Hog stride parameter

scikit-image

#1

Hello,

I need to extract Histogram of Oriented Gradients (HOG) for a given image and I need to set a specific striding step. Is there an easy way to implement it by using the HOG scikit-image function?

Thank you for your answer.


#2

@araihC I believe the argument pixels_per_cell corresponds to what you might think of as “stride” here? Specifically, if you use the astronaut image with pixels_per_cell=(8, 8) and cells_per_block=(1, 1), you get an image of shape (64, 64, 1, 1, 9), ie, each cell is being computed independently (64 * 8 = 512) and the results concatenated. When you use the default cells_per_block=(3, 3), you get an output of shape (62, 62, ...) because of boundary effects.


#3

Note: use feature_vector=False to get a non-raveled output so you can play around with the shapes and get an understanding of how hog shape and stride parameters work.


#4

Thank you very much for your reply. As it might be obvious I am quite new to this kind of processing!

Thank you!