New Algorithm Based On Modified Markov Random Field In SAR Images Segmentation

This thesis describes the modified MRF(Markov Random Field) theory which is used for SAR(Synthetic Aperture Radar) images segmentation. MRF model have lots of applications in SAR images segmentation, SAR images segmentation is an important stage in the postprocessing of SAR images, so these is some practical application for the research in SAR images recognition.The method involves the intensity information in feature field and the distances information in label field between pixels based on the traditional MRF potential function, it can utilize more special information of SAR images in the potential function model so that the segmentation result can be better. At the same time, potential function combines the penalty term on the edge strength, it can be localized well on edge. So a new algorithm for SAR images segmentation based on MRF fused edge information is given in thesis. The results show that the proposed algorithm performs well in miss-classification ratio, higher localization than ICM (iterated condition mode) and SA (simulative anneal) algorithm on the traditional MRF.The method utilize different orders of Gaussian-Hermite moments as SAR image features, It can get initial segmentation by adaptive threshold. A balance is then achieved between a region labeling component and a feature modeling component by introducing a energy function-based parameter to weight the two components in MRF model, the accurate result is achieved by MAP (maximum a poteriori). So a new fusion algorithm for unsupervised SAR images segmentation based on Markov Random Field with Gaussian-Hermite moments is also proposed in thesis. The experimental results show that it performs well in restraining the speckle noise, miss-classification ratio, and well performation in vision.