Bayesian image segmentation based on an inhomogenous hidden Markov random field

This paper introduces a Bayesian image segmentation algorithm with the consideration of label scale variability in many images. An inhomogeneous hidden Markov random field is adopted in this algorithm to model the label scale variability as a prior probability. An EM algorithm is developed to estimate parameters for both the prior probability and likelihood probability. The image segmentation is established by a MAP estimator. Different images are tested to verify our algorithm. Comparisons with other segmentation algorithms are made. The segmentation results show that our algorithm has better performance than others.