SAR Images Segmentation Based On Fuzzy HMMS In Contourlet Domain

The segmentation of SAR images is a crucial point in the automatic analysis and interpretation of data, and the research for the SAR image segmentation algorithm has been a hot spot. Since real SAR images are corrupted with speckle noise and have the features of high gray level, sufficient texture and faint border, the traditional segmentation methods and the segmentation algorithm based on classical MRF model that work successfully on natural images do not perform well on SAR images.Aimed at the high levels of speckle noises and faint borders of SAR images, in this paper, we combine the MRF with fuzzy concepts, use a threshold standard to identify the fuzzy and hard MRF, and then an improved fuzzy Markov random field (FMRF) model is established, we extend the new FMRF model to Contourlet domain, then a new algorithm based on the improved FMRF model in Contourlet domain is proposed. In this algorithm, FMRF is used to character the clustering of Contourlet coefficients, and through the analysis of the FMRF model, we deduce the membership formula of the pixel to a class, which is an efficient,unsupervised image segmentation algorithm. The new segmentation method is used to segment the texture and real SAR images, the results and analysis show that the proposed algorithm effectively suppress the affect of noise and provide a superior contour representation and an excellent visual performance.We also propose a new segmentation algorithm of SAR images based on HHMT-FMRF mixed model in Contourlet domain. This algorithm fuses the FMRF model and the half hidden Markov tree (HHMT) model at the scales of Contourlet transform, as a result, the clustering and persistence of Contourlet transform is fused into the segmentation of SAR images. The segmentation experiments of SAR images demonstrate that the proposed algorithm effectively suppresses the affect of noise, reduces the misclassification phenomena, and improves the efficiency of algorithm.