetBox(regions_to_clear, i, L_CLONE);
pixClearInRect(splitted_image_, box);
boxDestroy(&box);
}
boxaDestroy(®ions_to_clear);
pixaDestroy(&ccs);
if (devanagari_split_debugimage) {
DumpDebugImage(split_for_pageseg ? "pageseg_split_debug.png" :
"ocr_split_debug.png");
}
return true;
}
/*!
* \brief pixWriteMemBmp()
*
* \param[out] pfdata data of bmp formatted image
* \param[out] pfsize size of returned data
* \param[in] pixs 1, 2, 4, 8, 16, 32 bpp
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) 2 bpp bmp files are not valid in the spec, and are
* written as 8 bpp.
* (2) pix with depth <= 8 bpp are written with a colormap.
* 16 bpp gray and 32 bpp rgb pix are written without a colormap.
* (3) The transparency component in an rgb pix is ignored.
* All 32 bpp pix have the bmp alpha component set to 255 (opaque).
* (4) The bmp colormap entries, RGBA_QUAD, are the same as
* the ones used for colormaps in leptonica. This allows
* a simple memcpy for bmp output.
* </pre>
*/
l_int32
pixWriteMemBmp(l_uint8 **pfdata,
size_t *pfsize,
PIX *pixs)
{
l_uint8 pel[4];
l_uint8 *cta = NULL; /* address of the bmp color table array */
l_uint8 *fdata, *data, *fmdata;
l_int32 cmaplen; /* number of bytes in the bmp colormap */
l_int32 ncolors, val, stepsize;
l_int32 w, h, d, fdepth, xres, yres;
l_int32 pixWpl, pixBpl, extrabytes, fBpl, fWpl, i, j, k;
l_int32 heapcm; /* extra copy of cta on the heap ? 1 : 0 */
l_uint32 offbytes, fimagebytes;
l_uint32 *line, *pword;
size_t fsize;
BMP_FH *bmpfh;
BMP_IH *bmpih;
PIX *pix;
PIXCMAP *cmap;
RGBA_QUAD *pquad;
PROCNAME("pixWriteMemBmp");
if (pfdata) *pfdata = NULL;
if (pfsize) *pfsize = 0;
if (!pfdata)
return ERROR_INT("&fdata not defined", procName, 1 );
if (!pfsize)
return ERROR_INT("&fsize not defined", procName, 1 );
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
pixGetDimensions(pixs, &w, &h, &d);
if (d == 2) {
L_WARNING("2 bpp files can't be read; converting to 8 bpp\n", procName);
pix = pixConvert2To8(pixs, 0, 85, 170, 255, 1);
d = 8;
} else {
pix = pixCopy(NULL, pixs);
}
fdepth = (d == 32) ? 24 : d;
/* Resolution is given in pixels/meter */
xres = (l_int32)(39.37 * (l_float32)pixGetXRes(pix) + 0.5);
yres = (l_int32)(39.37 * (l_float32)pixGetYRes(pix) + 0.5);
pixWpl = pixGetWpl(pix);
pixBpl = 4 * pixWpl;
fWpl = (w * fdepth + 31) / 32;
fBpl = 4 * fWpl;
fimagebytes = h * fBpl;
if (fimagebytes > 4LL * L_MAX_ALLOWED_PIXELS) {
pixDestroy(&pix);
return ERROR_INT("image data is too large", procName, 1);
}
/* If not rgb or 16 bpp, the bmp data is required to have a colormap */
heapcm = 0;
if (d == 32 || d == 16) { /* 24 bpp rgb or 16 bpp: no colormap */
ncolors = 0;
cmaplen = 0;
} else if ((cmap = pixGetColormap(pix))) { /* existing colormap */
ncolors = pixcmapGetCount(cmap);
cmaplen = ncolors * sizeof(RGBA_QUAD);
cta = (l_uint8 *)cmap->array;
} else { /* no existing colormap; d <= 8; make a binary or gray one */
if (d == 1) {
cmaplen = sizeof(bwmap);
ncolors = 2;
cta = (l_uint8 *)bwmap;
} else { /* d = 2,4,8; use a grayscale output colormap */
ncolors = 1 << fdepth;
cmaplen = ncolors * sizeof(RGBA_QUAD);
heapcm = 1;
cta = (l_uint8 *)LEPT_CALLOC(cmaplen, 1);
stepsize = 255 / (ncolors - 1);
for (i = 0, val = 0, pquad = (RGBA_QUAD *)cta;
i < ncolors;
i++, val += stepsize, pquad++) {
pquad->blue = pquad->green = pquad->red = val;
pquad->alpha = 255; /* opaque */
}
}
}
#if DEBUG
{l_uint8 *pcmptr;
pcmptr = (l_uint8 *)pixGetColormap(pix)->array;
fprintf(stderr, "Pix colormap[0] = %c%c%c%d\n",
pcmptr[0], pcmptr[1], pcmptr[2], pcmptr[3]);
fprintf(stderr, "Pix colormap[1] = %c%c%c%d\n",
pcmptr[4], pcmptr[5], pcmptr[6], pcmptr[7]);
}
#endif /* DEBUG */
offbytes = BMP_FHBYTES + BMP_IHBYTES + cmaplen;
fsize = offbytes + fimagebytes;
fdata = (l_uint8 *)LEPT_CALLOC(fsize, 1);
*pfdata = fdata;
*pfsize = fsize;
/* Convert to little-endian and write the file header data */
bmpfh = (BMP_FH *)fdata;
bmpfh->bfType = convertOnBigEnd16(BMP_ID);
bmpfh->bfSize = convertOnBigEnd16(fsize & 0x0000ffff);
bmpfh->bfFill1 = convertOnBigEnd16((fsize >> 16) & 0x0000ffff);
bmpfh->bfOffBits = convertOnBigEnd16(offbytes & 0x0000ffff);
bmpfh->bfFill2 = convertOnBigEnd16((offbytes >> 16) & 0x0000ffff);
/* Convert to little-endian and write the info header data */
bmpih = (BMP_IH *)(fdata + BMP_FHBYTES);
bmpih->biSize = convertOnBigEnd32(BMP_IHBYTES);
bmpih->biWidth = convertOnBigEnd32(w);
bmpih->biHeight = convertOnBigEnd32(h);
bmpih->biPlanes = convertOnBigEnd16(1);
bmpih->biBitCount = convertOnBigEnd16(fdepth);
bmpih->biSizeImage = convertOnBigEnd32(fimagebytes);
bmpih->biXPelsPerMeter = convertOnBigEnd32(xres);
bmpih->biYPelsPerMeter = convertOnBigEnd32(yres);
bmpih->biClrUsed = convertOnBigEnd32(ncolors);
bmpih->biClrImportant = convertOnBigEnd32(ncolors);
/* Copy the colormap data and free the cta if necessary */
if (ncolors > 0) {
memcpy(fdata + BMP_FHBYTES + BMP_IHBYTES, cta, cmaplen);
if (heapcm) LEPT_FREE(cta);
}
/* When you write a binary image with a colormap
* that sets BLACK to 0, you must invert the data */
if (fdepth == 1 && cmap && ((l_uint8 *)(cmap->array))[0] == 0x0) {
pixInvert(pix, pix);
}
/* An endian byte swap is also required */
pixEndianByteSwap(pix);
/* Transfer the image data. Image origin for bmp is at lower right. */
fmdata = fdata + offbytes;
if (fdepth != 24) { /* typ 1 or 8 bpp */
data = (l_uint8 *)pixGetData(pix) + pixBpl * (h - 1);
for (i = 0; i < h; i++) {
memcpy(fmdata, data, fBpl);
data -= pixBpl;
fmdata += fBpl;
}
} else { /* 32 bpp pix; 24 bpp file
* See the comments in pixReadStreamBmp() to
* understand the logic behind the pixel ordering below.
* Note that we have again done an endian swap on
* little endian machines before arriving here, so that
* the bytes are ordered on both platforms as:
Red Green Blue --
|-----------|------------|-----------|-----------|
*/
extrabytes = fBpl - 3 * w;
line = pixGetData(pix) + pixWpl * (h - 1);
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
pword = line + j;
pel[2] = *((l_uint8 *)pword + COLOR_RED);
pel[1] = *((l_uint8 *)pword + COLOR_GREEN);
pel[0] = *((l_uint8 *)pword + COLOR_BLUE);
memcpy(fmdata, &pel, 3);
fmdata += 3;
}
if (extrabytes) {
for (k = 0; k < extrabytes; k++) {
memcpy(fmdata, &pel, 1);
fmdata++;
}
}
line -= pixWpl;
}
}
pixDestroy(&pix);
return 0;
}
/*
* pixWriteSegmentedPageToPS()
*
* Input: pixs (all depths; colormap ok)
* pixm (<optional> 1 bpp segmentation mask over image region)
* textscale (scale of text output relative to pixs)
* imagescale (scale of image output relative to pixs)
* threshold (threshold for binarization; typ. 190)
* pageno (page number in set; use 1 for new output file)
* fileout (output ps file)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This generates the PS string for a mixed text/image page,
* and adds it to an existing file if @pageno > 1.
* The PS output is determined by fitting the result to
* a letter-size (8.5 x 11 inch) page.
* (2) The two images (pixs and pixm) are at the same resolution
* (typically 300 ppi). They are used to generate two compressed
* images, pixb and pixc, that are put directly into the output
* PS file.
* (3) pixb is the text component. In the PostScript world, we think of
* it as a mask through which we paint black. It is produced by
* scaling pixs by @textscale, and thresholding to 1 bpp.
* (4) pixc is the image component, which is that part of pixs under
* the mask pixm. It is scaled from pixs by @imagescale.
* (5) Typical values are textscale = 2.0 and imagescale = 0.5.
* (6) If pixm == NULL, the page has only text. If it is all black,
* the page is all image and has no text.
* (7) This can be used to write a multi-page PS file, by using
* sequential page numbers with the same output file. It can
* also be used to write separate PS files for each page,
* by using different output files with @pageno = 0 or 1.
*/
l_int32
pixWriteSegmentedPageToPS(PIX *pixs,
PIX *pixm,
l_float32 textscale,
l_float32 imagescale,
l_int32 threshold,
l_int32 pageno,
const char *fileout)
{
l_int32 alltext, notext, d, ret;
l_uint32 val;
l_float32 scaleratio;
PIX *pixmi, *pixmis, *pixt, *pixg, *pixsc, *pixb, *pixc;
PROCNAME("pixWriteSegmentedPageToPS");
if (!pixs)
return ERROR_INT("pixs not defined", procName, 1);
if (!fileout)
return ERROR_INT("fileout not defined", procName, 1);
if (imagescale <= 0.0 || textscale <= 0.0)
return ERROR_INT("relative scales must be > 0.0", procName, 1);
/* Analyze the page. Determine the ratio by which the
* binary text mask is scaled relative to the image part.
* If there is no image region (alltext == TRUE), the
* text mask will be rendered directly to fit the page,
* and scaleratio = 1.0. */
alltext = TRUE;
notext = FALSE;
scaleratio = 1.0;
if (pixm) {
pixZero(pixm, &alltext); /* pixm empty: all text */
if (alltext)
pixm = NULL; /* treat it as not existing here */
else {
pixmi = pixInvert(NULL, pixm);
pixZero(pixmi, ¬ext); /* pixm full; no text */
pixDestroy(&pixmi);
scaleratio = textscale / imagescale;
}
}
if (pixGetDepth(pixs) == 1) { /* render tiff g4 */
pixb = pixClone(pixs);
pixc = NULL;
}
else {
pixt = pixConvertTo8Or32(pixs, 0, 0); /* this can be a clone of pixs */
/* Get the binary text mask. Note that pixg cannot be a
* clone of pixs, because it may be altered by pixSetMasked(). */
pixb = NULL;
if (notext == FALSE) {
d = pixGetDepth(pixt);
if (d == 8)
pixg = pixCopy(NULL, pixt);
else /* d == 32 */
pixg = pixConvertRGBToLuminance(pixt);
if (pixm) /* clear out the image parts */
pixSetMasked(pixg, pixm, 255);
if (textscale == 1.0)
pixsc = pixClone(pixg);
else if (textscale >= 0.7)
pixsc = pixScaleGrayLI(pixg, textscale, textscale);
else
pixsc = pixScaleAreaMap(pixg, textscale, textscale);
pixb = pixThresholdToBinary(pixsc, threshold);
pixDestroy(&pixg);
pixDestroy(&pixsc);
}
/* Get the scaled image region */
pixc = NULL;
if (pixm) {
if (imagescale == 1.0)
pixsc = pixClone(pixt); /* can possibly be a clone of pixs */
else
pixsc = pixScale(pixt, imagescale, imagescale);
/* If pixm is not full, clear the pixels in pixsc
* corresponding to bg in pixm, where there can be text
* that is written through the mask pixb. Note that
* we could skip this and use pixsc directly in
* pixWriteMixedToPS(); however, clearing these
* non-image regions to a white background will reduce
* the size of pixc (relative to pixsc), and hence
* reduce the size of the PS file that is generated.
* Use a copy so that we don't accidentally alter pixs. */
if (notext == FALSE) {
pixmis = pixScale(pixm, imagescale, imagescale);
pixmi = pixInvert(NULL, pixmis);
val = (d == 8) ? 0xff : 0xffffff00;
pixc = pixCopy(NULL, pixsc);
pixSetMasked(pixc, pixmi, val); /* clear non-image part */
pixDestroy(&pixmis);
pixDestroy(&pixmi);
}
else
pixc = pixClone(pixsc);
pixDestroy(&pixsc);
}
pixDestroy(&pixt);
}
ret = pixWriteMixedToPS(pixb, pixc, scaleratio, pageno, fileout);
pixDestroy(&pixb);
pixDestroy(&pixc);
return ret;
}
/*!
* pixColorMorph()
*
* Input: pixs
* type (L_MORPH_DILATE, L_MORPH_ERODE, L_MORPH_OPEN,
* or L_MORPH_CLOSE)
* hsize (of Sel; must be odd; origin implicitly in center)
* vsize (ditto)
* Return: pixd
*
* Notes:
* (1) This does the morph operation on each component separately,
* and recombines the result.
* (2) Sel is a brick with all elements being hits.
* (3) If hsize = vsize = 1, just returns a copy.
*/
PIX *
pixColorMorph(PIX *pixs,
l_int32 type,
l_int32 hsize,
l_int32 vsize)
{
PIX *pixr, *pixg, *pixb, *pixrm, *pixgm, *pixbm, *pixd;
PROCNAME("pixColorMorph");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 32)
return (PIX *)ERROR_PTR("pixs not 32 bpp", procName, NULL);
if (type != L_MORPH_DILATE && type != L_MORPH_ERODE &&
type != L_MORPH_OPEN && type != L_MORPH_CLOSE)
return (PIX *)ERROR_PTR("invalid morph type", procName, NULL);
if (hsize < 1 || vsize < 1)
return (PIX *)ERROR_PTR("hsize or vsize < 1", procName, NULL);
if ((hsize & 1) == 0 ) {
L_WARNING("horiz sel size must be odd; increasing by 1", procName);
hsize++;
}
if ((vsize & 1) == 0 ) {
L_WARNING("vert sel size must be odd; increasing by 1", procName);
vsize++;
}
if (hsize == 1 && vsize == 1)
return pixCopy(NULL, pixs);
pixr = pixGetRGBComponent(pixs, COLOR_RED);
pixg = pixGetRGBComponent(pixs, COLOR_GREEN);
pixb = pixGetRGBComponent(pixs, COLOR_BLUE);
if (type == L_MORPH_DILATE) {
pixrm = pixDilateGray(pixr, hsize, vsize);
pixgm = pixDilateGray(pixg, hsize, vsize);
pixbm = pixDilateGray(pixb, hsize, vsize);
}
else if (type == L_MORPH_ERODE) {
pixrm = pixErodeGray(pixr, hsize, vsize);
pixgm = pixErodeGray(pixg, hsize, vsize);
pixbm = pixErodeGray(pixb, hsize, vsize);
}
else if (type == L_MORPH_OPEN) {
pixrm = pixOpenGray(pixr, hsize, vsize);
pixgm = pixOpenGray(pixg, hsize, vsize);
pixbm = pixOpenGray(pixb, hsize, vsize);
}
else { /* type == L_MORPH_CLOSE */
pixrm = pixCloseGray(pixr, hsize, vsize);
pixgm = pixCloseGray(pixg, hsize, vsize);
pixbm = pixCloseGray(pixb, hsize, vsize);
}
pixd = pixCreateRGBImage(pixrm, pixgm, pixbm);
pixDestroy(&pixr);
pixDestroy(&pixrm);
pixDestroy(&pixg);
pixDestroy(&pixgm);
pixDestroy(&pixb);
pixDestroy(&pixbm);
return pixd;
}
l_int32 main(int argc,
char **argv) {
l_int32 i, n;
l_float32 a, b, c, d, e;
NUMA *nax, *nafit;
PIX *pixs, *pixn, *pixg, *pixb, *pixt1, *pixt2;
PIXA *pixa;
PTA *pta, *ptad;
PTAA *ptaa1, *ptaa2;
pixs = pixRead("cat-35.jpg");
/* pixs = pixRead("zanotti-78.jpg"); */
/* Normalize for varying background and binarize */
pixn = pixBackgroundNormSimple(pixs, NULL, NULL);
pixg = pixConvertRGBToGray(pixn, 0.5, 0.3, 0.2);
pixb = pixThresholdToBinary(pixg, 130);
pixDestroy(&pixn);
pixDestroy(&pixg);
/* Get the textline centers */
pixa = pixaCreate(6);
ptaa1 = dewarpGetTextlineCenters(pixb, 0);
pixt1 = pixCreateTemplate(pixs);
pixSetAll(pixt1);
pixt2 = pixDisplayPtaa(pixt1, ptaa1);
pixWrite("/tmp/textline1.png", pixt2, IFF_PNG);
pixDisplayWithTitle(pixt2, 0, 100, "textline centers 1", 1);
pixaAddPix(pixa, pixt2, L_INSERT);
pixDestroy(&pixt1);
/* Remove short lines */
fprintf(stderr, "Num all lines = %d\n", ptaaGetCount(ptaa1));
ptaa2 = dewarpRemoveShortLines(pixb, ptaa1, 0.8, 0);
pixt1 = pixCreateTemplate(pixs);
pixSetAll(pixt1);
pixt2 = pixDisplayPtaa(pixt1, ptaa2);
pixWrite("/tmp/textline2.png", pixt2, IFF_PNG);
pixDisplayWithTitle(pixt2, 300, 100, "textline centers 2", 1);
pixaAddPix(pixa, pixt2, L_INSERT);
pixDestroy(&pixt1);
n = ptaaGetCount(ptaa2);
fprintf(stderr, "Num long lines = %d\n", n);
ptaaDestroy(&ptaa1);
pixDestroy(&pixb);
/* Long lines over input image */
pixt1 = pixCopy(NULL, pixs);
pixt2 = pixDisplayPtaa(pixt1, ptaa2);
pixWrite("/tmp/textline3.png", pixt2, IFF_PNG);
pixDisplayWithTitle(pixt2, 600, 100, "textline centers 3", 1);
pixaAddPix(pixa, pixt2, L_INSERT);
pixDestroy(&pixt1);
/* Quadratic fit to curve */
pixt1 = pixCopy(NULL, pixs);
for (i = 0; i < n; i++) {
pta = ptaaGetPta(ptaa2, i, L_CLONE);
ptaGetArrays(pta, &nax, NULL);
ptaGetQuadraticLSF(pta, &a, &b, &c, &nafit);
fprintf(stderr, "Quadratic: a = %10.6f, b = %7.3f, c = %7.3f\n",
a, b, c);
ptad = ptaCreateFromNuma(nax, nafit);
pixDisplayPta(pixt1, pixt1, ptad);
ptaDestroy(&pta);
ptaDestroy(&ptad);
numaDestroy(&nax);
numaDestroy(&nafit);
}
pixWrite("/tmp/textline4.png", pixt1, IFF_PNG);
pixDisplayWithTitle(pixt1, 900, 100, "textline centers 4", 1);
pixaAddPix(pixa, pixt1, L_INSERT);
/* Cubic fit to curve */
pixt1 = pixCopy(NULL, pixs);
for (i = 0; i < n; i++) {
pta = ptaaGetPta(ptaa2, i, L_CLONE);
ptaGetArrays(pta, &nax, NULL);
ptaGetCubicLSF(pta, &a, &b, &c, &d, &nafit);
fprintf(stderr, "Cubic: a = %10.6f, b = %10.6f, c = %7.3f, d = %7.3f\n",
a, b, c, d);
ptad = ptaCreateFromNuma(nax, nafit);
pixDisplayPta(pixt1, pixt1, ptad);
ptaDestroy(&pta);
ptaDestroy(&ptad);
numaDestroy(&nax);
numaDestroy(&nafit);
}
pixWrite("/tmp/textline5.png", pixt1, IFF_PNG);
pixDisplayWithTitle(pixt1, 1200, 100, "textline centers 5", 1);
pixaAddPix(pixa, pixt1, L_INSERT);
/* Quartic fit to curve */
pixt1 = pixCopy(NULL, pixs);
for (i = 0; i < n; i++) {
pta = ptaaGetPta(ptaa2, i, L_CLONE);
ptaGetArrays(pta, &nax, NULL);
ptaGetQuarticLSF(pta, &a, &b, &c, &d, &e, &nafit);
fprintf(stderr,
"Quartic: a = %7.3f, b = %7.3f, c = %9.5f, d = %7.3f, e = %7.3f\n",
a, b, c, d, e);
ptad = ptaCreateFromNuma(nax, nafit);
pixDisplayPta(pixt1, pixt1, ptad);
ptaDestroy(&pta);
ptaDestroy(&ptad);
numaDestroy(&nax);
numaDestroy(&nafit);
}
pixWrite("/tmp/textline6.png", pixt1, IFF_PNG);
pixDisplayWithTitle(pixt1, 1500, 100, "textline centers 6", 1);
pixaAddPix(pixa, pixt1, L_INSERT);
pixaConvertToPdf(pixa, 300, 0.5, L_JPEG_ENCODE, 75,
"LS fittings to textlines", "/tmp/dewarp_fittings.pdf");
pixaDestroy(&pixa);
pixDestroy(&pixs);
ptaaDestroy(&ptaa2);
return 0;
}
int main(int argc,
char **argv)
{
l_int32 i, j;
l_int32 w, h, bw, bh, wpls, rval, gval, bval, same;
l_uint32 pixel;
l_uint32 *lines, *datas;
l_float32 sum1, sum2, ave1, ave2, ave3, ave4, diff1, diff2;
l_float32 var1, var2, var3;
BOX *box1, *box2;
NUMA *na, *na1, *na2, *na3, *na4;
PIX *pix, *pixs, *pix1, *pix2, *pix3, *pix4, *pix5, *pixg, *pixd;
PIXA *pixa;
static char mainName[] = "numa2_reg";
if (argc != 1)
return ERROR_INT(" Syntax: numa2_reg", mainName, 1);
lept_mkdir("lept/numa2");
/* -------------------------------------------------------------------*
* Numa-windowed stats *
* -------------------------------------------------------------------*/
#if DO_ALL
na = numaRead("lyra.5.na");
numaWindowedStats(na, 5, &na1, &na2, &na3, &na4);
gplotSimple1(na, GPLOT_PNG, "/tmp/lept/numa2/lyra6", "Original");
gplotSimple1(na1, GPLOT_PNG, "/tmp/lept/numa2/lyra7", "Mean");
gplotSimple1(na2, GPLOT_PNG, "/tmp/lept/numa2/lyra8", "Mean Square");
gplotSimple1(na3, GPLOT_PNG, "/tmp/lept/numa2/lyra9", "Variance");
gplotSimple1(na4, GPLOT_PNG, "/tmp/lept/numa2/lyra10", "RMS Difference");
pixa = pixaCreate(5);
pix1 = pixRead("/tmp/lept/numa2/lyra6.png");
pix2 = pixRead("/tmp/lept/numa2/lyra7.png");
pix3 = pixRead("/tmp/lept/numa2/lyra8.png");
pix4 = pixRead("/tmp/lept/numa2/lyra9.png");
pix5 = pixRead("/tmp/lept/numa2/lyra10.png");
pixaAddPix(pixa, pix1, L_INSERT);
pixaAddPix(pixa, pix2, L_INSERT);
pixaAddPix(pixa, pix3, L_INSERT);
pixaAddPix(pixa, pix4, L_INSERT);
pixaAddPix(pixa, pix5, L_INSERT);
pixd = pixaDisplayTiledInRows(pixa, 32, 1500, 1.0, 0, 20, 2);
pixDisplay(pixd, 100, 0);
pixWrite("/tmp/lept/numa2/window.png", pixd, IFF_PNG);
pixDestroy(&pixd);
pixaDestroy(&pixa);
numaDestroy(&na);
numaDestroy(&na1);
numaDestroy(&na2);
numaDestroy(&na3);
numaDestroy(&na4);
#endif
/* -------------------------------------------------------------------*
* Extraction on a line *
* -------------------------------------------------------------------*/
#if DO_ALL
/* First, make a pretty image */
w = h = 200;
pixs = pixCreate(w, h, 32);
wpls = pixGetWpl(pixs);
datas = pixGetData(pixs);
for (i = 0; i < 200; i++) {
lines = datas + i * wpls;
for (j = 0; j < 200; j++) {
rval = (l_int32)((255. * j) / w + (255. * i) / h);
gval = (l_int32)((255. * 2 * j) / w + (255. * 2 * i) / h) % 255;
bval = (l_int32)((255. * 4 * j) / w + (255. * 4 * i) / h) % 255;
composeRGBPixel(rval, gval, bval, &pixel);
lines[j] = pixel;
}
}
pixg = pixConvertTo8(pixs, 0); /* and a grayscale version */
pixWrite("/tmp/lept/numa_pixg.png", pixg, IFF_PNG);
pixDisplay(pixg, 450, 100);
na1 = pixExtractOnLine(pixg, 20, 20, 180, 20, 1);
na2 = pixExtractOnLine(pixg, 40, 30, 40, 170, 1);
na3 = pixExtractOnLine(pixg, 20, 170, 180, 30, 1);
na4 = pixExtractOnLine(pixg, 20, 190, 180, 10, 1);
gplotSimple1(na1, GPLOT_PNG, "/tmp/lept/numa2/ext1", "Horizontal");
gplotSimple1(na2, GPLOT_PNG, "/tmp/lept/numa2/ext2", "Vertical");
gplotSimple1(na3, GPLOT_PNG, "/tmp/lept/numa2/ext3",
"Slightly more horizontal than vertical");
gplotSimple1(na4, GPLOT_PNG, "/tmp/lept/numa2/ext4",
"Slightly more vertical than horizontal");
pixa = pixaCreate(4);
pix1 = pixRead("/tmp/lept/numa2/ext1.png");
pix2 = pixRead("/tmp/lept/numa2/ext2.png");
pix3 = pixRead("/tmp/lept/numa2/ext3.png");
pix4 = pixRead("/tmp/lept/numa2/ext4.png");
pixaAddPix(pixa, pix1, L_INSERT);
pixaAddPix(pixa, pix2, L_INSERT);
pixaAddPix(pixa, pix3, L_INSERT);
pixaAddPix(pixa, pix4, L_INSERT);
pixd = pixaDisplayTiledInRows(pixa, 32, 1500, 1.0, 0, 20, 2);
pixDisplay(pixd, 100, 450);
pixWrite("/tmp/lept/numa2/extract.png", pixd, IFF_PNG);
pixDestroy(&pixd);
pixaDestroy(&pixa);
pixDestroy(&pixg);
numaDestroy(&na1);
numaDestroy(&na2);
numaDestroy(&na3);
numaDestroy(&na4);
#endif
/* -------------------------------------------------------------------*
* Row and column pixel sums *
* -------------------------------------------------------------------*/
#if DO_ALL
/* Sum by columns in two halves (left and right) */
pixs = pixRead("test8.jpg");
pixGetDimensions(pixs, &w, &h, NULL);
box1 = boxCreate(0, 0, w / 2, h);
box2 = boxCreate(w / 2, 0, w - 2 / 2, h);
na1 = pixAverageByColumn(pixs, box1, L_BLACK_IS_MAX);
na2 = pixAverageByColumn(pixs, box2, L_BLACK_IS_MAX);
numaJoin(na1, na2, 0, -1);
na3 = pixAverageByColumn(pixs, NULL, L_BLACK_IS_MAX);
numaSimilar(na1, na3, 0.0, &same);
if (same)
fprintf(stderr, "Same for columns\n");
else
fprintf(stderr, "Error for columns\n");
pix = pixConvertTo32(pixs);
pixRenderPlotFromNumaGen(&pix, na3, L_HORIZONTAL_LINE, 3, h / 2, 80, 1,
0xff000000);
pixRenderPlotFromNuma(&pix, na3, L_PLOT_AT_BOT, 3, 80, 0xff000000);
boxDestroy(&box1);
boxDestroy(&box2);
numaDestroy(&na1);
numaDestroy(&na2);
numaDestroy(&na3);
/* Sum by rows in two halves (top and bottom) */
box1 = boxCreate(0, 0, w, h / 2);
box2 = boxCreate(0, h / 2, w, h - h / 2);
na1 = pixAverageByRow(pixs, box1, L_WHITE_IS_MAX);
na2 = pixAverageByRow(pixs, box2, L_WHITE_IS_MAX);
numaJoin(na1, na2, 0, -1);
na3 = pixAverageByRow(pixs, NULL, L_WHITE_IS_MAX);
numaSimilar(na1, na3, 0.0, &same);
if (same)
fprintf(stderr, "Same for rows\n");
else
fprintf(stderr, "Error for rows\n");
pixRenderPlotFromNumaGen(&pix, na3, L_VERTICAL_LINE, 3, w / 2, 80, 1,
0x00ff0000);
pixRenderPlotFromNuma(&pix, na3, L_PLOT_AT_RIGHT, 3, 80, 0x00ff0000);
pixDisplay(pix, 500, 200);
boxDestroy(&box1);
boxDestroy(&box2);
numaDestroy(&na1);
numaDestroy(&na2);
numaDestroy(&na3);
pixDestroy(&pix);
/* Average left by rows; right by columns; compare totals */
box1 = boxCreate(0, 0, w / 2, h);
box2 = boxCreate(w / 2, 0, w - 2 / 2, h);
na1 = pixAverageByRow(pixs, box1, L_WHITE_IS_MAX);
na2 = pixAverageByColumn(pixs, box2, L_WHITE_IS_MAX);
numaGetSum(na1, &sum1); /* sum of averages of left box */
numaGetSum(na2, &sum2); /* sum of averages of right box */
ave1 = sum1 / h;
ave2 = 2.0 * sum2 / w;
ave3 = 0.5 * (ave1 + ave2); /* average over both halves */
fprintf(stderr, "ave1 = %8.4f\n", sum1 / h);
fprintf(stderr, "ave2 = %8.4f\n", 2.0 * sum2 / w);
pixAverageInRect(pixs, NULL, &ave4); /* entire image */
diff1 = ave4 - ave3;
diff2 = w * h * ave4 - (0.5 * w * sum1 + h * sum2);
if (diff1 < 0.001)
fprintf(stderr, "Average diffs are correct\n");
else
fprintf(stderr, "Average diffs are wrong: diff1 = %7.5f\n", diff1);
if (diff2 < 20) /* float-to-integer roundoff */
fprintf(stderr, "Pixel sums are correct\n");
else
fprintf(stderr, "Pixel sums are in error: diff = %7.0f\n", diff2);
/* Variance left and right halves. Variance doesn't average
* in a simple way, unlike pixel sums. */
pixVarianceInRect(pixs, box1, &var1); /* entire image */
pixVarianceInRect(pixs, box2, &var2); /* entire image */
pixVarianceInRect(pixs, NULL, &var3); /* entire image */
fprintf(stderr, "0.5 * (var1 + var2) = %7.3f, var3 = %7.3f\n",
0.5 * (var1 + var2), var3);
boxDestroy(&box1);
boxDestroy(&box2);
numaDestroy(&na1);
numaDestroy(&na2);
#endif
/* -------------------------------------------------------------------*
* Row and column variances *
* -------------------------------------------------------------------*/
#if DO_ALL
/* Display variance by rows and columns */
box1 = boxCreate(415, 0, 130, 425);
boxGetGeometry(box1, NULL, NULL, &bw, &bh);
na1 = pixVarianceByRow(pixs, box1);
na2 = pixVarianceByColumn(pixs, box1);
pix = pixConvertTo32(pixs);
pix1 = pixCopy(NULL, pix);
pixRenderPlotFromNumaGen(&pix, na1, L_VERTICAL_LINE, 3, 415, 100, 1,
0xff000000);
pixRenderPlotFromNumaGen(&pix, na2, L_HORIZONTAL_LINE, 3, bh / 2, 100, 1,
0x00ff0000);
pixRenderPlotFromNuma(&pix1, na1, L_PLOT_AT_LEFT, 3, 60, 0x00ff0000);
pixRenderPlotFromNuma(&pix1, na1, L_PLOT_AT_MID_VERT, 3, 60, 0x0000ff00);
pixRenderPlotFromNuma(&pix1, na1, L_PLOT_AT_RIGHT, 3, 60, 0xff000000);
pixRenderPlotFromNuma(&pix1, na2, L_PLOT_AT_TOP, 3, 60, 0x0000ff00);
pixRenderPlotFromNuma(&pix1, na2, L_PLOT_AT_MID_HORIZ, 3, 60, 0xff000000);
pixRenderPlotFromNuma(&pix1, na2, L_PLOT_AT_BOT, 3, 60, 0x00ff0000);
pixDisplay(pix, 500, 900);
pixDisplay(pix1, 500, 1000);
boxDestroy(&box1);
numaDestroy(&na1);
numaDestroy(&na2);
pixDestroy(&pix);
pixDestroy(&pix1);
pixDestroy(&pixs);
/* Again on a different image */
pix1 = pixRead("boxedpage.jpg");
pix2 = pixConvertTo8(pix1, 0);
pixGetDimensions(pix2, &w, &h, NULL);
na1 = pixVarianceByRow(pix2, NULL);
pix3 = pixConvertTo32(pix1);
pixRenderPlotFromNumaGen(&pix3, na1, L_VERTICAL_LINE, 3, 0, 70, 1,
0xff000000);
na2 = pixVarianceByColumn(pix2, NULL);
pixRenderPlotFromNumaGen(&pix3, na2, L_HORIZONTAL_LINE, 3, bh - 1, 70, 1,
0x00ff0000);
pixDisplay(pix3, 1000, 0);
numaDestroy(&na1);
numaDestroy(&na2);
pixDestroy(&pix3);
/* Again, with an erosion */
pix3 = pixErodeGray(pix2, 3, 21);
pixDisplay(pix3, 1400, 0);
na1 = pixVarianceByRow(pix3, NULL);
pix4 = pixConvertTo32(pix1);
pixRenderPlotFromNumaGen(&pix4, na1, L_VERTICAL_LINE, 3, 30, 70, 1,
0xff000000);
na2 = pixVarianceByColumn(pix3, NULL);
pixRenderPlotFromNumaGen(&pix4, na2, L_HORIZONTAL_LINE, 3, bh - 1, 70, 1,
0x00ff0000);
pixDisplay(pix4, 1000, 550);
numaDestroy(&na1);
numaDestroy(&na2);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
pixDestroy(&pix4);
#endif
/* -------------------------------------------------------------------*
* Windowed variance along a line *
* -------------------------------------------------------------------*/
#if DO_ALL
pix1 = pixRead("boxedpage.jpg");
pix2 = pixConvertTo8(pix1, 0);
pixGetDimensions(pix2, &w, &h, NULL);
pix3 = pixCopy(NULL, pix1);
/* Plot along horizontal line */
pixWindowedVarianceOnLine(pix2, L_HORIZONTAL_LINE, h / 2 - 30, 0,
w, 5, &na1);
pixRenderPlotFromNumaGen(&pix1, na1, L_HORIZONTAL_LINE, 3, h / 2 - 30,
80, 1, 0xff000000);
pixRenderPlotFromNuma(&pix3, na1, L_PLOT_AT_TOP, 3, 60, 0x00ff0000);
pixRenderPlotFromNuma(&pix3, na1, L_PLOT_AT_BOT, 3, 60, 0x0000ff00);
/* Plot along vertical line */
pixWindowedVarianceOnLine(pix2, L_VERTICAL_LINE, 0.78 * w, 0,
h, 5, &na2);
pixRenderPlotFromNumaGen(&pix1, na2, L_VERTICAL_LINE, 3, 0.78 * w, 60,
1, 0x00ff0000);
pixRenderPlotFromNuma(&pix3, na2, L_PLOT_AT_LEFT, 3, 60, 0xff000000);
pixRenderPlotFromNuma(&pix3, na2, L_PLOT_AT_RIGHT, 3, 60, 0x00ff0000);
pixDisplay(pix1, 1000, 1000);
pixDisplay(pix3, 1500, 1000);
pixDestroy(&pix1);
pixDestroy(&pix2);
pixDestroy(&pix3);
numaDestroy(&na1);
numaDestroy(&na2);
#endif
return 0;
}
main(int argc,
char **argv)
{
char bufname[256];
l_int32 i, j, w, h, d, x, y, wpls;
l_uint32 *datas, *lines;
l_float32 *vc;
l_float32 *mat1, *mat2, *mat3, *mat1i, *mat2i, *mat3i, *matdinv;
l_float32 matd[9], matdi[9];
BOXA *boxa, *boxa2;
PIX *pix, *pixs, *pixb, *pixg, *pixc, *pixcs;
PIX *pixd, *pixt1, *pixt2, *pixt3;
PIXA *pixa;
PTA *ptas, *ptad;
static char mainName[] = "affine_reg";
if (argc != 1)
exit(ERROR_INT(" Syntax: affine_reg", mainName, 1));
if ((pixs = pixRead("feyn.tif")) == NULL)
exit(ERROR_INT("pixs not made", mainName, 1));
#if ALL
/* Test invertability of sequential. */
pixa = pixaCreate(0);
for (i = 0; i < 3; i++) {
pixb = pixAddBorder(pixs, ADDED_BORDER_PIXELS, 0);
MakePtas(i, &ptas, &ptad);
pixt1 = pixAffineSequential(pixb, ptad, ptas, 0, 0);
pixSaveTiled(pixt1, pixa, 3, 1, 20, 8);
pixt2 = pixAffineSequential(pixt1, ptas, ptad, 0, 0);
pixSaveTiled(pixt2, pixa, 3, 0, 20, 0);
pixd = pixRemoveBorder(pixt2, ADDED_BORDER_PIXELS);
pixXor(pixd, pixd, pixs);
pixSaveTiled(pixd, pixa, 3, 0, 20, 0);
sprintf(bufname, "/tmp/junkseq%d.png", i);
pixWrite(bufname, pixd, IFF_PNG);
pixDestroy(&pixb);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixd);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
}
pixt1 = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine1.png", pixt1, IFF_PNG);
pixDisplay(pixt1, 100, 100);
pixDestroy(&pixt1);
pixaDestroy(&pixa);
#endif
#if ALL
/* Test invertability of sampling */
pixa = pixaCreate(0);
for (i = 0; i < 3; i++) {
pixb = pixAddBorder(pixs, ADDED_BORDER_PIXELS, 0);
MakePtas(i, &ptas, &ptad);
pixt1 = pixAffineSampledPta(pixb, ptad, ptas, L_BRING_IN_WHITE);
pixSaveTiled(pixt1, pixa, 3, 1, 20, 8);
pixt2 = pixAffineSampledPta(pixt1, ptas, ptad, L_BRING_IN_WHITE);
pixSaveTiled(pixt2, pixa, 3, 0, 20, 0);
pixd = pixRemoveBorder(pixt2, ADDED_BORDER_PIXELS);
pixXor(pixd, pixd, pixs);
pixSaveTiled(pixd, pixa, 3, 0, 20, 0);
if (i == 0) pixWrite("/tmp/junksamp.png", pixt1, IFF_PNG);
pixDestroy(&pixb);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixd);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
}
pixt1 = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine2.png", pixt1, IFF_PNG);
pixDisplay(pixt1, 100, 300);
pixDestroy(&pixt1);
pixaDestroy(&pixa);
#endif
#if ALL
/* Test invertability of interpolation on grayscale */
pixa = pixaCreate(0);
pixg = pixScaleToGray3(pixs);
for (i = 0; i < 3; i++) {
pixb = pixAddBorder(pixg, ADDED_BORDER_PIXELS / 3, 255);
MakePtas(i, &ptas, &ptad);
pixt1 = pixAffinePta(pixb, ptad, ptas, L_BRING_IN_WHITE);
pixSaveTiled(pixt1, pixa, 1, 1, 20, 8);
pixt2 = pixAffinePta(pixt1, ptas, ptad, L_BRING_IN_WHITE);
pixSaveTiled(pixt2, pixa, 1, 0, 20, 0);
pixd = pixRemoveBorder(pixt2, ADDED_BORDER_PIXELS / 3);
pixXor(pixd, pixd, pixg);
pixSaveTiled(pixd, pixa, 1, 0, 20, 0);
if (i == 0) pixWrite("/tmp/junkinterp.png", pixt1, IFF_PNG);
pixDestroy(&pixb);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixd);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
}
pixt1 = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine3.png", pixt1, IFF_PNG);
pixDisplay(pixt1, 100, 500);
pixDestroy(&pixt1);
pixaDestroy(&pixa);
pixDestroy(&pixg);
#endif
#if ALL
/* Test invertability of interpolation on color */
pixa = pixaCreate(0);
pixc = pixRead("test24.jpg");
pixcs = pixScale(pixc, 0.3, 0.3);
for (i = 0; i < 3; i++) {
pixb = pixAddBorder(pixcs, ADDED_BORDER_PIXELS / 4, 0xffffff00);
MakePtas(i, &ptas, &ptad);
pixt1 = pixAffinePta(pixb, ptad, ptas, L_BRING_IN_WHITE);
pixSaveTiled(pixt1, pixa, 1, 1, 20, 32);
pixt2 = pixAffinePta(pixt1, ptas, ptad, L_BRING_IN_WHITE);
pixSaveTiled(pixt2, pixa, 1, 0, 20, 0);
pixd = pixRemoveBorder(pixt2, ADDED_BORDER_PIXELS / 4);
pixXor(pixd, pixd, pixcs);
pixSaveTiled(pixd, pixa, 1, 0, 20, 0);
pixDestroy(&pixb);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixd);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
}
pixt1 = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine4.png", pixt1, IFF_PNG);
pixDisplay(pixt1, 100, 500);
pixDestroy(&pixt1);
pixaDestroy(&pixa);
pixDestroy(&pixc);
pixDestroy(&pixcs);
#endif
#if ALL
/* Comparison between sequential and sampling */
MakePtas(3, &ptas, &ptad);
pixa = pixaCreate(0);
/* Use sequential transforms */
pixt1 = pixAffineSequential(pixs, ptas, ptad,
ADDED_BORDER_PIXELS, ADDED_BORDER_PIXELS);
pixSaveTiled(pixt1, pixa, 2, 0, 20, 8);
/* Use sampled transform */
pixt2 = pixAffineSampledPta(pixs, ptas, ptad, L_BRING_IN_WHITE);
pixSaveTiled(pixt2, pixa, 2, 0, 20, 8);
/* Compare the results */
pixXor(pixt2, pixt2, pixt1);
pixSaveTiled(pixt2, pixa, 2, 0, 20, 8);
pixd = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine5.png", pixd, IFF_PNG);
pixDisplay(pixd, 100, 700);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixd);
pixaDestroy(&pixa);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
#endif
#if ALL
/* Get timings and test with large distortion */
MakePtas(4, &ptas, &ptad);
pixa = pixaCreate(0);
pixg = pixScaleToGray3(pixs);
startTimer();
pixt1 = pixAffineSequential(pixg, ptas, ptad, 0, 0);
fprintf(stderr, " Time for pixAffineSequentialPta(): %6.2f sec\n",
stopTimer());
pixSaveTiled(pixt1, pixa, 1, 1, 20, 8);
startTimer();
pixt2 = pixAffineSampledPta(pixg, ptas, ptad, L_BRING_IN_WHITE);
fprintf(stderr, " Time for pixAffineSampledPta(): %6.2f sec\n", stopTimer());
pixSaveTiled(pixt2, pixa, 1, 0, 20, 8);
startTimer();
pixt3 = pixAffinePta(pixg, ptas, ptad, L_BRING_IN_WHITE);
fprintf(stderr, " Time for pixAffinePta(): %6.2f sec\n", stopTimer());
pixSaveTiled(pixt3, pixa, 1, 0, 20, 8);
pixXor(pixt1, pixt1, pixt2);
pixSaveTiled(pixt1, pixa, 1, 1, 20, 8);
pixXor(pixt2, pixt2, pixt3);
pixSaveTiled(pixt2, pixa, 1, 0, 20, 8);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
pixDestroy(&pixt3);
pixd = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine6.png", pixd, IFF_PNG);
pixDisplay(pixd, 100, 900);
pixDestroy(&pixd);
pixDestroy(&pixg);
pixaDestroy(&pixa);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
#endif
pixDestroy(&pixs);
#if 1
/* Set up pix and boxa */
pixa = pixaCreate(0);
pix = pixRead("lucasta.1.300.tif");
pixTranslate(pix, pix, 70, 0, L_BRING_IN_WHITE);
pixt1 = pixCloseBrick(NULL, pix, 14, 5);
pixOpenBrick(pixt1, pixt1, 1, 2);
boxa = pixConnComp(pixt1, NULL, 8);
pixs = pixConvertTo32(pix);
pixGetDimensions(pixs, &w, &h, NULL);
pixc = pixCopy(NULL, pixs);
RenderHashedBoxa(pixc, boxa, 113);
pixSaveTiled(pixc, pixa, 2, 1, 30, 32);
pixDestroy(&pix);
pixDestroy(&pixc);
pixDestroy(&pixt1);
/* Set up an affine transform in matd, and apply it to boxa */
mat1 = createMatrix2dTranslate(SHIFTX, SHIFTY);
mat2 = createMatrix2dScale(SCALEX, SCALEY);
mat3 = createMatrix2dRotate(w / 2, h / 2, ROTATION);
l_productMat3(mat3, mat2, mat1, matd, 3);
boxa2 = boxaAffineTransform(boxa, matd);
/* Set up the inverse transform in matdi */
mat1i = createMatrix2dTranslate(-SHIFTX, -SHIFTY);
mat2i = createMatrix2dScale(1.0/ SCALEX, 1.0 / SCALEY);
mat3i = createMatrix2dRotate(w / 2, h / 2, -ROTATION);
l_productMat3(mat1i, mat2i, mat3i, matdi, 3);
/* Invert the original affine transform in matdinv */
affineInvertXform(matd, &matdinv);
fprintf(stderr, "Affine transform, applied to boxa\n");
for (i = 0; i < 9; i++) {
if (i && (i % 3 == 0)) fprintf(stderr, "\n");
fprintf(stderr, " %7.3f ", matd[i]);
}
fprintf(stderr, "\nInverse transform, made by composing inverse parts");
for (i = 0; i < 9; i++) {
if (i % 3 == 0) fprintf(stderr, "\n");
fprintf(stderr, " %7.3f ", matdi[i]);
}
fprintf(stderr, "\nInverse transform, made by inverting the affine xform");
for (i = 0; i < 6; i++) {
if (i % 3 == 0) fprintf(stderr, "\n");
fprintf(stderr, " %7.3f ", matdinv[i]);
}
fprintf(stderr, "\n");
/* Apply the inverted affine transform pixs */
pixd = pixAffine(pixs, matdinv, L_BRING_IN_WHITE);
RenderHashedBoxa(pixd, boxa2, 513);
pixSaveTiled(pixd, pixa, 2, 0, 30, 32);
pixDestroy(&pixd);
pixd = pixaDisplay(pixa, 0, 0);
pixWrite("/tmp/junkaffine7.png", pixd, IFF_PNG);
pixDisplay(pixd, 100, 900);
pixDestroy(&pixd);
pixDestroy(&pixs);
pixaDestroy(&pixa);
boxaDestroy(&boxa);
boxaDestroy(&boxa2);
FREE(mat1);
FREE(mat2);
FREE(mat3);
FREE(mat1i);
FREE(mat2i);
FREE(mat3i);
#endif
return 0;
}
/*!
* \brief pixConnCompBB()
*
* \param[in] pixs 1 bpp
* \param[in] connectivity 4 or 8
* \return boxa, or NULL on error
*
* <pre>
* Notes:
* (1) Finds bounding boxes of 4- or 8-connected components
* in a binary image.
* (2) This works on a copy of the input pix. The c.c. are located
* in raster order and erased one at a time. In the process,
* the b.b. is computed and saved.
* </pre>
*/
BOXA *
pixConnCompBB(PIX *pixs,
l_int32 connectivity)
{
l_int32 h, iszero;
l_int32 x, y, xstart, ystart;
PIX *pixt;
BOX *box;
BOXA *boxa;
L_STACK *stack, *auxstack;
PROCNAME("pixConnCompBB");
if (!pixs || pixGetDepth(pixs) != 1)
return (BOXA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
if (connectivity != 4 && connectivity != 8)
return (BOXA *)ERROR_PTR("connectivity not 4 or 8", procName, NULL);
boxa = NULL;
pixt = NULL;
stack = NULL;
pixZero(pixs, &iszero);
if (iszero)
return boxaCreate(1); /* return empty boxa */
if ((pixt = pixCopy(NULL, pixs)) == NULL)
return (BOXA *)ERROR_PTR("pixt not made", procName, NULL);
h = pixGetHeight(pixs);
if ((stack = lstackCreate(h)) == NULL) {
L_ERROR("stack not made\n", procName);
goto cleanup;
}
auxstack = lstackCreate(0);
stack->auxstack = auxstack;
boxa = boxaCreate(0);
xstart = 0;
ystart = 0;
while (1) {
if (!nextOnPixelInRaster(pixt, xstart, ystart, &x, &y))
break;
if ((box = pixSeedfillBB(pixt, stack, x, y, connectivity)) == NULL) {
L_ERROR("box not made\n", procName);
boxaDestroy(&boxa);
goto cleanup;
}
boxaAddBox(boxa, box, L_INSERT);
xstart = x;
ystart = y;
}
#if DEBUG
pixCountPixels(pixt, &iszero, NULL);
fprintf(stderr, "Number of remaining pixels = %d\n", iszero);
pixWrite("junkremain", pixt1, IFF_PNG);
#endif /* DEBUG */
/* Cleanup, freeing the fillsegs on each stack */
cleanup:
lstackDestroy(&stack, TRUE);
pixDestroy(&pixt);
return boxa;
}
/*!
* pixSelectBySize()
*
* Input: pixs (1 bpp)
* width, height (threshold dimensions)
* connectivity (4 or 8)
* type (L_SELECT_WIDTH, L_SELECT_HEIGHT,
* L_SELECT_IF_EITHER, L_SELECT_IF_BOTH)
* relation (L_SELECT_IF_LT, L_SELECT_IF_GT,
* L_SELECT_IF_LTE, L_SELECT_IF_GTE)
* &changed (<optional return> 1 if changed; 0 otherwise)
* Return: filtered pixd, or null on error
*
* Notes:
* (1) The args specify constraints on the size of the
* components that are kept.
* (2) If unchanged, returns a copy of pixs. Otherwise,
* returns a new pix with the filtered components.
* (3) If the selection type is L_SELECT_WIDTH, the input
* height is ignored, and v.v.
* (4) To keep small components, use relation = L_SELECT_IF_LT or
* L_SELECT_IF_LTE.
* To keep large components, use relation = L_SELECT_IF_GT or
* L_SELECT_IF_GTE.
*/
PIX *
pixSelectBySize(PIX *pixs,
l_int32 width,
l_int32 height,
l_int32 connectivity,
l_int32 type,
l_int32 relation,
l_int32 *pchanged)
{
l_int32 w, h, empty, changed, count;
BOXA *boxa;
PIX *pixd;
PIXA *pixas, *pixad;
PROCNAME("pixSelectBySize");
if (!pixs)
return (PIX *)ERROR_PTR("pixs not defined", procName, NULL);
if (connectivity != 4 && connectivity != 8)
return (PIX *)ERROR_PTR("connectivity not 4 or 8", procName, NULL);
if (type != L_SELECT_WIDTH && type != L_SELECT_HEIGHT &&
type != L_SELECT_IF_EITHER && type != L_SELECT_IF_BOTH)
return (PIX *)ERROR_PTR("invalid type", procName, NULL);
if (relation != L_SELECT_IF_LT && relation != L_SELECT_IF_GT &&
relation != L_SELECT_IF_LTE && relation != L_SELECT_IF_GTE)
return (PIX *)ERROR_PTR("invalid relation", procName, NULL);
if (pchanged) *pchanged = FALSE;
/* Check if any components exist */
pixZero(pixs, &empty);
if (empty)
return pixCopy(NULL, pixs);
/* Identify and select the components */
boxa = pixConnComp(pixs, &pixas, connectivity);
pixad = pixaSelectBySize(pixas, width, height, type, relation, &changed);
boxaDestroy(&boxa);
pixaDestroy(&pixas);
/* Render the result */
if (!changed) {
pixaDestroy(&pixad);
return pixCopy(NULL, pixs);
}
else {
if (pchanged) *pchanged = TRUE;
pixGetDimensions(pixs, &w, &h, NULL);
count = pixaGetCount(pixad);
if (count == 0) /* return empty pix */
pixd = pixCreateTemplate(pixs);
else {
pixd = pixaDisplay(pixad, w, h);
pixCopyResolution(pixd, pixs);
pixCopyColormap(pixd, pixs);
pixCopyText(pixd, pixs);
pixCopyInputFormat(pixd, pixs);
}
pixaDestroy(&pixad);
return pixd;
}
}
// Returns a list of regions (boxes) which should be cleared in the original
// image so as to perform shiro-rekha splitting. Pix is assumed to carry one
// (or less) word only. Xheight measure could be the global estimate, the row
// estimate, or unspecified. If unspecified, over splitting may occur, since a
// conservative estimate of stroke width along with an associated multiplier
// is used in its place. It is advisable to have a specified xheight when
// splitting for classification/training.
// A vertical projection histogram of all the on-pixels in the input pix is
// computed. The maxima of this histogram is regarded as an approximate location
// of the shiro-rekha. By descending on the maxima's peak on both sides,
// stroke width of shiro-rekha is estimated.
// A horizontal projection histogram is computed for a sub-image of the input
// image, which extends from just below the shiro-rekha down to a certain
// leeway. The leeway depends on the input xheight, if provided, else a
// conservative multiplier on approximate stroke width is used (which may lead
// to over-splitting).
void ShiroRekhaSplitter::SplitWordShiroRekha(SplitStrategy split_strategy,
Pix* pix,
int xheight,
int word_left,
int word_top,
Boxa* regions_to_clear) {
if (split_strategy == NO_SPLIT) {
return;
}
int width = pixGetWidth(pix);
int height = pixGetHeight(pix);
// Statistically determine the yextents of the shiro-rekha.
int shirorekha_top, shirorekha_bottom, shirorekha_ylevel;
GetShiroRekhaYExtents(pix, &shirorekha_top, &shirorekha_bottom,
&shirorekha_ylevel);
// Since the shiro rekha is also a stroke, its width is equal to the stroke
// width.
int stroke_width = shirorekha_bottom - shirorekha_top + 1;
// Some safeguards to protect CCs we do not want to be split.
// These are particularly useful when the word wasn't eliminated earlier
// because xheight information was unavailable.
if (shirorekha_ylevel > height / 2) {
// Shirorekha shouldn't be in the bottom half of the word.
if (devanagari_split_debuglevel > 0) {
tprintf("Skipping splitting CC at (%d, %d): shirorekha in lower half..\n",
word_left, word_top);
}
return;
}
if (stroke_width > height / 3) {
// Even the boldest of fonts shouldn't do this.
if (devanagari_split_debuglevel > 0) {
tprintf("Skipping splitting CC at (%d, %d): stroke width too huge..\n",
word_left, word_top);
}
return;
}
// Clear the ascender and descender regions of the word.
// Obtain a vertical projection histogram for the resulting image.
Box* box_to_clear = boxCreate(0, shirorekha_top - stroke_width / 3,
width, 5 * stroke_width / 3);
Pix* word_in_xheight = pixCopy(NULL, pix);
pixClearInRect(word_in_xheight, box_to_clear);
// Also clear any pixels which are below shirorekha_bottom + some leeway.
// The leeway is set to xheight if the information is available, else it is a
// multiplier applied to the stroke width.
int leeway_to_keep = stroke_width * 3;
if (xheight != kUnspecifiedXheight) {
// This is because the xheight-region typically includes the shiro-rekha
// inside it, i.e., the top of the xheight range corresponds to the top of
// shiro-rekha.
leeway_to_keep = xheight - stroke_width;
}
box_to_clear->y = shirorekha_bottom + leeway_to_keep;
box_to_clear->h = height - box_to_clear->y;
pixClearInRect(word_in_xheight, box_to_clear);
boxDestroy(&box_to_clear);
PixelHistogram vert_hist;
vert_hist.ConstructVerticalCountHist(word_in_xheight);
pixDestroy(&word_in_xheight);
// If the number of black pixel in any column of the image is less than a
// fraction of the stroke width, treat it as noise / a stray mark. Perform
// these changes inside the vert_hist data itself, as that is used later on as
// a bit vector for the final split decision at every column.
for (int i = 0; i < width; ++i) {
if (vert_hist.hist()[i] <= stroke_width / 4)
vert_hist.hist()[i] = 0;
else
vert_hist.hist()[i] = 1;
}
// In order to split the line at any point, we make sure that the width of the
// gap is atleast half the stroke width.
int i = 0;
int cur_component_width = 0;
while (i < width) {
if (!vert_hist.hist()[i]) {
int j = 0;
while (i + j < width && !vert_hist.hist()[i+j])
++j;
if (j >= stroke_width / 2 && cur_component_width >= stroke_width / 2) {
// Perform a shiro-rekha split. The intervening region lies from i to
// i+j-1.
// A minimal single-pixel split makes the estimation of intra- and
// inter-word spacing easier during page layout analysis,
// whereas a maximal split may be needed for OCR, depending on
// how the engine was trained.
bool minimal_split = (split_strategy == MINIMAL_SPLIT);
int split_width = minimal_split ? 1 : j;
int split_left = minimal_split ? i + (j / 2) - (split_width / 2) : i;
if (!minimal_split || (i != 0 && i + j != width)) {
Box* box_to_clear =
boxCreate(word_left + split_left,
word_top + shirorekha_top - stroke_width / 3,
split_width,
5 * stroke_width / 3);
if (box_to_clear) {
boxaAddBox(regions_to_clear, box_to_clear, L_CLONE);
// Mark this in the debug image if needed.
if (devanagari_split_debugimage) {
pixRenderBoxArb(debug_image_, box_to_clear, 1, 128, 255, 128);
}
boxDestroy(&box_to_clear);
cur_component_width = 0;
}
}
}
i += j;
} else {
++i;
++cur_component_width;
}
}
}
/*!
* pixWriteMemWebP()
*
* Input: &encdata (<return> webp encoded data of pixs)
* &encsize (<return> size of webp encoded data)
* pixs (any depth, cmapped OK)
* quality (0 - 100; default ~80)
* lossless (use 1 for lossless; 0 for lossy)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) Lossless and lossy encoding are entirely different in webp.
* @quality applies to lossy, and is ignored for lossless.
* (2) The input image is converted to RGB if necessary. If spp == 3,
* we set the alpha channel to fully opaque (255), and
* WebPEncodeRGBA() then removes the alpha chunk when encoding,
* setting the internal header field has_alpha to 0.
*/
l_int32
pixWriteMemWebP(l_uint8 **pencdata,
size_t *pencsize,
PIX *pixs,
l_int32 quality,
l_int32 lossless)
{
l_int32 w, h, d, wpl, stride;
l_uint32 *data;
PIX *pix1, *pix2;
PROCNAME("pixWriteMemWebP");
if (!pencdata)
return ERROR_INT("&encdata not defined", procName, 1);
*pencdata = NULL;
if (!pencsize)
return ERROR_INT("&encsize not defined", procName, 1);
*pencsize = 0;
if (!pixs)
return ERROR_INT("&pixs not defined", procName, 1);
if (lossless == 0 && (quality < 0 || quality > 100))
return ERROR_INT("quality not in [0 ... 100]", procName, 1);
if ((pix1 = pixRemoveColormap(pixs, REMOVE_CMAP_TO_FULL_COLOR)) == NULL)
return ERROR_INT("failure to remove color map", procName, 1);
/* Convert to rgb if not 32 bpp; pix2 must not be a clone of pixs. */
if (pixGetDepth(pix1) != 32)
pix2 = pixConvertTo32(pix1);
else
pix2 = pixCopy(NULL, pix1);
pixDestroy(&pix1);
pixGetDimensions(pix2, &w, &h, &d);
if (w <= 0 || h <= 0 || d != 32) {
pixDestroy(&pix2);
return ERROR_INT("pix2 not 32 bpp or of 0 size", procName, 1);
}
/* If spp == 3, need to set alpha layer to opaque (all 1s). */
if (pixGetSpp(pix2) == 3)
pixSetComponentArbitrary(pix2, L_ALPHA_CHANNEL, 255);
/* Webp encoder assumes big-endian byte order for RGBA components */
pixEndianByteSwap(pix2);
wpl = pixGetWpl(pix2);
data = pixGetData(pix2);
stride = wpl * 4;
if (lossless) {
*pencsize = WebPEncodeLosslessRGBA((uint8_t *)data, w, h,
stride, pencdata);
} else {
*pencsize = WebPEncodeRGBA((uint8_t *)data, w, h, stride,
quality, pencdata);
}
pixDestroy(&pix2);
if (*pencsize == 0) {
free(pencdata);
*pencdata = NULL;
return ERROR_INT("webp encoding failed", procName, 1);
}
return 0;
}
main(int argc,
char **argv)
{
char *filein;
l_int32 i, orient;
l_float32 upconf1, upconf2, leftconf1, leftconf2, conf1, conf2;
PIX *pixs, *pixt1, *pixt2;
static char mainName[] = "flipdetect_reg";
if (argc != 2)
exit(ERROR_INT(" Syntax: flipdetect_reg filein", mainName, 1));
filein = argv[1];
if ((pixt1 = pixRead(filein)) == NULL)
exit(ERROR_INT("pixt1 not made", mainName, 1));
pixs = pixConvertTo1(pixt1, 130);
pixDestroy(&pixt1);
fprintf(stderr, "\nTest orientation detection\n");
startTimer();
pixOrientDetect(pixs, &upconf1, &leftconf1, 0, 0);
fprintf(stderr, "Time for rop orient test: %7.3f sec\n", stopTimer());
makeOrientDecision(upconf1, leftconf1, 0, 0, &orient, 1);
startTimer();
pixOrientDetectDwa(pixs, &upconf2, &leftconf2, 0, 0);
fprintf(stderr, "Time for dwa orient test: %7.3f sec\n", stopTimer());
if (upconf1 == upconf2 && leftconf1 == leftconf2) {
printStarredMessage("Orient results identical");
fprintf(stderr, "upconf = %7.3f, leftconf = %7.3f\n",
upconf1, leftconf1);
}
else {
printStarredMessage("Orient results differ");
fprintf(stderr, "upconf1 = %7.3f, upconf2 = %7.3f\n", upconf1, upconf2);
fprintf(stderr, "leftconf1 = %7.3f, leftconf2 = %7.3f\n",
leftconf1, leftconf2);
}
pixt1 = pixCopy(NULL, pixs);
fprintf(stderr, "\nTest orient detection for 4 orientations\n");
for (i = 0; i < 4; i++) {
pixOrientDetectDwa(pixt1, &upconf2, &leftconf2, 0, 0);
makeOrientDecision(upconf2, leftconf2, 0, 0, &orient, 1);
if (i == 3) break;
pixt2 = pixRotate90(pixt1, 1);
pixDestroy(&pixt1);
pixt1 = pixt2;
}
pixDestroy(&pixt1);
fprintf(stderr, "\nTest mirror reverse detection\n");
startTimer();
pixMirrorDetect(pixs, &conf1, 0, 1);
fprintf(stderr, "Time for rop mirror flip test: %7.3f sec\n", stopTimer());
startTimer();
pixMirrorDetectDwa(pixs, &conf2, 0, 0);
fprintf(stderr, "Time for dwa mirror flip test: %7.3f sec\n", stopTimer());
if (conf1 == conf2) {
printStarredMessage("Mirror results identical");
fprintf(stderr, "conf = %7.3f\n", conf1);
}
else {
printStarredMessage("Mirror results differ");
fprintf(stderr, "conf1 = %7.3f, conf2 = %7.3f\n", conf1, conf2);
}
fprintf(stderr, "\nSafer version of up-down tests\n");
pixUpDownDetectGeneral(pixs, &conf1, 0, 10, 1);
pixUpDownDetectGeneralDwa(pixs, &conf2, 0, 10, 1);
if (conf1 == conf2)
fprintf(stderr, "Confidence results are identical\n");
else
fprintf(stderr, "Confidence results differ\n");
pixDestroy(&pixs);
exit(0);
}