confidence should be defined by bimodality not range #21
Description
the real measure of "confidence" in a pixel isn't how big the difference between bright and dark is.
but when you look at all the values for a single pixel across all photos, it's how bimodal the distribution is. if it's very bimodal, that's good. the degree to which there are outliers indicates overlapping reflections, and if it is well approximated by a single gaussian then it's just background noise.