Hi David, thanks for the response. That makes sense now. The key part I was missing was that the output is N dimensional with a unique bin for each possible combination of components. I was thinking it was simply one 1D histogram per component - e.g. bin count * 3 for RGB.