How many bins are applied to an image of different types in skimage.measure.shannon_entropy()

How does the skimage.measure.shannon_entropy() compute the histogram before returning the entropy result? I pass images in the format of 1,2 and 4 byte signed and unsigned grayscale images and also 32bit floating point grayscale images.


  • The source code is here:

    As you can see it is quite short, two lines of code. The histogram does not do binning at all, it uses the count output of numpy.unique. This means that the histogram will have as many elements as unique values in the image. For an 8-bit image, you’ll have up to 256 bins, but possibly less. For a 16-bit image you have up to 65536 bins, etc.

    For images with higher bit depth, and especially for floating-point images, you should quantize your image to get meaningful results. For a floating-point image you can assume each pixel has a unique value, which makes the entropy computation not so meaningful. Because Shannon entropy does not take the relationship between neighboring pixels into account, so a floating-point image of a smooth ramp has the same Shannon entropy as a floating-point image with random values.