On context-based Bayesian image segmentation: joint multi-context and multiscale approach and wavelet-domain hidden Markov models

Abstract

We show that context-based Bayesian image segmentation can be improved by strengthening both contextual modeling and statistical texture characterization. Firstly, we develop a joint multi-context and multiscale segmentation algorithm to achieve more robust contextual modeling by using multiple context models. Secondly, we study statistical texture characterization using wavelet-domain hidden Markov models (HMMs), and in particular, we use an improved HMM, HMT-3S to obtain more accurate multiscale texture characterization. Experimental results on two synthetic mosaic show that both contextual modeling and texture characterization play important roles in context-based Bayesian image segmentation.