Hi,
I just observed a strange behaviour of cv2.calcHist. When calculating the sum of all bin-values it sometimes does not add up to the number of pixels of an image.
img = cv2.imread("Lenna.png")
print img.shape
# => (512L, 512L, 3L)
for i in range(3):
hist = cv2.calcHist([img], [i], None, [256], [0,255])
print np.sum(hist)
# => 262144.0
# => 262144.0
# => 262032.0
The number of pixels in each channel should be 262144, which is correct for the blue and green channel. The value of the red channel differes by 112. This seems to be a serious bug at least in the Python wrapper. I currently have no time to reproduce this in C++.
I am using OpenCV-Python 2.4.8 (64Bit) compiled for Windows by Christoph Gohlke: http://www.lfd.uci.edu/~gohlke/pythonlibs/#opencv and Numpy 1.7.1
The error remains also after updating to Opencv 2.4.8.1 and Numpy 1.8.1
The image used was: http://upload.wikimedia.org/wikipedia/en/2/24/Lenna.png
edit 2014-03-29:
Some more debug data. The following two histograms have been created for the red channel. The first one is the result of the OpenCV function, the second was calculated using Numpy.
OpenCV:
print cv2.calcHist([img], [2], None, [256], [0,255]).astype(np.int).transpose()
[ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 1 1 1 8
6 10 9 20 37 32 41 53 74 103 132 179 193 242 238
336 386 404 483 607 628 771 785 1001 1016 1177 1269 1332 1446 1487
1457 1574 1585 1591 1557 1569 1660 1648 1420 1559 1418 1384 1319 1342 1156
1120 955 969 828 782 752 737 719 700 628 673 587 617 610 592
557 593 552 566 582 559 571 520 664 650 618 730 594 667 675
685 771 715 667 740 744 766 765 772 817 817 744 806 760 777
812 797 799 861 814 910 907 918 888 1011 879 996 912 952 884
1074 977 1073 1040 1216 1250 1403 1534 1639 1682 1776 1874 1769 1582 1743
1441 1477 1483 1409 1437 1449 1389 1479 1592 1655 1657 1666 1857 1896 1813
1979 1814 1956 1928 2055 2012 2303 2333 2670 2787 3232 3154 3476 3424 3516
3102 3176 2787 2885 2630 2731 2664 2955 2955 3360 3554 4138 3987 4057 4327
3713 3185 2929 2551 2432 2195 2256 1960 2126 2186 2265 2417 2445 2282 1826
1972 1456 1137 986 748 749 667 582 428 357 313 302 242 178 67
0]
Numpy:
hist, _ = np.histogram(img[:,:,i],np.arange(0,257))
print hist
[ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1 0 1 1 1 8
6 10 9 20 37 32 41 53 74 103 132 179 193 242 238
336 386 404 483 607 628 771 785 1001 1016 1177 1269 1332 1446 1487
1457 1574 1585 1591 1557 1569 1660 1648 1420 1559 1418 1384 1319 1342 1156
1120 955 969 828 782 752 737 719 700 628 673 587 617 610 592
557 593 552 566 582 559 571 520 664 650 618 730 594 667 675
685 771 715 667 740 744 766 765 772 817 817 744 806 760 777
812 797 799 861 814 910 907 918 888 1011 879 996 912 952 884
1074 977 1073 1040 1216 1250 1403 1534 1639 1682 1776 1874 1769 1582 1743
1441 1477 1483 1409 1437 1449 1389 1479 1592 1655 1657 1666 1857 1896 1813
1979 1814 1956 1928 2055 2012 2303 2333 2670 2787 3232 3154 3476 3424 3516
3102 3176 2787 2885 2630 2731 2664 2955 2955 3360 3554 4138 3987 4057 4327
3713 3185 2929 2551 2432 2195 2256 1960 2126 2186 2265 2417 2445 2282 1826
1972 1456 1137 986 748 749 667 582 428 357 313 302 242 178 67
112]
So there are the missing 112 pixels. The last bin of the OpenCV histogram has a zero value. Explaination still missing.
