SAR image interpretation based on Markov mesh random fields models

An unsupervised segmentation approach is presented for the analysis of intensity synthetic aperture radar (SAR) images. It utilizes hidden Markov mesh random fields (HMMRF), assuming that the image elements have a causal relationship: a pixel´s past is considered more important than its future, a hypothesis that is of particular interest to SAR images, given the way these images are generated. The approach is build upon the HMMRF theory of Devijver (1993), substituting his Gaussian distribution with the Gamma distribution, using an new initial image segmentation algorithm. The method is suited for real time image analysis, making the method powerful in, for instance, SAR image interpretation and data base browsing