Clutter patch identification based on Markov random field models

This paper addresses the problem of clutter patch identification based on Markov random field (MRF) models. MRF has long been recognized by the image processing community to be an accurate model to describe a variety of image characteristics such as texture. Here, we use the MRF to model clutter patch characteristics, captured by a radar receiver or radar imagery equipment, due to the fact that clutter patches usually occur in connected regions. Furthermore, we assume that observations inside each clutter patch are homogenous, i.e., observations follow a single probability distribution. We use the Metropolis-Hasting algorithm and the reversible jump Markov chain algorithm to search for solutions based on the Maximum a Posteriori (MAP) criterion. Several examples are provided to illustrate the performance of our algorithm.