An Analysis of Histogram-Based Thresholding Algorithms

Abstract

Eleven histogram-based global thresholding algorithms are presented in a common notational framework. Relationships among them are identified from 654 mixtures of two Gaussian distributions, plus effects of mixed pixels. The iterated version of Kittler and Illingworth′s minimum error algorithm (Pattern Recognition, 19, 1986, 41-47) is found to be best.