/*!
* \brief boxConvertToPta()
*
* \param[in] box
* \param[in] ncorners 2 or 4 for the representation of the box
* \return pta with %ncorners points, or NULL on error
*
* <pre>
* Notes:
* (1) If ncorners == 2, we select the UL and LR corners.
* Otherwise we save all 4 corners in this order: UL, UR, LL, LR.
* </pre>
*/
PTA *
boxConvertToPta(BOX *box,
l_int32 ncorners)
{
l_int32 x, y, w, h;
PTA *pta;
PROCNAME("boxConvertToPta");
if (!box)
return (PTA *)ERROR_PTR("box not defined", procName, NULL);
if (ncorners != 2 && ncorners != 4)
return (PTA *)ERROR_PTR("ncorners not 2 or 4", procName, NULL);
if ((pta = ptaCreate(ncorners)) == NULL)
return (PTA *)ERROR_PTR("pta not made", procName, NULL);
boxGetGeometry(box, &x, &y, &w, &h);
ptaAddPt(pta, x, y);
if (ncorners == 2) {
ptaAddPt(pta, x + w - 1, y + h - 1);
} else {
ptaAddPt(pta, x + w - 1, y);
ptaAddPt(pta, x, y + h - 1);
ptaAddPt(pta, x + w - 1, y + h - 1);
}
return pta;
}
/*!
* boxaConvertToPta()
*
* Input: boxa
* ncorners (2 or 4 for the representation of each box)
* Return: pta (with @ncorners points for each box in the boxa),
* or null on error
*
* Notes:
* (1) If ncorners == 2, we select the UL and LR corners.
* Otherwise we save all 4 corners in this order: UL, UR, LL, LR.
*/
PTA *
boxaConvertToPta(BOXA *boxa,
l_int32 ncorners)
{
l_int32 i, n, x, y, w, h;
PTA *pta;
PROCNAME("boxaConvertToPta");
if (!boxa)
return (PTA *)ERROR_PTR("boxa not defined", procName, NULL);
if (ncorners != 2 && ncorners != 4)
return (PTA *)ERROR_PTR("ncorners not 2 or 4", procName, NULL);
n = boxaGetCount(boxa);
if ((pta = ptaCreate(n)) == NULL)
return (PTA *)ERROR_PTR("pta not made", procName, NULL);
for (i = 0; i < n; i++) {
boxaGetBoxGeometry(boxa, i, &x, &y, &w, &h);
ptaAddPt(pta, x, y);
if (ncorners == 2)
ptaAddPt(pta, x + w - 1, y + h - 1);
else {
ptaAddPt(pta, x + w - 1, y);
ptaAddPt(pta, x, y + h - 1);
ptaAddPt(pta, x + w - 1, y + h - 1);
}
}
return pta;
}
main(int argc,
char **argv)
{
char *filein, *fileout;
l_int32 d;
BOX *box1, *box2, *box3, *box4;
BOXA *boxa;
PIX *pixs, *pixt1, *pixt2, *pixt3;
PTA *pta;
static char mainName[] = "graphicstest";
if (argc != 3)
exit(ERROR_INT(" Syntax: graphicstest filein fileout", mainName, 1));
filein = argv[1];
fileout = argv[2];
if ((pixs = pixRead(filein)) == NULL)
exit(ERROR_INT(" Syntax: pixs not made", mainName, 1));
d = pixGetDepth(pixs);
if (d <= 8)
pixt1 = pixConvertTo32(pixs);
else
pixt1 = pixClone(pixs);
/* Paint on RGB */
pixRenderLineArb(pixt1, 450, 20, 850, 320, 5, 200, 50, 125);
pixRenderLineArb(pixt1, 30, 40, 440, 40, 5, 100, 200, 25);
pixRenderLineBlend(pixt1, 30, 60, 440, 70, 5, 115, 200, 120, 0.3);
pixRenderLineBlend(pixt1, 30, 600, 440, 670, 9, 215, 115, 30, 0.5);
pixRenderLineBlend(pixt1, 130, 700, 540, 770, 9, 255, 255, 250, 0.4);
pixRenderLineBlend(pixt1, 130, 800, 540, 870, 9, 0, 0, 0, 0.4);
box1 = boxCreate(70, 80, 300, 245);
box2 = boxCreate(470, 180, 150, 205);
box3 = boxCreate(520, 220, 160, 220);
box4 = boxCreate(570, 260, 160, 220);
boxa = boxaCreate(3);
boxaAddBox(boxa, box2, L_INSERT);
boxaAddBox(boxa, box3, L_INSERT);
boxaAddBox(boxa, box4, L_INSERT);
pixRenderBoxArb(pixt1, box1, 3, 200, 200, 25);
pixRenderBoxaBlend(pixt1, boxa, 17, 200, 200, 25, 0.4, 1);
pta = ptaCreate(5);
ptaAddPt(pta, 250, 300);
ptaAddPt(pta, 350, 450);
ptaAddPt(pta, 400, 600);
ptaAddPt(pta, 212, 512);
ptaAddPt(pta, 180, 375);
pixRenderPolylineBlend(pixt1, pta, 17, 25, 200, 200, 0.5, 1, 1);
pixWrite(fileout, pixt1, IFF_JFIF_JPEG);
pixDisplay(pixt1, 200, 200);
pixDestroy(&pixs);
pixDestroy(&pixt1);
boxDestroy(&box1);
boxaDestroy(&boxa);
ptaDestroy(&pta);
pixDestroy(&pixs);
return 0;
}
/*!
* pixSubsampleBoundaryPixels()
*
* Input: pixs (1 bpp, with only boundary pixels in fg)
* skip (number to skip between samples as you traverse boundary)
* Return: pta, or null on error
*
* Notes:
* (1) If skip = 0, we take all the fg pixels.
* (2) We try to traverse the boundaries in a regular way.
* Some pixels may be missed, and these are then subsampled
* randomly with a fraction determined by 'skip'.
* (3) The most natural approach is to use a depth first (stack-based)
* method to find the fg pixels. However, the pixel runs are
* 4-connected and there are relatively few branches. So
* instead of doing a proper depth-first search, we get nearly
* the same result using two nested while loops: the outer
* one continues a raster-based search for the next fg pixel,
* and the inner one does a reasonable job running along
* each 4-connected coutour.
*/
PTA *
pixSubsampleBoundaryPixels(PIX *pixs,
l_int32 skip)
{
l_int32 x, y, xn, yn, xs, ys, xa, ya, count;
PIX *pixt;
PTA *pta;
PROCNAME("pixSubsampleBoundaryPixels");
if (!pixs)
return (PTA *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 1)
return (PTA *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
if (skip < 0)
return (PTA *)ERROR_PTR("skip < 0", procName, NULL);
if (skip == 0)
return ptaGetPixelsFromPix(pixs, NULL);
pta = ptaCreate(0);
pixt = pixCopy(NULL, pixs);
xs = ys = 0;
while (nextOnPixelInRaster(pixt, xs, ys, &xn, &yn)) { /* new series */
xs = xn;
ys = yn;
/* Add first point in this series */
ptaAddPt(pta, xs, ys);
/* Trace out boundary, erasing all and saving every (skip + 1)th */
x = xs;
y = ys;
pixSetPixel(pixt, x, y, 0);
count = 0;
while (adjacentOnPixelInRaster(pixt, x, y, &xa, &ya)) {
x = xa;
y = ya;
pixSetPixel(pixt, x, y, 0);
if (count == skip) {
ptaAddPt(pta, x, y);
count = 0;
}
else {
count++;
}
}
}
pixDestroy(&pixt);
return pta;
}
/*!
* ptaRemoveDupsByAset()
*
* Input: ptas (assumed to be integer values)
* Return: ptad (with duplicates removed), or null on error
*
* Notes:
* (1) This is slower than ptaRemoveDupsByHash(), mostly because
* of the nlogn sort to build up the rbtree. Do not use for
* large numbers of points (say, > 1M).
*/
PTA *
ptaRemoveDupsByAset(PTA *ptas)
{
l_int32 i, n, x, y;
PTA *ptad;
l_uint64 hash;
L_ASET *set;
RB_TYPE key;
PROCNAME("ptaRemoveDupsByAset");
if (!ptas)
return (PTA *)ERROR_PTR("ptas not defined", procName, NULL);
set = l_asetCreate(L_UINT_TYPE);
n = ptaGetCount(ptas);
ptad = ptaCreate(n);
for (i = 0; i < n; i++) {
ptaGetIPt(ptas, i, &x, &y);
l_hashPtToUint64(x, y, &hash);
key.utype = hash;
if (!l_asetFind(set, key)) {
ptaAddPt(ptad, x, y);
l_asetInsert(set, key);
}
}
l_asetDestroy(&set);
return ptad;
}
/*!
* ptaSortByIndex()
*
* Input: ptas
* naindex (na that maps from the new pta to the input pta)
* Return: ptad (sorted), or null on error
*/
PTA *
ptaSortByIndex(PTA *ptas,
NUMA *naindex)
{
l_int32 i, index, n;
l_float32 x, y;
PTA *ptad;
PROCNAME("ptaSortByIndex");
if (!ptas)
return (PTA *)ERROR_PTR("ptas not defined", procName, NULL);
if (!naindex)
return (PTA *)ERROR_PTR("naindex not defined", procName, NULL);
/* Build up sorted pta using sort index */
n = numaGetCount(naindex);
if ((ptad = ptaCreate(n)) == NULL)
return (PTA *)ERROR_PTR("ptad not made", procName, NULL);
for (i = 0; i < n; i++) {
numaGetIValue(naindex, i, &index);
ptaGetPt(ptas, index, &x, &y);
ptaAddPt(ptad, x, y);
}
return ptad;
}
/*!
* ptaRotate()
*
* Input: ptas (for initial points)
* (xc, yc) (location of center of rotation)
* angle (rotation in radians; clockwise is positive)
* (&ptad) (<return> new locations)
* Return: 0 if OK; 1 on error
*
* Notes;
* (1) See createMatrix2dScale() for details of transform.
*/
PTA *
ptaRotate(PTA *ptas,
l_float32 xc,
l_float32 yc,
l_float32 angle)
{
l_int32 i, npts;
l_float32 x, y, xp, yp, sina, cosa;
PTA *ptad;
PROCNAME("ptaRotate");
if (!ptas)
return (PTA *)ERROR_PTR("ptas not defined", procName, NULL);
npts = ptaGetCount(ptas);
if ((ptad = ptaCreate(npts)) == NULL)
return (PTA *)ERROR_PTR("ptad not made", procName, NULL);
sina = sin(angle);
cosa = cos(angle);
for (i = 0; i < npts; i++) {
ptaGetPt(ptas, i, &x, &y);
xp = xc + (x - xc) * cosa - (y - yc) * sina;
yp = yc + (x - xc) * sina + (y - yc) * cosa;
ptaAddPt(ptad, xp, yp);
}
return ptad;
}
/*!
* ptaGetMeanVerticals()
*
* Input: pixs (1 bpp, single c.c.)
* x,y (location of UL corner of pixs with respect to page image
* Return: pta (mean y-values in component for each x-value,
* both translated by (x,y)
*/
PTA *
pixGetMeanVerticals(PIX *pixs,
l_int32 x,
l_int32 y)
{
l_int32 w, h, i, j, wpl, sum, count;
l_uint32 *line, *data;
PTA *pta;
PROCNAME("pixGetMeanVerticals");
if (!pixs || pixGetDepth(pixs) != 1)
return (PTA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
pixGetDimensions(pixs, &w, &h, NULL);
pta = ptaCreate(w);
data = pixGetData(pixs);
wpl = pixGetWpl(pixs);
for (j = 0; j < w; j++) {
line = data;
sum = count = 0;
for (i = 0; i < h; i++) {
if (GET_DATA_BIT(line, j) == 1) {
sum += i;
count += 1;
}
line += wpl;
}
if (count == 0) continue;
ptaAddPt(pta, x + j, y + (sum / count));
}
return pta;
}
/*!
* ptaAffineTransform()
*
* Input: ptas (for initial points)
* mat (3x3 transform matrix; canonical form)
* Return: ptad (transformed points), or null on error
*/
PTA *
ptaAffineTransform(PTA *ptas,
l_float32 *mat)
{
l_int32 i, npts;
l_float32 vecs[3], vecd[3];
PTA *ptad;
PROCNAME("ptaAffineTransform");
if (!ptas)
return (PTA *)ERROR_PTR("ptas not defined", procName, NULL);
if (!mat)
return (PTA *)ERROR_PTR("transform not defined", procName, NULL);
vecs[2] = 1;
npts = ptaGetCount(ptas);
if ((ptad = ptaCreate(npts)) == NULL)
return (PTA *)ERROR_PTR("ptad not made", procName, NULL);
for (i = 0; i < npts; i++) {
ptaGetPt(ptas, i, &vecs[0], &vecs[1]);
l_productMatVec(mat, vecs, vecd, 3);
ptaAddPt(ptad, vecd[0], vecd[1]);
}
return ptad;
}
static l_float32 *
Generate3PtTransformVector() {
l_int32 i;
l_float32 *vc;
PTA *ptas, *ptad;
ptas = ptaCreate(3);
ptad = ptaCreate(3);
for (i = 0; i < 3; i++) {
ptaAddPt(ptas, xs[i], ys[i]);
ptaAddPt(ptad, xd[i], yd[i]);
}
getAffineXformCoeffs(ptad, ptas, &vc);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
return vc;
}
static PTA *
BuildPointSet(l_int32 w, l_int32 h, l_int32 add_dups)
{
l_int32 i, j;
PTA *pta;
pta = ptaCreate(w * h);
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++)
ptaAddPt(pta, j, i);
if (add_dups) { /* extra (0.2 * w * h) points */
for (j = 0.4 * w; j < 0.6 * w; j++)
ptaAddPt(pta, j, i);
}
}
return pta;
}
/*!
* boxaExtractAsPta()
*
* Input: boxa
* &ptal (<optional return> array of left locations vs. index)
* &ptat (<optional return> array of top locations vs. index)
* &ptar (<optional return> array of right locations vs. index)
* &ptab (<optional return> array of bottom locations vs. index)
* keepinvalid (1 to keep invalid boxes; 0 to remove them)
* Return: 0 if OK, 1 on error
*/
l_int32
boxaExtractAsPta(BOXA *boxa,
PTA **pptal,
PTA **pptat,
PTA **pptar,
PTA **pptab,
l_int32 keepinvalid)
{
l_int32 i, n, left, top, right, bot, w, h;
PROCNAME("boxaExtractAsPta");
if (!pptal && !pptar && !pptat && !pptab)
return ERROR_INT("no output requested", procName, 1);
if (pptal) *pptal = NULL;
if (pptat) *pptat = NULL;
if (pptar) *pptar = NULL;
if (pptab) *pptab = NULL;
if (!boxa)
return ERROR_INT("boxa not defined", procName, 1);
if (!keepinvalid && boxaGetValidCount(boxa) == 0)
return ERROR_INT("no valid boxes", procName, 1);
n = boxaGetCount(boxa);
if (pptal) *pptal = ptaCreate(n);
if (pptat) *pptat = ptaCreate(n);
if (pptar) *pptar = ptaCreate(n);
if (pptab) *pptab = ptaCreate(n);
for (i = 0; i < n; i++) {
boxaGetBoxGeometry(boxa, i, &left, &top, &w, &h);
if (!keepinvalid && (w <= 0 || h <= 0))
continue;
right = left + w - 1;
bot = top + h - 1;
if (pptal) ptaAddPt(*pptal, i, left);
if (pptat) ptaAddPt(*pptat, i, top);
if (pptar) ptaAddPt(*pptar, i, right);
if (pptab) ptaAddPt(*pptab, i, bot);
}
return 0;
}
static l_float32 *
Generate4PtTransformVector(l_int32 type) {
l_int32 i;
l_float32 *vc;
PTA *ptas, *ptad;
ptas = ptaCreate(4);
ptad = ptaCreate(4);
for (i = 0; i < 4; i++) {
ptaAddPt(ptas, xs[i], ys[i]);
ptaAddPt(ptad, xd[i], yd[i]);
}
if (type == PROJECTIVE)
getProjectiveXformCoeffs(ptad, ptas, &vc);
else /* BILINEAR */
getBilinearXformCoeffs(ptad, ptas, &vc);
ptaDestroy(&ptas);
ptaDestroy(&ptad);
return vc;
}
static void
MakePtas(l_int32 i,
PTA **pptas,
PTA **pptad)
{
*pptas = ptaCreate(4);
ptaAddPt(*pptas, x1[i], y1[i]);
ptaAddPt(*pptas, x2[i], y2[i]);
ptaAddPt(*pptas, x3[i], y3[i]);
ptaAddPt(*pptas, x4[i], y4[i]);
*pptad = ptaCreate(4);
ptaAddPt(*pptad, xp1[i], yp1[i]);
ptaAddPt(*pptad, xp2[i], yp2[i]);
ptaAddPt(*pptad, xp3[i], yp3[i]);
ptaAddPt(*pptad, xp4[i], yp4[i]);
return;
}
/*!
* ptaIntersectionByHash()
*
* Input: pta1, pta2
* Return: ptad (intersection of the point sets), or null on error
*
* Notes:
* (1) This is faster than ptaIntersectionByAset(), because the
* bucket lookup is O(n). It should be used if the pts are
* integers (e.g., representing pixel positions).
*/
PTA *
ptaIntersectionByHash(PTA *pta1,
PTA *pta2)
{
l_int32 n1, n2, nsmall, i, x, y, index1, index2;
l_uint32 nsize2;
l_uint64 key;
L_DNAHASH *dahash1, *dahash2;
PTA *pta_small, *pta_big, *ptad;
PROCNAME("ptaIntersectionByHash");
if (!pta1)
return (PTA *)ERROR_PTR("pta1 not defined", procName, NULL);
if (!pta2)
return (PTA *)ERROR_PTR("pta2 not defined", procName, NULL);
/* Put the elements of the biggest pta into a dnahash */
n1 = ptaGetCount(pta1);
n2 = ptaGetCount(pta2);
pta_small = (n1 < n2) ? pta1 : pta2; /* do not destroy pta_small */
pta_big = (n1 < n2) ? pta2 : pta1; /* do not destroy pta_big */
dahash1 = l_dnaHashCreateFromPta(pta_big);
/* Build up the intersection of points. Add to ptad
* if the point is in pta_big (using dahash1) but hasn't
* yet been seen in the traversal of pta_small (using dahash2). */
ptad = ptaCreate(0);
nsmall = ptaGetCount(pta_small);
findNextLargerPrime(nsmall / 20, &nsize2); /* buckets in hash table */
dahash2 = l_dnaHashCreate(nsize2, 0);
for (i = 0; i < nsmall; i++) {
ptaGetIPt(pta_small, i, &x, &y);
ptaFindPtByHash(pta_big, dahash1, x, y, &index1);
if (index1 >= 0) { /* found */
ptaFindPtByHash(pta_small, dahash2, x, y, &index2);
if (index2 == -1) { /* not found */
ptaAddPt(ptad, x, y);
l_hashPtToUint64Fast(nsize2, x, y, &key);
l_dnaHashAdd(dahash2, key, (l_float64)i);
}
}
}
l_dnaHashDestroy(&dahash1);
l_dnaHashDestroy(&dahash2);
return ptad;
}
// Computes the coefficients of a randomized projective transformation.
// The image transform requires backward transformation coefficient, and the
// box transform the forward coefficients.
// Returns the incolor arg to pixProjective.
int ProjectiveCoeffs(int width, int height, TRand* randomizer,
float** im_coeffs, float** box_coeffs) {
// Setup "from" points.
Pta* src_pts = ptaCreate(4);
ptaAddPt(src_pts, 0.0f, 0.0f);
ptaAddPt(src_pts, width, 0.0f);
ptaAddPt(src_pts, width, height);
ptaAddPt(src_pts, 0.0f, height);
// Extract factors from pseudo-random sequence.
float factors[FN_NUM_FACTORS];
float shear = 0.0f; // Shear is signed.
for (int i = 0; i < FN_NUM_FACTORS; ++i) {
// Everything is squared to make wild values rarer.
if (i == FN_SHEAR) {
// Shear is signed.
shear = randomizer->SignedRand(0.5 / 3.0);
shear = shear >= 0.0 ? shear * shear : -shear * shear;
// Keep the sheared points within the original rectangle.
if (shear < -factors[FN_X0]) shear = -factors[FN_X0];
if (shear > factors[FN_X1]) shear = factors[FN_X1];
factors[i] = shear;
} else if (i != FN_INCOLOR) {
factors[i] = fabs(randomizer->SignedRand(1.0));
if (i <= FN_Y3)
factors[i] *= 5.0 / 8.0;
else
factors[i] *= 0.5;
factors[i] *= factors[i];
}
}
// Setup "to" points.
Pta* dest_pts = ptaCreate(4);
ptaAddPt(dest_pts, factors[FN_X0] * width, factors[FN_Y0] * height);
ptaAddPt(dest_pts, (1.0f - factors[FN_X1]) * width, factors[FN_Y1] * height);
ptaAddPt(dest_pts, (1.0f - factors[FN_X1] + shear) * width,
(1 - factors[FN_Y2]) * height);
ptaAddPt(dest_pts, (factors[FN_X0] + shear) * width,
(1 - factors[FN_Y3]) * height);
getProjectiveXformCoeffs(dest_pts, src_pts, im_coeffs);
getProjectiveXformCoeffs(src_pts, dest_pts, box_coeffs);
ptaDestroy(&src_pts);
ptaDestroy(&dest_pts);
return factors[FN_INCOLOR] > 0.5f ? L_BRING_IN_WHITE : L_BRING_IN_BLACK;
}
/** Returns the polygon outline of the current block. The returned Pta must
* be ptaDestroy-ed after use. */
Pta* PageIterator::BlockPolygon() const {
if (it_->block() == NULL || it_->block()->block == NULL)
return NULL; // Already at the end!
if (it_->block()->block->poly_block() == NULL)
return NULL; // No layout analysis used - no polygon.
ICOORDELT_IT it(it_->block()->block->poly_block()->points());
Pta* pta = ptaCreate(it.length());
int num_pts = 0;
for (it.mark_cycle_pt(); !it.cycled_list(); it.forward(), ++num_pts) {
ICOORD* pt = it.data();
// Convert to top-down coords within the input image.
float x = static_cast<float>(pt->x()) / scale_ + rect_left_;
float y = rect_top_ + rect_height_ - static_cast<float>(pt->y()) / scale_;
ptaAddPt(pta, x, y);
}
return pta;
}
static void
MakePtas(l_int32 i,
l_int32 npts, /* 3 or 4 */
PTA **pptas,
PTA **pptad)
{
*pptas = ptaCreate(npts);
ptaAddPt(*pptas, x1[i], y1[i]);
ptaAddPt(*pptas, x2[i], y2[i]);
ptaAddPt(*pptas, x3[i], y3[i]);
if (npts == 4) ptaAddPt(*pptas, x4[i], y4[i]);
*pptad = ptaCreate(npts);
ptaAddPt(*pptad, xp1[i], yp1[i]);
ptaAddPt(*pptad, xp2[i], yp2[i]);
ptaAddPt(*pptad, xp3[i], yp3[i]);
if (npts == 4) ptaAddPt(*pptad, xp4[i], yp4[i]);
return;
}
static void
MakePtas(l_int32 i,
PTA **pptas,
PTA **pptad)
{
PTA *ptas, *ptad;
ptas = ptaCreate(4);
ptaAddPt(ptas, xs1[i], ys1[i]);
ptaAddPt(ptas, xs2[i], ys2[i]);
ptaAddPt(ptas, xs3[i], ys3[i]);
ptaAddPt(ptas, xs4[i], ys4[i]);
ptad = ptaCreate(4);
ptaAddPt(ptad, xd1[i], yd1[i]);
ptaAddPt(ptad, xd2[i], yd2[i]);
ptaAddPt(ptad, xd3[i], yd3[i]);
ptaAddPt(ptad, xd4[i], yd4[i]);
*pptas = ptas;
*pptad = ptad;
return;
}
/*!
* ptaRemoveDupsByHash()
*
* Input: ptas (assumed to be integer values)
* &ptad (<return> unique set of pts; duplicates removed)
* &dahash (<optional return> dnahash used for lookup)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) Generates a pta with unique values.
* (2) The dnahash is built up with ptad to assure uniqueness.
* It can be used to find if a point is in the set:
* ptaFindPtByHash(ptad, dahash, x, y, &index)
* (3) The hash of the (x,y) location is simple and fast. It scales
* up with the number of buckets to insure a fairly random
* bucket selection for adjacent points.
* (4) A Dna is used rather than a Numa because we need accurate
* representation of 32-bit integers that are indices into ptas.
* Integer --> float --> integer conversion makes errors for
* integers larger than 10M.
* (5) This is faster than ptaRemoveDupsByAset(), because the
* bucket lookup is O(n), although there is a double-loop
* lookup within the dna in each bucket.
*/
l_int32
ptaRemoveDupsByHash(PTA *ptas,
PTA **pptad,
L_DNAHASH **pdahash)
{
l_int32 i, n, index, items, x, y;
l_uint32 nsize;
l_uint64 key;
l_float64 val;
PTA *ptad;
L_DNAHASH *dahash;
PROCNAME("ptaRemoveDupsByHash");
if (pdahash) *pdahash = NULL;
if (!pptad)
return ERROR_INT("&ptad not defined", procName, 1);
*pptad = NULL;
if (!ptas)
return ERROR_INT("ptas not defined", procName, 1);
n = ptaGetCount(ptas);
findNextLargerPrime(n / 20, &nsize); /* buckets in hash table */
dahash = l_dnaHashCreate(nsize, 8);
ptad = ptaCreate(n);
*pptad = ptad;
for (i = 0, items = 0; i < n; i++) {
ptaGetIPt(ptas, i, &x, &y);
ptaFindPtByHash(ptad, dahash, x, y, &index);
if (index < 0) { /* not found */
l_hashPtToUint64Fast(nsize, x, y, &key);
l_dnaHashAdd(dahash, key, (l_float64)items);
ptaAddPt(ptad, x, y);
items++;
}
}
if (pdahash)
*pdahash = dahash;
else
l_dnaHashDestroy(&dahash);
return 0;
}
/*!
* ptaIntersectionByAset()
*
* Input: pta1, pta2
* Return: ptad (intersection of the point sets), or null on error
*
* Notes:
* (1) See sarrayIntersectionByAset() for the approach.
* (2) The key is a 64-bit hash from the (x,y) pair.
* (3) This is slower than ptaIntersectionByHash(), mostly because
* of the nlogn sort to build up the rbtree. Do not use for
* large numbers of points (say, > 1M).
*/
PTA *
ptaIntersectionByAset(PTA *pta1,
PTA *pta2)
{
l_int32 n1, n2, i, n, x, y;
l_uint64 hash;
L_ASET *set1, *set2;
RB_TYPE key;
PTA *pta_small, *pta_big, *ptad;
PROCNAME("ptaIntersectionByAset");
if (!pta1)
return (PTA *)ERROR_PTR("pta1 not defined", procName, NULL);
if (!pta2)
return (PTA *)ERROR_PTR("pta2 not defined", procName, NULL);
/* Put the elements of the biggest array into a set */
n1 = ptaGetCount(pta1);
n2 = ptaGetCount(pta2);
pta_small = (n1 < n2) ? pta1 : pta2; /* do not destroy pta_small */
pta_big = (n1 < n2) ? pta2 : pta1; /* do not destroy pta_big */
set1 = l_asetCreateFromPta(pta_big);
/* Build up the intersection of points */
ptad = ptaCreate(0);
n = ptaGetCount(pta_small);
set2 = l_asetCreate(L_UINT_TYPE);
for (i = 0; i < n; i++) {
ptaGetIPt(pta_small, i, &x, &y);
l_hashPtToUint64(x, y, &hash);
key.utype = hash;
if (l_asetFind(set1, key) && !l_asetFind(set2, key)) {
ptaAddPt(ptad, x, y);
l_asetInsert(set2, key);
}
}
l_asetDestroy(&set1);
l_asetDestroy(&set2);
return ptad;
}
void QSPLINE::plot(Pix *pix) const {
if (pix == NULL) {
return;
}
int32_t segment; // Index of segment
int16_t step; // Index of poly piece
double increment; // x increment
double x; // x coord
double height = static_cast<double>(pixGetHeight(pix));
Pta* points = ptaCreate(QSPLINE_PRECISION * segments);
const int kLineWidth = 5;
for (segment = 0; segment < segments; segment++) {
increment = static_cast<double>((xcoords[segment + 1] -
xcoords[segment])) / QSPLINE_PRECISION;
x = xcoords[segment];
for (step = 0; step <= QSPLINE_PRECISION; step++) {
double y = height - quadratics[segment].y(x);
ptaAddPt(points, x, y);
x += increment;
}
}
switch (pixGetDepth(pix)) {
case 1:
pixRenderPolyline(pix, points, kLineWidth, L_SET_PIXELS, 1);
break;
case 32:
pixRenderPolylineArb(pix, points, kLineWidth, 255, 0, 0, 1);
break;
default:
pixRenderPolyline(pix, points, kLineWidth, L_CLEAR_PIXELS, 1);
break;
}
ptaDestroy(&points);
}
/*!
* ptaScale()
*
* Input: ptas (for initial points)
* scalex (horizontal scale factor)
* scaley (vertical scale factor)
* Return: 0 if OK; 1 on error
*
* Notes;
* (1) See createMatrix2dScale() for details of transform.
*/
PTA *
ptaScale(PTA *ptas,
l_float32 scalex,
l_float32 scaley)
{
l_int32 i, npts;
l_float32 x, y;
PTA *ptad;
PROCNAME("ptaScale");
if (!ptas)
return (PTA *)ERROR_PTR("ptas not defined", procName, NULL);
npts = ptaGetCount(ptas);
if ((ptad = ptaCreate(npts)) == NULL)
return (PTA *)ERROR_PTR("ptad not made", procName, NULL);
for (i = 0; i < npts; i++) {
ptaGetPt(ptas, i, &x, &y);
ptaAddPt(ptad, scalex * x, scaley * y);
}
return ptad;
}
/*!
* \brief pixGetLocalSkewAngles()
*
* \param[in] pixs 1 bpp
* \param[in] nslices the number of horizontal overlapping slices; must
* be larger than 1 and not exceed 20; 0 for default
* \param[in] redsweep sweep reduction factor: 1, 2, 4 or 8;
* use 0 for default value
* \param[in] redsearch search reduction factor: 1, 2, 4 or 8, and not
* larger than redsweep; use 0 for default value
* \param[in] sweeprange half the full range, assumed about 0; in degrees;
* use 0.0 for default value
* \param[in] sweepdelta angle increment of sweep; in degrees;
* use 0.0 for default value
* \param[in] minbsdelta min binary search increment angle; in degrees;
* use 0.0 for default value
* \param[out] pa [optional] slope of skew as fctn of y
* \param[out] pb [optional] intercept at y=0 of skew as fctn of y
* \param[in] debug 1 for generating plot of skew angle vs. y; 0 otherwise
* \return naskew, or NULL on error
*
* <pre>
* Notes:
* (1) The local skew is measured in a set of overlapping strips.
* We then do a least square linear fit parameters to get
* the slope and intercept parameters a and b in
* skew-angle = a * y + b (degrees)
* for the local skew as a function of raster line y.
* This is then used to make naskew, which can be interpreted
* as the computed skew angle (in degrees) at the left edge
* of each raster line.
* (2) naskew can then be used to find the baselines of text, because
* each text line has a baseline that should intersect
* the left edge of the image with the angle given by this
* array, evaluated at the raster line of intersection.
* </pre>
*/
NUMA *
pixGetLocalSkewAngles(PIX *pixs,
l_int32 nslices,
l_int32 redsweep,
l_int32 redsearch,
l_float32 sweeprange,
l_float32 sweepdelta,
l_float32 minbsdelta,
l_float32 *pa,
l_float32 *pb,
l_int32 debug)
{
l_int32 w, h, hs, i, ystart, yend, ovlap, npts;
l_float32 angle, conf, ycenter, a, b;
BOX *box;
GPLOT *gplot;
NUMA *naskew, *nax, *nay;
PIX *pix;
PTA *pta;
PROCNAME("pixGetLocalSkewAngles");
if (!pixs || pixGetDepth(pixs) != 1)
return (NUMA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
if (nslices < 2 || nslices > 20)
nslices = DEFAULT_SLICES;
if (redsweep < 1 || redsweep > 8)
redsweep = DEFAULT_SWEEP_REDUCTION;
if (redsearch < 1 || redsearch > redsweep)
redsearch = DEFAULT_BS_REDUCTION;
if (sweeprange == 0.0)
sweeprange = DEFAULT_SWEEP_RANGE;
if (sweepdelta == 0.0)
sweepdelta = DEFAULT_SWEEP_DELTA;
if (minbsdelta == 0.0)
minbsdelta = DEFAULT_MINBS_DELTA;
pixGetDimensions(pixs, &w, &h, NULL);
hs = h / nslices;
ovlap = (l_int32)(OVERLAP_FRACTION * hs);
pta = ptaCreate(nslices);
for (i = 0; i < nslices; i++) {
ystart = L_MAX(0, hs * i - ovlap);
yend = L_MIN(h - 1, hs * (i + 1) + ovlap);
ycenter = (ystart + yend) / 2;
box = boxCreate(0, ystart, w, yend - ystart + 1);
pix = pixClipRectangle(pixs, box, NULL);
pixFindSkewSweepAndSearch(pix, &angle, &conf, redsweep, redsearch,
sweeprange, sweepdelta, minbsdelta);
if (conf > MIN_ALLOWED_CONFIDENCE)
ptaAddPt(pta, ycenter, angle);
pixDestroy(&pix);
boxDestroy(&box);
}
/* Do linear least squares fit */
if ((npts = ptaGetCount(pta)) < 2) {
ptaDestroy(&pta);
return (NUMA *)ERROR_PTR("can't fit skew", procName, NULL);
}
ptaGetLinearLSF(pta, &a, &b, NULL);
if (pa) *pa = a;
if (pb) *pb = b;
/* Make skew angle array as function of raster line */
naskew = numaCreate(h);
for (i = 0; i < h; i++) {
angle = a * i + b;
numaAddNumber(naskew, angle);
}
if (debug) {
lept_mkdir("lept/baseline");
ptaGetArrays(pta, &nax, &nay);
gplot = gplotCreate("/tmp/lept/baseline/skew", GPLOT_PNG,
"skew as fctn of y", "y (in raster lines from top)",
"angle (in degrees)");
gplotAddPlot(gplot, NULL, naskew, GPLOT_POINTS, "linear lsf");
gplotAddPlot(gplot, nax, nay, GPLOT_POINTS, "actual data pts");
gplotMakeOutput(gplot);
gplotDestroy(&gplot);
numaDestroy(&nax);
numaDestroy(&nay);
}
ptaDestroy(&pta);
return naskew;
}
/*!
* \brief pixGetLocalSkewTransform()
*
* \param[in] pixs
* \param[in] nslices the number of horizontal overlapping slices; must
* be larger than 1 and not exceed 20; use 0 for default
* \param[in] redsweep sweep reduction factor: 1, 2, 4 or 8;
* use 0 for default value
* \param[in] redsearch search reduction factor: 1, 2, 4 or 8, and
* not larger than redsweep; use 0 for default value
* \param[in] sweeprange half the full range, assumed about 0; in degrees;
* use 0.0 for default value
* \param[in] sweepdelta angle increment of sweep; in degrees;
* use 0.0 for default value
* \param[in] minbsdelta min binary search increment angle; in degrees;
* use 0.0 for default value
* \param[out] pptas 4 points in the source
* \param[out] pptad the corresponding 4 pts in the dest
* \return 0 if OK, 1 on error
*
* <pre>
* Notes:
* (1) This generates two pairs of points in the src, each pair
* corresponding to a pair of points that would lie along
* the same raster line in a transformed (dewarped) image.
* (2) The sets of 4 src and 4 dest points returned by this function
* can then be used, in a projective or bilinear transform,
* to remove keystoning in the src.
* </pre>
*/
l_int32
pixGetLocalSkewTransform(PIX *pixs,
l_int32 nslices,
l_int32 redsweep,
l_int32 redsearch,
l_float32 sweeprange,
l_float32 sweepdelta,
l_float32 minbsdelta,
PTA **pptas,
PTA **pptad)
{
l_int32 w, h, i;
l_float32 deg2rad, angr, angd, dely;
NUMA *naskew;
PTA *ptas, *ptad;
PROCNAME("pixGetLocalSkewTransform");
if (!pptas || !pptad)
return ERROR_INT("&ptas and &ptad not defined", procName, 1);
*pptas = *pptad = NULL;
if (!pixs || pixGetDepth(pixs) != 1)
return ERROR_INT("pixs not defined or not 1 bpp", procName, 1);
if (nslices < 2 || nslices > 20)
nslices = DEFAULT_SLICES;
if (redsweep < 1 || redsweep > 8)
redsweep = DEFAULT_SWEEP_REDUCTION;
if (redsearch < 1 || redsearch > redsweep)
redsearch = DEFAULT_BS_REDUCTION;
if (sweeprange == 0.0)
sweeprange = DEFAULT_SWEEP_RANGE;
if (sweepdelta == 0.0)
sweepdelta = DEFAULT_SWEEP_DELTA;
if (minbsdelta == 0.0)
minbsdelta = DEFAULT_MINBS_DELTA;
naskew = pixGetLocalSkewAngles(pixs, nslices, redsweep, redsearch,
sweeprange, sweepdelta, minbsdelta,
NULL, NULL, 0);
if (!naskew)
return ERROR_INT("naskew not made", procName, 1);
deg2rad = 3.14159265 / 180.;
w = pixGetWidth(pixs);
h = pixGetHeight(pixs);
ptas = ptaCreate(4);
ptad = ptaCreate(4);
*pptas = ptas;
*pptad = ptad;
/* Find i for skew line that intersects LHS at i and RHS at h / 20 */
for (i = 0; i < h; i++) {
numaGetFValue(naskew, i, &angd);
angr = angd * deg2rad;
dely = w * tan(angr);
if (i - dely > 0.05 * h)
break;
}
ptaAddPt(ptas, 0, i);
ptaAddPt(ptas, w - 1, i - dely);
ptaAddPt(ptad, 0, i);
ptaAddPt(ptad, w - 1, i);
/* Find i for skew line that intersects LHS at i and RHS at 19h / 20 */
for (i = h - 1; i > 0; i--) {
numaGetFValue(naskew, i, &angd);
angr = angd * deg2rad;
dely = w * tan(angr);
if (i - dely < 0.95 * h)
break;
}
ptaAddPt(ptas, 0, i);
ptaAddPt(ptas, w - 1, i - dely);
ptaAddPt(ptad, 0, i);
ptaAddPt(ptad, w - 1, i);
numaDestroy(&naskew);
return 0;
}
/*!
* \brief pixFindBaselines()
*
* \param[in] pixs 1 bpp, 300 ppi
* \param[out] ppta [optional] pairs of pts corresponding to
* approx. ends of each text line
* \param[in] pixadb for debug output; use NULL to skip
* \return na of baseline y values, or NULL on error
*
* <pre>
* Notes:
* (1) Input binary image must have text lines already aligned
* horizontally. This can be done by either rotating the
* image with pixDeskew(), or, if a projective transform
* is required, by doing pixDeskewLocal() first.
* (2) Input null for &pta if you don't want this returned.
* The pta will come in pairs of points (left and right end
* of each baseline).
* (3) Caution: this will not work properly on text with multiple
* columns, where the lines are not aligned between columns.
* If there are multiple columns, they should be extracted
* separately before finding the baselines.
* (4) This function constructs different types of output
* for baselines; namely, a set of raster line values and
* a set of end points of each baseline.
* (5) This function was designed to handle short and long text lines
* without using dangerous thresholds on the peak heights. It does
* this by combining the differential signal with a morphological
* analysis of the locations of the text lines. One can also
* combine this data to normalize the peak heights, by weighting
* the differential signal in the region of each baseline
* by the inverse of the width of the text line found there.
* </pre>
*/
NUMA *
pixFindBaselines(PIX *pixs,
PTA **ppta,
PIXA *pixadb)
{
l_int32 h, i, j, nbox, val1, val2, ndiff, bx, by, bw, bh;
l_int32 imaxloc, peakthresh, zerothresh, inpeak;
l_int32 mintosearch, max, maxloc, nloc, locval;
l_int32 *array;
l_float32 maxval;
BOXA *boxa1, *boxa2, *boxa3;
GPLOT *gplot;
NUMA *nasum, *nadiff, *naloc, *naval;
PIX *pix1, *pix2;
PTA *pta;
PROCNAME("pixFindBaselines");
if (ppta) *ppta = NULL;
if (!pixs || pixGetDepth(pixs) != 1)
return (NUMA *)ERROR_PTR("pixs undefined or not 1 bpp", procName, NULL);
/* Close up the text characters, removing noise */
pix1 = pixMorphSequence(pixs, "c25.1 + e15.1", 0);
/* Estimate the resolution */
if (pixadb) pixaAddPix(pixadb, pixScale(pix1, 0.25, 0.25), L_INSERT);
/* Save the difference of adjacent row sums.
* The high positive-going peaks are the baselines */
if ((nasum = pixCountPixelsByRow(pix1, NULL)) == NULL) {
pixDestroy(&pix1);
return (NUMA *)ERROR_PTR("nasum not made", procName, NULL);
}
h = pixGetHeight(pixs);
nadiff = numaCreate(h);
numaGetIValue(nasum, 0, &val2);
for (i = 0; i < h - 1; i++) {
val1 = val2;
numaGetIValue(nasum, i + 1, &val2);
numaAddNumber(nadiff, val1 - val2);
}
numaDestroy(&nasum);
if (pixadb) { /* show the difference signal */
lept_mkdir("lept/baseline");
gplotSimple1(nadiff, GPLOT_PNG, "/tmp/lept/baseline/diff", "Diff Sig");
pix2 = pixRead("/tmp/lept/baseline/diff.png");
pixaAddPix(pixadb, pix2, L_INSERT);
}
/* Use the zeroes of the profile to locate each baseline. */
array = numaGetIArray(nadiff);
ndiff = numaGetCount(nadiff);
numaGetMax(nadiff, &maxval, &imaxloc);
numaDestroy(&nadiff);
/* Use this to begin locating a new peak: */
peakthresh = (l_int32)maxval / PEAK_THRESHOLD_RATIO;
/* Use this to begin a region between peaks: */
zerothresh = (l_int32)maxval / ZERO_THRESHOLD_RATIO;
naloc = numaCreate(0);
naval = numaCreate(0);
inpeak = FALSE;
for (i = 0; i < ndiff; i++) {
if (inpeak == FALSE) {
if (array[i] > peakthresh) { /* transition to in-peak */
inpeak = TRUE;
mintosearch = i + MIN_DIST_IN_PEAK; /* accept no zeros
* between i and mintosearch */
max = array[i];
maxloc = i;
}
} else { /* inpeak == TRUE; look for max */
if (array[i] > max) {
max = array[i];
maxloc = i;
mintosearch = i + MIN_DIST_IN_PEAK;
} else if (i > mintosearch && array[i] <= zerothresh) { /* leave */
inpeak = FALSE;
numaAddNumber(naval, max);
numaAddNumber(naloc, maxloc);
}
}
}
LEPT_FREE(array);
/* If array[ndiff-1] is max, eg. no descenders, baseline at bottom */
if (inpeak) {
numaAddNumber(naval, max);
numaAddNumber(naloc, maxloc);
}
if (pixadb) { /* show the raster locations for the peaks */
gplot = gplotCreate("/tmp/lept/baseline/loc", GPLOT_PNG, "Peak locs",
"rasterline", "height");
gplotAddPlot(gplot, naloc, naval, GPLOT_POINTS, "locs");
gplotMakeOutput(gplot);
gplotDestroy(&gplot);
pix2 = pixRead("/tmp/lept/baseline/loc.png");
pixaAddPix(pixadb, pix2, L_INSERT);
}
numaDestroy(&naval);
/* Generate an approximate profile of text line width.
* First, filter the boxes of text, where there may be
* more than one box for a given textline. */
pix2 = pixMorphSequence(pix1, "r11 + c20.1 + o30.1 +c1.3", 0);
if (pixadb) pixaAddPix(pixadb, pix2, L_COPY);
boxa1 = pixConnComp(pix2, NULL, 4);
pixDestroy(&pix1);
pixDestroy(&pix2);
if (boxaGetCount(boxa1) == 0) {
numaDestroy(&naloc);
boxaDestroy(&boxa1);
L_INFO("no compnents after filtering\n", procName);
return NULL;
}
boxa2 = boxaTransform(boxa1, 0, 0, 4., 4.);
boxa3 = boxaSort(boxa2, L_SORT_BY_Y, L_SORT_INCREASING, NULL);
boxaDestroy(&boxa1);
boxaDestroy(&boxa2);
/* Optionally, find the baseline segments */
pta = NULL;
if (ppta) {
pta = ptaCreate(0);
*ppta = pta;
}
if (pta) {
nloc = numaGetCount(naloc);
nbox = boxaGetCount(boxa3);
for (i = 0; i < nbox; i++) {
boxaGetBoxGeometry(boxa3, i, &bx, &by, &bw, &bh);
for (j = 0; j < nloc; j++) {
numaGetIValue(naloc, j, &locval);
if (L_ABS(locval - (by + bh)) > 25)
continue;
ptaAddPt(pta, bx, locval);
ptaAddPt(pta, bx + bw, locval);
break;
}
}
}
boxaDestroy(&boxa3);
if (pixadb && pta) { /* display baselines */
l_int32 npts, x1, y1, x2, y2;
pix1 = pixConvertTo32(pixs);
npts = ptaGetCount(pta);
for (i = 0; i < npts; i += 2) {
ptaGetIPt(pta, i, &x1, &y1);
ptaGetIPt(pta, i + 1, &x2, &y2);
pixRenderLineArb(pix1, x1, y1, x2, y2, 2, 255, 0, 0);
}
pixWrite("/tmp/lept/baseline/baselines.png", pix1, IFF_PNG);
pixaAddPix(pixadb, pixScale(pix1, 0.25, 0.25), L_INSERT);
pixDestroy(&pix1);
}
return naloc;
}
/*!
* pixGenerateSelWithRuns()
*
* Input: pix (1 bpp, typically small, to be used as a pattern)
* nhlines (number of hor lines along which elements are found)
* nvlines (number of vert lines along which elements are found)
* distance (min distance from boundary pixel; use 0 for default)
* minlength (min runlength to set hit or miss; use 0 for default)
* toppix (number of extra pixels of bg added above)
* botpix (number of extra pixels of bg added below)
* leftpix (number of extra pixels of bg added to left)
* rightpix (number of extra pixels of bg added to right)
* &pixe (<optional return> input pix expanded by extra pixels)
* Return: sel (hit-miss for input pattern), or null on error
*
* Notes:
* (1) The horizontal and vertical lines along which elements are
* selected are roughly equally spaced. The actual locations of
* the hits and misses are the centers of respective run-lengths.
* (2) No elements are selected that are less than 'distance' pixels away
* from a boundary pixel of the same color. This makes the
* match much more robust to edge noise. Valid inputs of
* 'distance' are 0, 1, 2, 3 and 4. If distance is either 0 or
* greater than 4, we reset it to the default value.
* (3) The 4 numbers for adding rectangles of pixels outside the fg
* can be use if the pattern is expected to be surrounded by bg
* (white) pixels. On the other hand, if the pattern may be near
* other fg (black) components on some sides, use 0 for those sides.
* (4) The pixels added to a side allow you to have miss elements there.
* There is a constraint between distance, minlength, and
* the added pixels for this to work. We illustrate using the
* default values. If you add 5 pixels to the top, and use a
* distance of 1, then you end up with a vertical run of at least
* 4 bg pixels along the top edge of the image. If you use a
* minimum runlength of 3, each vertical line will always find
* a miss near the center of its run. However, if you use a
* minimum runlength of 5, you will not get a miss on every vertical
* line. As another example, if you have 7 added pixels and a
* distance of 2, you can use a runlength up to 5 to guarantee
* that the miss element is recorded. We give a warning if the
* contraint does not guarantee a miss element outside the
* image proper.
* (5) The input pix, as extended by the extra pixels on selected sides,
* can optionally be returned. For debugging, call
* pixDisplayHitMissSel() to visualize the hit-miss sel superimposed
* on the generating bitmap.
*/
SEL *
pixGenerateSelWithRuns(PIX *pixs,
l_int32 nhlines,
l_int32 nvlines,
l_int32 distance,
l_int32 minlength,
l_int32 toppix,
l_int32 botpix,
l_int32 leftpix,
l_int32 rightpix,
PIX **ppixe)
{
l_int32 ws, hs, w, h, x, y, xval, yval, i, j, nh, nm;
l_float32 delh, delw;
NUMA *nah, *nam;
PIX *pixt1, *pixt2, *pixfg, *pixbg;
PTA *ptah, *ptam;
SEL *seld, *sel;
PROCNAME("pixGenerateSelWithRuns");
if (ppixe) *ppixe = NULL;
if (!pixs)
return (SEL *)ERROR_PTR("pixs not defined", procName, NULL);
if (pixGetDepth(pixs) != 1)
return (SEL *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
if (nhlines < 1 && nvlines < 1)
return (SEL *)ERROR_PTR("nvlines and nhlines both < 1", procName, NULL);
if (distance <= 0)
distance = DEFAULT_DISTANCE_TO_BOUNDARY;
if (minlength <= 0)
minlength = DEFAULT_MIN_RUNLENGTH;
if (distance > MAX_DISTANCE_TO_BOUNDARY) {
L_WARNING("distance too large; setting to max value", procName);
distance = MAX_DISTANCE_TO_BOUNDARY;
}
/* Locate the foreground */
pixClipToForeground(pixs, &pixt1, NULL);
if (!pixt1)
return (SEL *)ERROR_PTR("pixt1 not made", procName, NULL);
ws = pixGetWidth(pixt1);
hs = pixGetHeight(pixt1);
w = ws;
h = hs;
/* Crop out a region including the foreground, and add pixels
* on sides depending on the side flags */
if (toppix || botpix || leftpix || rightpix) {
x = y = 0;
if (toppix) {
h += toppix;
y = toppix;
if (toppix < distance + minlength)
L_WARNING("no miss elements in added top pixels", procName);
}
if (botpix) {
h += botpix;
if (botpix < distance + minlength)
L_WARNING("no miss elements in added bot pixels", procName);
}
if (leftpix) {
w += leftpix;
x = leftpix;
if (leftpix < distance + minlength)
L_WARNING("no miss elements in added left pixels", procName);
}
if (rightpix) {
w += rightpix;
if (rightpix < distance + minlength)
L_WARNING("no miss elements in added right pixels", procName);
}
pixt2 = pixCreate(w, h, 1);
pixRasterop(pixt2, x, y, ws, hs, PIX_SRC, pixt1, 0, 0);
}
else
pixt2 = pixClone(pixt1);
if (ppixe)
*ppixe = pixClone(pixt2);
pixDestroy(&pixt1);
/* Identify fg and bg pixels that are at least 'distance' pixels
* away from the boundary pixels in their set */
seld = selCreateBrick(2 * distance + 1, 2 * distance + 1,
distance, distance, SEL_HIT);
pixfg = pixErode(NULL, pixt2, seld);
pixbg = pixDilate(NULL, pixt2, seld);
pixInvert(pixbg, pixbg);
selDestroy(&seld);
pixDestroy(&pixt2);
/* Accumulate hit and miss points */
ptah = ptaCreate(0);
ptam = ptaCreate(0);
if (nhlines >= 1) {
delh = (l_float32)h / (l_float32)(nhlines + 1);
for (i = 0, y = 0; i < nhlines; i++) {
y += (l_int32)(delh + 0.5);
nah = pixGetRunCentersOnLine(pixfg, -1, y, minlength);
nam = pixGetRunCentersOnLine(pixbg, -1, y, minlength);
nh = numaGetCount(nah);
nm = numaGetCount(nam);
for (j = 0; j < nh; j++) {
numaGetIValue(nah, j, &xval);
ptaAddPt(ptah, xval, y);
}
for (j = 0; j < nm; j++) {
numaGetIValue(nam, j, &xval);
ptaAddPt(ptam, xval, y);
}
numaDestroy(&nah);
numaDestroy(&nam);
}
}
if (nvlines >= 1) {
delw = (l_float32)w / (l_float32)(nvlines + 1);
for (i = 0, x = 0; i < nvlines; i++) {
x += (l_int32)(delw + 0.5);
nah = pixGetRunCentersOnLine(pixfg, x, -1, minlength);
nam = pixGetRunCentersOnLine(pixbg, x, -1, minlength);
nh = numaGetCount(nah);
nm = numaGetCount(nam);
for (j = 0; j < nh; j++) {
numaGetIValue(nah, j, &yval);
ptaAddPt(ptah, x, yval);
}
for (j = 0; j < nm; j++) {
numaGetIValue(nam, j, &yval);
ptaAddPt(ptam, x, yval);
}
numaDestroy(&nah);
numaDestroy(&nam);
}
}
/* Make the Sel with those points */
sel = selCreateBrick(h, w, h / 2, w / 2, SEL_DONT_CARE);
nh = ptaGetCount(ptah);
for (i = 0; i < nh; i++) {
ptaGetIPt(ptah, i, &x, &y);
selSetElement(sel, y, x, SEL_HIT);
}
nm = ptaGetCount(ptam);
for (i = 0; i < nm; i++) {
ptaGetIPt(ptam, i, &x, &y);
selSetElement(sel, y, x, SEL_MISS);
}
pixDestroy(&pixfg);
pixDestroy(&pixbg);
ptaDestroy(&ptah);
ptaDestroy(&ptam);
return sel;
}
/*!
* pixSearchBinaryMaze()
*
* Input: pixs (1 bpp, maze)
* xi, yi (beginning point; use same initial point
* that was used to generate the maze)
* xf, yf (end point, or close to it)
* &ppixd (<optional return> maze with path illustrated, or
* if no path possible, the part of the maze
* that was searched)
* Return: pta (shortest path), or null if either no path
* exists or on error
*
* Notes:
* (1) Because of the overhead in calling pixGetPixel() and
* pixSetPixel(), we have used raster line pointers and the
* GET_DATA* and SET_DATA* macros for many of the pix accesses.
* (2) Commentary:
* The goal is to find the shortest path between beginning and
* end points, without going through walls, and there are many
* ways to solve this problem.
* We use a queue to implement a breadth-first search. Two auxiliary
* "image" data structures can be used: one to mark the visited
* pixels and one to give the direction to the parent for each
* visited pixels. The first structure is used to avoid putting
* pixels on the queue more than once, and the second is used
* for retracing back to the origin, like the breadcrumbs in
* Hansel and Gretel. Each pixel taken off the queue is destroyed
* after it is used to locate the allowed neighbors. In fact,
* only one distance image is required, if you initialize it
* to some value that signifies "not yet visited." (We use
* a binary image for marking visited pixels because it is clearer.)
* This method for a simple search of a binary maze is implemented in
* searchBinaryMaze().
* An alternative method would store the (manhattan) distance
* from the start point with each pixel on the queue. The children
* of each pixel get a distance one larger than the parent. These
* values can be stored in an auxiliary distance map image
* that is constructed simultaneously with the search. Once the
* end point is reached, the distance map is used to backtrack
* along a minimum path. There may be several equal length
* minimum paths, any one of which can be chosen this way.
*/
PTA *
pixSearchBinaryMaze(PIX *pixs,
l_int32 xi,
l_int32 yi,
l_int32 xf,
l_int32 yf,
PIX **ppixd)
{
l_int32 i, j, x, y, w, h, d, found;
l_uint32 val, rpixel, gpixel, bpixel;
void **lines1, **linem1, **linep8, **lined32;
MAZEEL *el, *elp;
PIX *pixd; /* the shortest path written on the maze image */
PIX *pixm; /* for bookkeeping, to indicate pixels already visited */
PIX *pixp; /* for bookkeeping, to indicate direction to parent */
L_QUEUE *lq;
PTA *pta;
PROCNAME("pixSearchBinaryMaze");
if (ppixd) *ppixd = NULL;
if (!pixs)
return (PTA *)ERROR_PTR("pixs not defined", procName, NULL);
pixGetDimensions(pixs, &w, &h, &d);
if (d != 1)
return (PTA *)ERROR_PTR("pixs not 1 bpp", procName, NULL);
if (xi <= 0 || xi >= w)
return (PTA *)ERROR_PTR("xi not valid", procName, NULL);
if (yi <= 0 || yi >= h)
return (PTA *)ERROR_PTR("yi not valid", procName, NULL);
pixGetPixel(pixs, xi, yi, &val);
if (val != 0)
return (PTA *)ERROR_PTR("(xi,yi) not bg pixel", procName, NULL);
pixd = NULL;
pta = NULL;
/* Find a bg pixel near input point (xf, yf) */
localSearchForBackground(pixs, &xf, &yf, 5);
#if DEBUG_MAZE
fprintf(stderr, "(xi, yi) = (%d, %d), (xf, yf) = (%d, %d)\n",
xi, yi, xf, yf);
#endif /* DEBUG_MAZE */
pixm = pixCreate(w, h, 1); /* initialized to OFF */
pixp = pixCreate(w, h, 8); /* direction to parent stored as enum val */
lines1 = pixGetLinePtrs(pixs, NULL);
linem1 = pixGetLinePtrs(pixm, NULL);
linep8 = pixGetLinePtrs(pixp, NULL);
lq = lqueueCreate(0);
/* Prime the queue with the first pixel; it is OFF */
el = mazeelCreate(xi, yi, 0); /* don't need direction here */
pixSetPixel(pixm, xi, yi, 1); /* mark visited */
lqueueAdd(lq, el);
/* Fill up the pix storing directions to parents,
* stopping when we hit the point (xf, yf) */
found = FALSE;
while (lqueueGetCount(lq) > 0) {
elp = (MAZEEL *)lqueueRemove(lq);
x = elp->x;
y = elp->y;
if (x == xf && y == yf) {
found = TRUE;
FREE(elp);
break;
}
if (x > 0) { /* check to west */
val = GET_DATA_BIT(linem1[y], x - 1);
if (val == 0) { /* not yet visited */
SET_DATA_BIT(linem1[y], x - 1); /* mark visited */
val = GET_DATA_BIT(lines1[y], x - 1);
if (val == 0) { /* bg, not a wall */
SET_DATA_BYTE(linep8[y], x - 1, DIR_EAST); /* parent E */
el = mazeelCreate(x - 1, y, 0);
lqueueAdd(lq, el);
}
}
}
if (y > 0) { /* check north */
val = GET_DATA_BIT(linem1[y - 1], x);
if (val == 0) { /* not yet visited */
SET_DATA_BIT(linem1[y - 1], x); /* mark visited */
val = GET_DATA_BIT(lines1[y - 1], x);
if (val == 0) { /* bg, not a wall */
SET_DATA_BYTE(linep8[y - 1], x, DIR_SOUTH); /* parent S */
el = mazeelCreate(x, y - 1, 0);
lqueueAdd(lq, el);
}
}
}
if (x < w - 1) { /* check east */
val = GET_DATA_BIT(linem1[y], x + 1);
if (val == 0) { /* not yet visited */
SET_DATA_BIT(linem1[y], x + 1); /* mark visited */
val = GET_DATA_BIT(lines1[y], x + 1);
if (val == 0) { /* bg, not a wall */
SET_DATA_BYTE(linep8[y], x + 1, DIR_WEST); /* parent W */
el = mazeelCreate(x + 1, y, 0);
lqueueAdd(lq, el);
}
}
}
if (y < h - 1) { /* check south */
val = GET_DATA_BIT(linem1[y + 1], x);
if (val == 0) { /* not yet visited */
SET_DATA_BIT(linem1[y + 1], x); /* mark visited */
val = GET_DATA_BIT(lines1[y + 1], x);
if (val == 0) { /* bg, not a wall */
SET_DATA_BYTE(linep8[y + 1], x, DIR_NORTH); /* parent N */
el = mazeelCreate(x, y + 1, 0);
lqueueAdd(lq, el);
}
}
}
FREE(elp);
}
lqueueDestroy(&lq, TRUE);
pixDestroy(&pixm);
FREE(linem1);
if (ppixd) {
pixd = pixUnpackBinary(pixs, 32, 1);
*ppixd = pixd;
}
composeRGBPixel(255, 0, 0, &rpixel); /* start point */
composeRGBPixel(0, 255, 0, &gpixel);
composeRGBPixel(0, 0, 255, &bpixel); /* end point */
if (!found) {
L_INFO(" No path found", procName);
if (pixd) { /* paint all visited locations */
lined32 = pixGetLinePtrs(pixd, NULL);
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
val = GET_DATA_BYTE(linep8[i], j);
if (val != 0 && pixd)
SET_DATA_FOUR_BYTES(lined32[i], j, gpixel);
}
}
FREE(lined32);
}
}
else { /* write path onto pixd */
L_INFO(" Path found", procName);
pta = ptaCreate(0);
x = xf;
y = yf;
while (1) {
ptaAddPt(pta, x, y);
if (x == xi && y == yi)
break;
if (pixd)
pixSetPixel(pixd, x, y, gpixel);
pixGetPixel(pixp, x, y, &val);
if (val == DIR_NORTH)
y--;
else if (val == DIR_SOUTH)
y++;
else if (val == DIR_EAST)
x++;
else if (val == DIR_WEST)
x--;
}
}
if (pixd) {
pixSetPixel(pixd, xi, yi, rpixel);
pixSetPixel(pixd, xf, yf, bpixel);
}
pixDestroy(&pixp);
FREE(lines1);
FREE(linep8);
return pta;
}
/*!
* pixSearchGrayMaze()
*
* Input: pixs (1 bpp, maze)
* xi, yi (beginning point; use same initial point
* that was used to generate the maze)
* xf, yf (end point, or close to it)
* &ppixd (<optional return> maze with path illustrated, or
* if no path possible, the part of the maze
* that was searched)
* Return: pta (shortest path), or null if either no path
* exists or on error
*
* Commentary:
* Consider first a slight generalization of the binary maze
* search problem. Suppose that you can go through walls,
* but the cost is higher (say, an increment of 3 to go into
* a wall pixel rather than 1)? You're still trying to find
* the shortest path. One way to do this is with an ordered
* queue, and a simple way to visualize an ordered queue is as
* a set of stacks, each stack being marked with the distance
* of each pixel in the stack from the start. We place the
* start pixel in stack 0, pop it, and process its 4 children.
* Each pixel is given a distance that is incremented from that
* of its parent (0 in this case), depending on if it is a wall
* pixel or not. That value may be recorded on a distance map,
* according to the algorithm below. For children of the first
* pixel, those not on a wall go in stack 1, and wall
* children go in stack 3. Stack 0 being emptied, the process
* then continues with pixels being popped from stack 1.
* Here is the algorithm for each child pixel. The pixel's
* distance value, were it to be placed on a stack, is compared
* with the value for it that is on the distance map. There
* are three possible cases:
* (1) If the pixel has not yet been registered, it is pushed
* on its stack and the distance is written to the map.
* (2) If it has previously been registered with a higher distance,
* the distance on the map is relaxed to that of the
* current pixel, which is then placed on its stack.
* (3) If it has previously been registered with an equal
* or lower value, the pixel is discarded.
* The pixels are popped and processed successively from
* stack 1, and when stack 1 is empty, popping starts on stack 2.
* This continues until the destination pixel is popped off
* a stack. The minimum path is then derived from the distance map,
* going back from the end point as before. This is just Dijkstra's
* algorithm for a directed graph; here, the underlying graph
* (consisting of the pixels and four edges connecting each pixel
* to its 4-neighbor) is a special case of a directed graph, where
* each edge is bi-directional. The implementation of this generalized
* maze search is left as an exercise to the reader.
*
* Let's generalize a bit further. Suppose the "maze" is just
* a grayscale image -- think of it as an elevation map. The cost
* of moving on this surface depends on the height, or the gradient,
* or whatever you want. All that is required is that the cost
* is specified and non-negative on each link between adjacent
* pixels. Now the problem becomes: find the least cost path
* moving on this surface between two specified end points.
* For example, if the cost across an edge between two pixels
* depends on the "gradient", you can use:
* cost = 1 + L_ABS(deltaV)
* where deltaV is the difference in value between two adjacent
* pixels. If the costs are all integers, we can still use an array
* of stacks to avoid ordering the queue (e.g., by using a heap sort.)
* This is a neat problem, because you don't even have to build a
* maze -- you can can use it on any grayscale image!
*
* Rather than using an array of stacks, a more practical
* approach is to implement with a priority queue, which is
* a queue that is sorted so that the elements with the largest
* (or smallest) key values always come off first. The
* priority queue is efficiently implemented as a heap, and
* this is how we do it. Suppose you run the algorithm
* using a priority queue, doing the bookkeeping with an
* auxiliary image data structure that saves the distance of
* each pixel put on the queue as before, according to the method
* described above. We implement it as a 2-way choice by
* initializing the distance array to a large value and putting
* a pixel on the queue if its distance is less than the value
* found on the array. When you finally pop the end pixel from
* the queue, you're done, and you can trace the path backward,
* either always going downhill or using an auxiliary image to
* give you the direction to go at each step. This is implemented
* here in searchGrayMaze().
*
* Do we really have to use a sorted queue? Can we solve this
* generalized maze with an unsorted queue of pixels? (Or even
* an unsorted stack, doing a depth-first search (DFS)?)
* Consider a different algorithm for this generalized maze, where
* we travel again breadth first, but this time use a single,
* unsorted queue. An auxiliary image is used as before to
* store the distances and to determine if pixels get pushed
* on the stack or dropped. As before, we must allow pixels
* to be revisited, with relaxation of the distance if a shorter
* path arrives later. As a result, we will in general have
* multiple instances of the same pixel on the stack with different
* distances. However, because the queue is not ordered, some of
* these pixels will be popped when another instance with a lower
* distance is still on the stack. Here, we're just popping them
* in the order they go on, rather than setting up a priority
* based on minimum distance. Thus, unlike the priority queue,
* when a pixel is popped we have to check the distance map to
* see if a pixel with a lower distance has been put on the queue,
* and, if so, we discard the pixel we just popped. So the
* "while" loop looks like this:
* - pop a pixel from the queue
* - check its distance against the distance stored in the
* distance map; if larger, discard
* - otherwise, for each of its neighbors:
* - compute its distance from the start pixel
* - compare this distance with that on the distance map:
* - if the distance map value higher, relax the distance
* and push the pixel on the queue
* - if the distance map value is lower, discard the pixel
*
* How does this loop terminate? Before, with an ordered queue,
* it terminates when you pop the end pixel. But with an unordered
* queue (or stack), the first time you hit the end pixel, the
* distance is not guaranteed to be correct, because the pixels
* along the shortest path may not have yet been visited and relaxed.
* Because the shortest path can theoretically go anywhere,
* we must keep going. How do we know when to stop? Dijkstra
* uses an ordered queue to systematically remove nodes from
* further consideration. (Each time a pixel is popped, we're
* done with it; it's "finalized" in the Dijkstra sense because
* we know the shortest path to it.) However, with an unordered
* queue, the brute force answer is: stop when the queue
* (or stack) is empty, because then every pixel in the image
* has been assigned its minimum "distance" from the start pixel.
*
* This is similar to the situation when you use a stack for the
* simpler uniform-step problem: with breadth-first search (BFS)
* the pixels on the queue are automatically ordered, so you are
* done when you locate the end pixel as a neighbor of a popped pixel;
* whereas depth-first search (DFS), using a stack, requires,
* in general, a search of every accessible pixel. Further, if
* a pixel is revisited with a smaller distance, that distance is
* recorded and the pixel is put on the stack again.
*
* But surely, you ask, can't we stop sooner? What if the
* start and end pixels are very close to each other?
* OK, suppose they are, and you have very high walls and a
* long snaking level path that is actually the minimum cost.
* That long path can wind back and forth across the entire
* maze many times before ending up at the end point, which
* could be just over a wall from the start. With the unordered
* queue, you very quickly get a high distance for the end
* pixel, which will be relaxed to the minimum distance only
* after all the pixels of the path have been visited and placed
* on the queue, multiple times for many of them. So that's the
* price for not ordering the queue!
*/
PTA *
pixSearchGrayMaze(PIX *pixs,
l_int32 xi,
l_int32 yi,
l_int32 xf,
l_int32 yf,
PIX **ppixd)
{
l_int32 x, y, w, h, d;
l_uint32 val, valr, vals, rpixel, gpixel, bpixel;
void **lines8, **liner32, **linep8;
l_int32 cost, dist, distparent, sival, sivals;
MAZEEL *el, *elp;
PIX *pixd; /* optionally plot the path on this RGB version of pixs */
PIX *pixr; /* for bookkeeping, to indicate the minimum distance */
/* to pixels already visited */
PIX *pixp; /* for bookkeeping, to indicate direction to parent */
L_HEAP *lh;
PTA *pta;
PROCNAME("pixSearchGrayMaze");
if (ppixd) *ppixd = NULL;
if (!pixs)
return (PTA *)ERROR_PTR("pixs not defined", procName, NULL);
pixGetDimensions(pixs, &w, &h, &d);
if (d != 8)
return (PTA *)ERROR_PTR("pixs not 8 bpp", procName, NULL);
if (xi <= 0 || xi >= w)
return (PTA *)ERROR_PTR("xi not valid", procName, NULL);
if (yi <= 0 || yi >= h)
return (PTA *)ERROR_PTR("yi not valid", procName, NULL);
pixd = NULL;
pta = NULL;
pixr = pixCreate(w, h, 32);
pixSetAll(pixr); /* initialize to max value */
pixp = pixCreate(w, h, 8); /* direction to parent stored as enum val */
lines8 = pixGetLinePtrs(pixs, NULL);
linep8 = pixGetLinePtrs(pixp, NULL);
liner32 = pixGetLinePtrs(pixr, NULL);
lh = lheapCreate(0, L_SORT_INCREASING); /* always remove closest pixels */
/* Prime the heap with the first pixel */
pixGetPixel(pixs, xi, yi, &val);
el = mazeelCreate(xi, yi, 0); /* don't need direction here */
el->distance = 0;
pixGetPixel(pixs, xi, yi, &val);
el->val = val;
pixSetPixel(pixr, xi, yi, 0); /* distance is 0 */
lheapAdd(lh, el);
/* Breadth-first search with priority queue (implemented by
a heap), labeling direction to parents in pixp and minimum
distance to visited pixels in pixr. Stop when we pull
the destination point (xf, yf) off the queue. */
while (lheapGetCount(lh) > 0) {
elp = (MAZEEL *)lheapRemove(lh);
if (!elp)
return (PTA *)ERROR_PTR("heap broken!!", procName, NULL);
x = elp->x;
y = elp->y;
if (x == xf && y == yf) { /* exit condition */
FREE(elp);
break;
}
distparent = (l_int32)elp->distance;
val = elp->val;
sival = val;
if (x > 0) { /* check to west */
vals = GET_DATA_BYTE(lines8[y], x - 1);
valr = GET_DATA_FOUR_BYTES(liner32[y], x - 1);
sivals = (l_int32)vals;
cost = 1 + L_ABS(sivals - sival); /* cost to move to this pixel */
dist = distparent + cost;
if (dist < valr) { /* shortest path so far to this pixel */
SET_DATA_FOUR_BYTES(liner32[y], x - 1, dist); /* new dist */
SET_DATA_BYTE(linep8[y], x - 1, DIR_EAST); /* parent to E */
el = mazeelCreate(x - 1, y, 0);
el->val = vals;
el->distance = dist;
lheapAdd(lh, el);
}
}
if (y > 0) { /* check north */
vals = GET_DATA_BYTE(lines8[y - 1], x);
valr = GET_DATA_FOUR_BYTES(liner32[y - 1], x);
sivals = (l_int32)vals;
cost = 1 + L_ABS(sivals - sival); /* cost to move to this pixel */
dist = distparent + cost;
if (dist < valr) { /* shortest path so far to this pixel */
SET_DATA_FOUR_BYTES(liner32[y - 1], x, dist); /* new dist */
SET_DATA_BYTE(linep8[y - 1], x, DIR_SOUTH); /* parent to S */
el = mazeelCreate(x, y - 1, 0);
el->val = vals;
el->distance = dist;
lheapAdd(lh, el);
}
}
if (x < w - 1) { /* check east */
vals = GET_DATA_BYTE(lines8[y], x + 1);
valr = GET_DATA_FOUR_BYTES(liner32[y], x + 1);
sivals = (l_int32)vals;
cost = 1 + L_ABS(sivals - sival); /* cost to move to this pixel */
dist = distparent + cost;
if (dist < valr) { /* shortest path so far to this pixel */
SET_DATA_FOUR_BYTES(liner32[y], x + 1, dist); /* new dist */
SET_DATA_BYTE(linep8[y], x + 1, DIR_WEST); /* parent to W */
el = mazeelCreate(x + 1, y, 0);
el->val = vals;
el->distance = dist;
lheapAdd(lh, el);
}
}
if (y < h - 1) { /* check south */
vals = GET_DATA_BYTE(lines8[y + 1], x);
valr = GET_DATA_FOUR_BYTES(liner32[y + 1], x);
sivals = (l_int32)vals;
cost = 1 + L_ABS(sivals - sival); /* cost to move to this pixel */
dist = distparent + cost;
if (dist < valr) { /* shortest path so far to this pixel */
SET_DATA_FOUR_BYTES(liner32[y + 1], x, dist); /* new dist */
SET_DATA_BYTE(linep8[y + 1], x, DIR_NORTH); /* parent to N */
el = mazeelCreate(x, y + 1, 0);
el->val = vals;
el->distance = dist;
lheapAdd(lh, el);
}
}
FREE(elp);
}
lheapDestroy(&lh, TRUE);
if (ppixd) {
pixd = pixConvert8To32(pixs);
*ppixd = pixd;
}
composeRGBPixel(255, 0, 0, &rpixel); /* start point */
composeRGBPixel(0, 255, 0, &gpixel);
composeRGBPixel(0, 0, 255, &bpixel); /* end point */
x = xf;
y = yf;
pta = ptaCreate(0);
while (1) { /* write path onto pixd */
ptaAddPt(pta, x, y);
if (x == xi && y == yi)
break;
if (pixd)
pixSetPixel(pixd, x, y, gpixel);
pixGetPixel(pixp, x, y, &val);
if (val == DIR_NORTH)
y--;
else if (val == DIR_SOUTH)
y++;
else if (val == DIR_EAST)
x++;
else if (val == DIR_WEST)
x--;
pixGetPixel(pixr, x, y, &val);
#if DEBUG_PATH
fprintf(stderr, "(x,y) = (%d, %d); dist = %d\n", x, y, val);
#endif /* DEBUG_PATH */
}
if (pixd) {
pixSetPixel(pixd, xi, yi, rpixel);
pixSetPixel(pixd, xf, yf, bpixel);
}
pixDestroy(&pixp);
pixDestroy(&pixr);
FREE(lines8);
FREE(linep8);
FREE(liner32);
return pta;
}
/*!
* dewarpBuildModel()
*
* Input: dew
* debugflag (1 for debugging output)
* Return: 0 if OK, 1 on error
*
* Notes:
* (1) This is the basic function that builds the vertical
* disparity array, which allows determination of the
* src pixel in the input image corresponding to each
* dest pixel in the dewarped image.
* (2) The method is as follows:
* * Estimate the centers of all the long textlines and
* fit a LS quadratic to each one. This smooths the curves.
* * Sample each curve at a regular interval, find the y-value
* of the flat point on each curve, and subtract the sampled
* curve value from this value. This is the vertical
* disparity.
* * Fit a LS quadratic to each set of vertically aligned
* disparity samples. This smooths the disparity values
* in the vertical direction. Then resample at the same
* regular interval, We now have a regular grid of smoothed
* vertical disparity valuels.
* * Interpolate this grid to get a full resolution disparity
* map. This can be applied directly to the src image
* pixels to dewarp the image in the vertical direction,
* making all textlines horizontal.
*/
l_int32
dewarpBuildModel(L_DEWARP *dew,
l_int32 debugflag)
{
char *tempname;
l_int32 i, j, nlines, nx, ny, sampling;
l_float32 c0, c1, c2, x, y, flaty, val;
l_float32 *faflats;
NUMA *nax, *nafit, *nacurve, *nacurves, *naflat, *naflats, *naflatsi;
PIX *pixs, *pixt1, *pixt2;
PTA *pta, *ptad;
PTAA *ptaa1, *ptaa2, *ptaa3, *ptaa4, *ptaa5, *ptaa6, *ptaa7;
FPIX *fpix1, *fpix2, *fpix3;
PROCNAME("dewarpBuildModel");
if (!dew)
return ERROR_INT("dew not defined", procName, 1);
pixs = dew->pixs;
if (debugflag) {
pixDisplayWithTitle(pixs, 0, 0, "pixs", 1);
pixWriteTempfile("/tmp", "pixs.png", pixs, IFF_PNG, NULL);
}
/* Make initial estimate of centers of textlines */
ptaa1 = pixGetTextlineCenters(pixs, DEBUG_TEXTLINE_CENTERS);
if (debugflag) {
pixt1 = pixConvertTo32(pixs);
pixt2 = pixDisplayPtaa(pixt1, ptaa1);
pixWriteTempfile("/tmp", "lines1.png", pixt2, IFF_PNG, NULL);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
}
/* Remove all lines that are not near the length
* of the longest line. */
ptaa2 = ptaaRemoveShortLines(pixs, ptaa1, 0.8, DEBUG_SHORT_LINES);
if (debugflag) {
pixt1 = pixConvertTo32(pixs);
pixt2 = pixDisplayPtaa(pixt1, ptaa2);
pixWriteTempfile("/tmp", "lines2.png", pixt2, IFF_PNG, NULL);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
}
nlines = ptaaGetCount(ptaa2);
if (nlines < dew->minlines)
return ERROR_INT("insufficient lines to build model", procName, 1);
/* Do quadratic fit to smooth each line. A single quadratic
* over the entire width of the line appears to be sufficient.
* Quartics tend to overfit to noise. Each line is thus
* represented by three coefficients: c2 * x^2 + c1 * x + c0.
* Using the coefficients, sample each fitted curve uniformly
* across the full width of the image. */
sampling = dew->sampling;
nx = dew->nx;
ny = dew->ny;
ptaa3 = ptaaCreate(nlines);
nacurve = numaCreate(nlines); /* stores curvature coeff c2 */
for (i = 0; i < nlines; i++) { /* for each line */
pta = ptaaGetPta(ptaa2, i, L_CLONE);
ptaGetQuadraticLSF(pta, &c2, &c1, &c0, NULL);
numaAddNumber(nacurve, c2);
ptad = ptaCreate(nx);
for (j = 0; j < nx; j++) { /* uniformly sampled in x */
x = j * sampling;
applyQuadraticFit(c2, c1, c0, x, &y);
ptaAddPt(ptad, x, y);
}
ptaaAddPta(ptaa3, ptad, L_INSERT);
ptaDestroy(&pta);
}
if (debugflag) {
ptaa4 = ptaaCreate(nlines);
for (i = 0; i < nlines; i++) {
pta = ptaaGetPta(ptaa2, i, L_CLONE);
ptaGetArrays(pta, &nax, NULL);
ptaGetQuadraticLSF(pta, NULL, NULL, NULL, &nafit);
ptad = ptaCreateFromNuma(nax, nafit);
ptaaAddPta(ptaa4, ptad, L_INSERT);
ptaDestroy(&pta);
numaDestroy(&nax);
numaDestroy(&nafit);
}
pixt1 = pixConvertTo32(pixs);
pixt2 = pixDisplayPtaa(pixt1, ptaa4);
pixWriteTempfile("/tmp", "lines3.png", pixt2, IFF_PNG, NULL);
pixDestroy(&pixt1);
pixDestroy(&pixt2);
ptaaDestroy(&ptaa4);
}
/* Find and save the flat points in each curve. */
naflat = numaCreate(nlines);
for (i = 0; i < nlines; i++) {
pta = ptaaGetPta(ptaa3, i, L_CLONE);
numaGetFValue(nacurve, i, &c2);
if (c2 <= 0) /* flat point at bottom; max value of y in curve */
ptaGetRange(pta, NULL, NULL, NULL, &flaty);
else /* flat point at top; min value of y in curve */
ptaGetRange(pta, NULL, NULL, &flaty, NULL);
numaAddNumber(naflat, flaty);
ptaDestroy(&pta);
}
/* Sort the lines in ptaa3 by their position */
naflatsi = numaGetSortIndex(naflat, L_SORT_INCREASING);
naflats = numaSortByIndex(naflat, naflatsi);
nacurves = numaSortByIndex(nacurve, naflatsi);
dew->naflats = naflats;
dew->nacurves = nacurves;
ptaa4 = ptaaSortByIndex(ptaa3, naflatsi);
numaDestroy(&naflat);
numaDestroy(&nacurve);
numaDestroy(&naflatsi);
if (debugflag) {
tempname = genTempFilename("/tmp", "naflats.na", 0);
numaWrite(tempname, naflats);
FREE(tempname);
}
/* Convert the sampled points in ptaa3 to a sampled disparity with
* with respect to the flat point in the curve. */
ptaa5 = ptaaCreate(nlines);
for (i = 0; i < nlines; i++) {
pta = ptaaGetPta(ptaa4, i, L_CLONE);
numaGetFValue(naflats, i, &flaty);
ptad = ptaCreate(nx);
for (j = 0; j < nx; j++) {
ptaGetPt(pta, j, &x, &y);
ptaAddPt(ptad, x, flaty - y);
}
ptaaAddPta(ptaa5, ptad, L_INSERT);
ptaDestroy(&pta);
}
if (debugflag) {
tempname = genTempFilename("/tmp", "ptaa5.ptaa", 0);
ptaaWrite(tempname, ptaa5, 0);
FREE(tempname);
}
/* Generate a ptaa taking vertical 'columns' from ptaa5.
* We want to fit the vertical disparity on the column to the
* vertical position of the line, which we call 'y' here and
* obtain from naflats. */
ptaa6 = ptaaCreate(nx);
faflats = numaGetFArray(naflats, L_NOCOPY);
for (j = 0; j < nx; j++) {
pta = ptaCreate(nlines);
for (i = 0; i < nlines; i++) {
y = faflats[i];
ptaaGetPt(ptaa5, i, j, NULL, &val); /* disparity value */
ptaAddPt(pta, y, val);
}
ptaaAddPta(ptaa6, pta, L_INSERT);
}
if (debugflag) {
tempname = genTempFilename("/tmp", "ptaa6.ptaa", 0);
ptaaWrite(tempname, ptaa6, 0);
FREE(tempname);
}
/* Do quadratic fit vertically on a subset of pixel columns
* for the vertical displacement, which identifies the
* src pixel(s) for each dest pixel. Sample the displacement
* on a regular grid in the vertical direction. */
ptaa7 = ptaaCreate(nx); /* uniformly sampled across full height of image */
for (j = 0; j < nx; j++) { /* for each column */
pta = ptaaGetPta(ptaa6, j, L_CLONE);
ptaGetQuadraticLSF(pta, &c2, &c1, &c0, NULL);
ptad = ptaCreate(ny);
for (i = 0; i < ny; i++) { /* uniformly sampled in y */
y = i * sampling;
applyQuadraticFit(c2, c1, c0, y, &val);
ptaAddPt(ptad, y, val);
}
ptaaAddPta(ptaa7, ptad, L_INSERT);
ptaDestroy(&pta);
}
if (debugflag) {
tempname = genTempFilename("/tmp", "ptaa7.ptaa", 0);
ptaaWrite(tempname, ptaa7, 0);
FREE(tempname);
}
/* Save the result in a fpix at the specified subsampling */
fpix1 = fpixCreate(nx, ny);
for (i = 0; i < ny; i++) {
for (j = 0; j < nx; j++) {
ptaaGetPt(ptaa7, j, i, NULL, &val);
fpixSetPixel(fpix1, j, i, val);
}
}
dew->sampvdispar = fpix1;
/* Generate a full res fpix for vertical dewarping. We require that
* the size of this fpix is at least as big as the input image. */
fpix2 = fpixScaleByInteger(fpix1, sampling);
dew->fullvdispar = fpix2;
if (debugflag) {
pixt1 = fpixRenderContours(fpix2, -2., 2.0, 0.2);
pixWriteTempfile("/tmp", "vert-contours.png", pixt1, IFF_PNG, NULL);
pixDisplay(pixt1, 1000, 0);
pixDestroy(&pixt1);
}
/* Generate full res and sampled fpix for horizontal dewarping. This
* works to the extent that the line curvature is due to bending
* out of the plane normal to the camera, and not wide-angle
* "fishbowl" distortion. Also generate the sampled horizontal
* disparity array. */
if (dew->applyhoriz) {
fpix3 = fpixBuildHorizontalDisparity(fpix2, 0, &dew->extraw);
dew->fullhdispar = fpix3;
dew->samphdispar = fpixSampledDisparity(fpix3, dew->sampling);
if (debugflag) {
pixt1 = fpixRenderContours(fpix3, -2., 2.0, 0.2);
pixWriteTempfile("/tmp", "horiz-contours.png", pixt1,
IFF_PNG, NULL);
pixDisplay(pixt1, 1000, 0);
pixDestroy(&pixt1);
}
}
dew->success = 1;
ptaaDestroy(&ptaa1);
ptaaDestroy(&ptaa2);
ptaaDestroy(&ptaa3);
ptaaDestroy(&ptaa4);
ptaaDestroy(&ptaa5);
ptaaDestroy(&ptaa6);
ptaaDestroy(&ptaa7);
return 0;
}