Information fusion in a Markov random field-based image segmentation approach using adaptive neighbourhoods

In this paper an image segmentation method is proposed that is a modification to the Markov random field (MRF) region label process used by Rignot and Chellappa (1992). Using Bayesian inference, the optimal shape of the neighbourhood system is determined on the basis of the Markovian property. This MRF segmentation approach with adaptive neighbourhood systems (MRF-AN) makes it possible to better preserve small features by the combination of evidence from different knowledge sources. The purpose of the article is to show the validity of the concept of MRF-AN for image segmentation. Results are shown using synthetic aperture radar data