compute_haralick_features works incorrectly when `im_intensity` is a float with [0, 1] values

[idc9]

My understanding of compute_haralick_features is that you can provide im_intensity as a 2d array of float intensity values between 0 and 1. However, as written, the function will always cast im_intensity to an int in this line. This leads to incorrect behavior for [0,1] intensity valued images i.e. these will be binarized, which is not what we want to do. For example, most of the Haralick features end up being constant.

A simple fix would be to simply remove the .astype(np.uint8) i.e.

subImage = im_intensity[minr:maxr+1, minc:maxc+1]

I'm not sure what the purpose of casing subImage to np.uint8 -- is there a problem with removing this?

[cooperlab]

@idc9 this is addressed by #948 . We'll get this merged after the reviewer returns from travel.

[adrtsc]

Hi, just wanted to add here as the PR that was mentioned seems to still be open. I've been using the haralick features function and I ran into a related issue. My input arrays are uint16 array and the current behaviour causes overflow once the arrays are cast to uint8 without any previous rescaling of the intensities.

[Aleksandr-Enin]

I've also encountered this problem, and at the very least it should be added in the documentation that the image has to be scaled to uint8 for compute_haralick_features to work properly.