Sar Image Segmentation Algorithm Based On Spatial Information Clustering

SAR image segmentation is one of the key technique in automatic SAR imagery interpretation, whose results have a significant impact on SAR imagery interpretation. The unique coherent nature of the SAR imaging system provide a higher resolution image than optical image. Due to the heavy unavoidable speckle noise, image segmentation algorithm for traditional additive noise is no longer applicable to SAR image.Gamma mixture model has been widely used in SAR image segmentation algorithm according to its unique statistical distribution with advantage of simple form and easy to calculate. But this algorithm only use the gray-scale image information, and neglect the spatial information between pixels, resulted in the sensitive to noise and large volume of incorrect segmentation. This dissertation proposed a new SAR image segmentation algorithm based on spatial information clustering, including the two aspects as follows:1) One SAR image segmentation algorithm was analysized through the combination of bilateral filtering and Gamma mixture model in this research. During the solution of the parameters in the model, the Bayesian posterior probability of each pixel is treated as the pixel value of the image. Then, the Bayesian posterior probability value of the adjacent pixels in the image get smooth with a similar values through bilateral filtering. Each iteration solution of the parameters was affected by the pixel position through integration one filtering operation into each iteration of the parameters. In this way, this algorithm both can get an accurate SAR image segmentation, and can better reduce the influence of the speckle noise on segmentation results.2) SAR image segmentation using regional Gamma mixture model was proposed. Watershed segmentation was acted as the initial segmentation, and the over-segmentation regions was served as the Gamma mixture model clustering samples, which make the Gamma mixture model clustering upgrade from traditional pixel level to the regional level. Considering the connection between the regions, one neighborhood factors was introduced into the iterative algorithm to get a weighted neighborhood generation probability, improved the results of segmentation.Adequate experiments and analysis was done with synthesis SAR image and real SAR image respectively. Comparing to traditional Gamma mixture clustering、regional Gaussian mixture model and regional MRF algorithm, the proposed algorithm have a significant improvement with a stronger anti-noise capability. The feasibility and effectiveness of the proposed algorithm in SAR image segmentation was proved.