Segmentation of SAR imagery using the Gaussian Markov random field model

Segmentation of SAR imagery by the use of Gaussian Markov random field (GMRF) model is studied. Following an initial segmentation obtained by watershed process, region merging based on the GMRF model is exploited. Three different merging criterions (merging of the most similar neighboring regions, merging of the most similar random regions and merging based on simulated annealing) are investigated. The results of segmentation of SAR images show that the first merging criterion is suitable for the images with continuous water areas, the second merging criterion is adapted to be implemented on the images with discontinuous water areas; the third one gives a good result with a huge burden of calculation.