SAR Image Processing Based On Markov Random Fields

SAR image processing has been widely applied in different areas, and is a primary approach for people to gain the information. As the SAR imaging technology has been developed rapidly, the speed and quality of SAR data acquisition has been increased as well. Both of these will increase the demand for SAR image interpretation. With a lot of high resolution SAR measured data, how to process SAR image rapidly and automaticly is becoming an important research direction. From the perspective of probability, based on markov random field model and the statistical characteristic of SAR image, this thesis study target detection and segmentation in SAR image. The main work of this thsis is summarized as follows:1.The research background, the state of art and the significance of road detection and the segmentation of the land and sea are introduced in SAR image processing. The first part also summarizes the main work of the thesis.2.The markov random field model and its basic theory are introduced. Besides the concept of field, the definition of neighborhood and clique have been introduced in detail. The vital important theoretical support in the application of MRF is the equivalence of MRF and gibbs random field. So in this chapter, we describe the GRF and show the equivalence of MRF and GRF. Finally a commonly used energy minimization algorithm is presented.3.The implementation of segmentation algorithm of land and sea in SAR image is studied. In this part, an appropriate probability model is applied to describe the statistical features of sea and land samples in SAR data. After that, the likelihood information of two class of samples are obtained. Then considering the priori information of samples and combining the neighborhood and clique definition, segmenation of land and sea is achieved by energy minimization algorithm.4.The road detection algorithm in SAR image is studied. To realize the change from points to lines, point detection is fulfilled by edge detection in first step. The result is converted to line through Hough transformation then. Because the roads in SAR imagehave some specific attributes, using these as prior information, the road detection is accomplished with the help of MRF. The experiment results show that the detection performance depends on complexity of the scene, and reasonable and effective preprocessing is very important.