Bayesian segmentation supported by neighborhood configurations

From the statistical point of view, segmentation methods are dependent upon how the characteristics in image are formulated and where they are extracted from. In this paper, the joint conditional probability is exploited to characterize the statistical properties and is also localized to better capture the local properties of the neighborhood. Two different neighborhood configurations are defined and each of them incorporates with given prior information through Bayesian formula. It is considered as a criterion function in the proposed method. The proposed method segments images by maximizing the given criterion function. The results show the comparison of the results from four different methods depending on the combination of neighborhood configurations with prior information.