Fast Algorithm Based On TMF For Unsupervised Multi-class Segmentation Of SAR Images

Synthetic Aperture Radar (SAR) is of significant sense in national defense and environment. So as a hot spot in radar signal processing, the interpretation of SAR image is a very important task, in which SAR images segmentation is an important stage. Since the SAR image contains a large amount of speckle noise, the classical segmentation techniques that work successfully on natural images do not perform well on SAR images.In this dissertation, based on hidden Markov field and pairwise Markov field, we do research on triple Markov random field. Triplet Markov random fields (TMF) model is suitable for dealing with multi-class segmentation of non-stationary, non-Gaussian SAR images. In order to reduce the complexity of the model and algorithm to satisfy the requirement of real-time, robust and efficient processing of SAR images, a fast algorithm based on TMF for unsupervised multi-class segmentation of SAR images is proposed in this paper. For the speckle in SAR images, numerical characteristic, threshold selection and the rule of quadtree decomposition are researched firstly. With the new method, a SAR image can quickly be mapped into an edge-based pixon-reprenation, which results in a coarse decomposition in smooth regions ,and a fine decomposition in edges. Then combining TMF model with the pixon-reprenation, a new potential energy function of TMF based on pixon-reprenation is deduced. Finally, the segmentation is finished by Bayesian maximum posteriori mode(MPM). In this paper, the algorithm is used to segment simulated data and real SAR images. Experimental results and analysis demonstrate that compared with the TMF segmentation algorithm, the proposed algorithm can maintain the performance of the TMF segmentation algorithm, even improve. But the computational efficiency is greatly improved in our algorithm. So our algorithm is an effective fast algorithm.