int main(int argc,
char **argv)
{
l_int32 i, n, w, h;
BOXA *boxa;
NUMA *naindex, *naw, *nah, *naw_med, *nah_med;
PIX *pixs, *pixt;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
/* Generate arrays of word widths and heights */
pixs = pixRead("feyn.tif");
pixGetWordBoxesInTextlines(pixs, 1, 6, 6, 500, 50, &boxa, &naindex);
n = boxaGetCount(boxa);
naw = numaCreate(0);
nah = numaCreate(0);
for (i = 0; i < n; i++) {
boxaGetBoxGeometry(boxa, i, NULL, NULL, &w, &h);
numaAddNumber(naw, w);
numaAddNumber(nah, h);
}
boxaDestroy(&boxa);
numaDestroy(&naindex);
/* Make the rank bin arrays of median values, with 10 bins */
lept_rmfile("/tmp/lept/regout/w_10bin.png"); /* remove existing ones */
lept_rmfile("/tmp/lept/regout/h_10bin.png");
lept_rmfile("/tmp/lept/regout/w_30bin.png");
lept_rmfile("/tmp/lept/regout/h_30bin.png");
numaGetRankBinValues(naw, 10, NULL, &naw_med);
numaGetRankBinValues(nah, 10, NULL, &nah_med);
gplotSimple1(naw_med, GPLOT_PNG, "/tmp/lept/regout/w_10bin",
"width vs rank bins (10)");
gplotSimple1(nah_med, GPLOT_PNG, "/tmp/lept/regout/h_10bin",
"height vs rank bins (10)");
numaDestroy(&naw_med);
numaDestroy(&nah_med);
/* Make the rank bin arrays of median values, with 30 bins */
numaGetRankBinValues(naw, 30, NULL, &naw_med);
numaGetRankBinValues(nah, 30, NULL, &nah_med);
gplotSimple1(naw_med, GPLOT_PNG, "/tmp/lept/regout/w_30bin",
"width vs rank bins (30)");
gplotSimple1(nah_med, GPLOT_PNG, "/tmp/lept/regout/h_30bin",
"height vs rank bins (30)");
numaDestroy(&naw_med);
numaDestroy(&nah_med);
/* Save as golden files, or check against them */
regTestCheckFile(rp, "/tmp/lept/regout/w_10bin.png"); /* 0 */
regTestCheckFile(rp, "/tmp/lept/regout/h_10bin.png"); /* 1 */
regTestCheckFile(rp, "/tmp/lept/regout/w_30bin.png"); /* 2 */
regTestCheckFile(rp, "/tmp/lept/regout/h_30bin.png"); /* 3 */
/* Display results for debugging */
pixt = pixRead("/tmp/lept/regout/w_10bin.png");
pixDisplayWithTitle(pixt, 0, 0, NULL, rp->display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/lept/regout/h_10bin.png");
pixDisplayWithTitle(pixt, 650, 0, NULL, rp->display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/lept/regout/w_30bin.png");
pixDisplayWithTitle(pixt, 0, 550, NULL, rp->display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/lept/regout/h_30bin.png");
pixDisplayWithTitle(pixt, 650, 550, NULL, rp->display);
pixDestroy(&pixt);
pixDestroy(&pixs);
numaDestroy(&naw);
numaDestroy(&nah);
return regTestCleanup(rp);
}
main(int argc,
char **argv)
{
l_int32 i, n, w, h, success, display;
FILE *fp;
BOXA *boxa;
NUMA *naindex, *naw, *nah, *naw_med, *nah_med;
PIX *pixs, *pixt, *pixd;
if (regTestSetup(argc, argv, &fp, &display, &success, NULL))
return 1;
/* Generate arrays of word widths and heights */
pixs = pixRead("feyn.tif");
pixGetWordBoxesInTextlines(pixs, 1, 6, 6, 500, 50, &boxa, &naindex);
n = boxaGetCount(boxa);
naw = numaCreate(0);
nah = numaCreate(0);
for (i = 0; i < n; i++) {
boxaGetBoxGeometry(boxa, i, NULL, NULL, &w, &h);
numaAddNumber(naw, w);
numaAddNumber(nah, h);
}
boxaDestroy(&boxa);
numaDestroy(&naindex);
/* Make the rank bin arrays of median values, with 10 bins */
numaGetRankBinValues(naw, 10, NULL, &naw_med);
numaGetRankBinValues(nah, 10, NULL, &nah_med);
gplotSimple1(naw_med, GPLOT_PNG, "/tmp/w_10bin",
"width vs rank bins (10)");
gplotSimple1(nah_med, GPLOT_PNG, "/tmp/h_10bin",
"height vs rank bins (10)");
numaDestroy(&naw_med);
numaDestroy(&nah_med);
/* Make the rank bin arrays of median values, with 30 bins */
numaGetRankBinValues(naw, 30, NULL, &naw_med);
numaGetRankBinValues(nah, 30, NULL, &nah_med);
gplotSimple1(naw_med, GPLOT_PNG, "/tmp/w_30bin",
"width vs rank bins (30)");
gplotSimple1(nah_med, GPLOT_PNG, "/tmp/h_30bin",
"height vs rank bins (30)");
numaDestroy(&naw_med);
numaDestroy(&nah_med);
/* Give gnuplot time to write out the files */
#ifndef _WIN32
sleep(2);
#else
Sleep(2000);
#endif /* _WIN32 */
/* Save as golden files, or check against them */
regTestCheckFile(fp, argv, "/tmp/w_10bin.png", 0, &success);
regTestCheckFile(fp, argv, "/tmp/h_10bin.png", 1, &success);
regTestCheckFile(fp, argv, "/tmp/w_30bin.png", 2, &success);
regTestCheckFile(fp, argv, "/tmp/h_30bin.png", 3, &success);
/* Display results for debugging */
pixt = pixRead("/tmp/w_10bin.png");
pixDisplayWithTitle(pixt, 0, 0, NULL, display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/h_10bin.png");
pixDisplayWithTitle(pixt, 650, 0, NULL, display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/w_30bin.png");
pixDisplayWithTitle(pixt, 0, 550, NULL, display);
pixDestroy(&pixt);
pixt = pixRead("/tmp/h_30bin.png");
pixDisplayWithTitle(pixt, 650, 550, NULL, display);
pixDestroy(&pixt);
pixDestroy(&pixs);
numaDestroy(&naw);
numaDestroy(&nah);
regTestCleanup(argc, argv, fp, success, NULL);
return 0;
}
l_int32 main(int argc,
char **argv)
{
l_int32 i, w, h, n, val, ne, no, nbins, minw, maxw, minh, maxh;
l_int32 mine, mino, maxe, maxo;
l_int32 w_diff, h_diff, median_w_diff, median_h_diff;
l_int32 noutw, nouth;
l_float32 medwe, medhe, medwo, medho;
BOXA *boxa1, *boxa2, *boxae, *boxao;
NUMA *na1, *nawe, *nahe, *nawo, *naho;
NUMA *nadiffw, *nadiffh; /* diff from median w and h */
NUMA *naiw, *naih; /* indicator arrays for small outlier dimensions */
NUMA *narbwe, *narbhe, *narbwo, *narbho; /* rank-binned w and h */
PIX *pix1;
PIXA *pixa1;
L_REGPARAMS *rp;
if (regTestSetup(argc, argv, &rp))
return 1;
lept_mkdir("lept/boxa");
boxa1 = boxaRead("boxa4.ba");
/* Fill invalid boxes */
n = boxaGetCount(boxa1);
na1 = boxaFindInvalidBoxes(boxa1);
if (na1)
boxa2 = boxaFillSequence(boxa1, L_USE_SAME_PARITY_BOXES, 0);
else
boxa2 = boxaCopy(boxa1, L_CLONE);
boxaDestroy(&boxa1);
/* Get the widths and heights for even and odd parity */
boxaSplitEvenOdd(boxa2, 0, &boxae, &boxao);
boxaGetSizes(boxae, &nawe, &nahe);
boxaGetSizes(boxao, &nawo, &naho);
boxaDestroy(&boxa2);
/* Find the medians */
numaGetMedian(nawe, &medwe);
numaGetMedian(nahe, &medhe);
numaGetMedian(nawo, &medwo);
numaGetMedian(naho, &medho);
/* Find the median even/odd differences for width and height */
median_w_diff = L_ABS(medwe - medwo);
median_h_diff = L_ABS(medhe - medho);
regTestCompareValues(rp, 210, median_w_diff, 0.0); /* 0 */
regTestCompareValues(rp, 15, median_h_diff, 0.0); /* 1 */
if (rp->display) {
fprintf(stderr, "diff of e/o median widths = %d\n", median_w_diff);
fprintf(stderr, "diff of e/o median heights = %d\n", median_h_diff);
}
/* Find the differences of box width and height from the median */
nadiffw = numaMakeConstant(0, n);
nadiffh = numaMakeConstant(0, n);
ne = numaGetCount(nawe);
no = numaGetCount(nawo);
for (i = 0; i < ne; i++) {
numaGetIValue(nawe, i, &val);
numaSetValue(nadiffw, 2 * i, L_ABS(val - medwe));
numaGetIValue(nahe, i, &val);
numaSetValue(nadiffh, 2 * i, L_ABS(val - medhe));
}
for (i = 0; i < no; i++) {
numaGetIValue(nawo, i, &val);
numaSetValue(nadiffw, 2 * i + 1, L_ABS(val - medwo));
numaGetIValue(naho, i, &val);
numaSetValue(nadiffh, 2 * i + 1, L_ABS(val - medho));
}
/* Don't count invalid boxes; set the diffs to 0 for them */
if (na1) {
for (i = 0; i < n; i++) {
numaGetIValue(na1, i, &val);
if (val == 1) {
numaSetValue(nadiffw, i, 0);
numaSetValue(nadiffh, i, 0);
}
}
}
/* Make an indicator array for boxes that differ from the
* median by more than a threshold value for outliers */
naiw = numaMakeThresholdIndicator(nadiffw, 90, L_SELECT_IF_GT);
naih = numaMakeThresholdIndicator(nadiffh, 90, L_SELECT_IF_GT);
numaGetCountRelativeToZero(naiw, L_GREATER_THAN_ZERO, &noutw);
numaGetCountRelativeToZero(naih, L_GREATER_THAN_ZERO, &nouth);
regTestCompareValues(rp, 24, noutw, 0.0); /* 2 */
regTestCompareValues(rp, 0, nouth, 0.0); /* 3 */
if (rp->display)
fprintf(stderr, "num width outliers = %d, num height outliers = %d\n",
noutw, nouth);
numaDestroy(&nadiffw);
numaDestroy(&nadiffh);
numaDestroy(&naiw);
numaDestroy(&naih);
/* Find the rank bins for width and height */
nbins = L_MAX(5, ne / 50); // up to 50 pages/bin
numaGetRankBinValues(nawe, nbins, NULL, &narbwe);
numaGetRankBinValues(nawo, nbins, NULL, &narbwo);
numaGetRankBinValues(nahe, nbins, NULL, &narbhe);
numaGetRankBinValues(naho, nbins, NULL, &narbho);
numaDestroy(&nawe);
numaDestroy(&nawo);
numaDestroy(&nahe);
numaDestroy(&naho);
/* Find min and max binned widths and heights; get the max diffs */
numaGetIValue(narbwe, 0, &mine);
numaGetIValue(narbwe, nbins - 1, &maxe);
numaGetIValue(narbwo, 0, &mino);
numaGetIValue(narbwo, nbins - 1, &maxo);
minw = L_MIN(mine, mino);
maxw = L_MAX(maxe, maxo);
w_diff = maxw - minw;
numaGetIValue(narbhe, 0, &mine);
numaGetIValue(narbhe, nbins - 1, &maxe);
numaGetIValue(narbho, 0, &mino);
numaGetIValue(narbho, nbins - 1, &maxo);
minh = L_MIN(mine, mino);
maxh = L_MAX(maxe, maxo);
h_diff = maxh - minh;
numaDestroy(&narbwe);
numaDestroy(&narbhe);
numaDestroy(&narbwo);
numaDestroy(&narbho);
regTestCompareValues(rp, 409, w_diff, 0.0); /* 4 */
regTestCompareValues(rp, 49, h_diff, 0.0); /* 5 */
if (rp->display)
fprintf(stderr, "Binned rank results: w_diff = %d, h_diff = %d\n",
w_diff, h_diff);
/* Plot the results */
if (noutw > 0 || nouth > 0) {
pixa1 = pixaCreate(2);
boxaPlotSizes(boxae, "even", NULL, NULL, &pix1);
pixaAddPix(pixa1, pix1, L_INSERT);
boxaPlotSizes(boxao, "odd", NULL, NULL, &pix1);
pixaAddPix(pixa1, pix1, L_INSERT);
pix1 = pixaDisplayTiledInRows(pixa1, 32, 1500, 1.0, 0, 30, 2);
regTestWritePixAndCheck(rp, pix1, IFF_PNG); /* 6 */
pixDisplayWithTitle(pix1, 100, 100, NULL, rp->display);
pixDestroy(&pix1);
pixaDestroy(&pixa1);
}
boxaDestroy(&boxae);
boxaDestroy(&boxao);
return regTestCleanup(rp);
}
