An irregular MRF region label model for multi-channel image segmentation

Abstract

A multi-channel image segmentation method is discussed that utilizes a Markov random field (MRF) region label model with adaptive neighbourhoods. Bayesian inference is applied to realize the combination of evidence from different knowledge sources. In such a way, optimization of the shape of a neighbourhood set is achieved by following a criterion that makes use of hypothesis on the Markovian property by exploiting the local image content. The purpose of the article is to show the theoretical validity of the approach by elucidating correspondences and differences with a similar concept. Results are shown using optical remote sensing data.