On Fri, Apr 20, 2012 at 12:41 PM, David Stolarsky firstname.lastname@example.org wrote:
It looks to me like the HoG computation ignores half of the possible gradient values (see attached image). There are 15 lines visible in the attached HoG visualization, likewise there are 15 non-zero values in the feature vector returned by feature.hog(). Is this a bug? Am I doing something wrong? I'm using the hog example script from http://scikits-image.org/docs/0.5/auto_examples/plot_hog.html%C2%A0, modified to take a file from the command line.
I don't have access to your input image, but I suspect what's happening is that when the image is divided up into 8x8 cells, the top edges fall on the boundaries of cells, while the bottom ones fall inside. If you send me the input image, I can verify, or you can modify the cell shape to see what happens.