In this article I shown a C++ implementation of the “max entropy” threshold algorithm using the CImg library. This implementation also performs for 16 and 32 bit / float images.

First, a little bit of context: Some time ago I implemented the OTSU threshold algorithm as a pre-processing step for image binarization. I used that threshold algorithm to distinguish the noise pixels from the potential “star” pixels. This worked quite well for the high contrast input images at that time. However, for weak stars it unfortunately failed badly.