Thresholding of biological images by using evolutionary algorithms

Abstract

This paper addresses the thresholding of biological images through multiobjective optimization techniques. Three objective functions are used during the optimization, which are combined at pairs: Shannon entropy versus Otsu's inter-class and Shannon entropy versus Otsu's intra-class. We show that although both combinations are obtaining the same vector of thresholds, the first objective function pair presents less computational effort to compute the Pareto front. Furthermore, we have also show that the size of the initial population of the evolutionary algorithm can be selected as 1/10 of the full space. As a consequence, Pareto fronts can quickly be computed and without affecting its performance and diversity.