A new algorithm for unsupervised image segmentation based on D-MRF model and ANOVA

Abstract

A new algorithm for unsupervised image segmentation is proposed in this paper, which is based on the D-MRF model and ANOVA. Firstly, ANOVA is incorporated to determine the number of clusters combining with several statistics. Compared with models based on information criteria, ANOVA avoids the parameter estimation error, which reduces time consumption. Secondly, histogram is adopted to verify the validity of the new algorithm. Secondly, D-MRF is adopted to setup modeling. Thirdly, based on MRF-MAP, image segmentation is realized through using ICM. In model fitting, DAEM is used to estimate parameters in image field; on the other hand, local entropy is simulated as parameters in label field. Finally, the validity and practicability of the new algorithm are verified by two experiments.