Research On Speckle Reduction And Segmentation Algorithms For SAR Images

Synthetic Aperture Radar (SAR) is widely used in civilian and military fields. Speckle noise is the main obstacle of SAR images due to the scattered signals. Therefore, restraining speckle noise is necessary for further application. Because of the rich texture information and the rough edges, traditional algorithm cannot achieve satisfactory segmentation quality. Therefore, speckle reduction and segmentation become one of the hottest research objects in the field of SAR image processing. This thesis takes SAR image as research object and studies speckle reduction and segmentation algorithms.Firstly, a speckle reduction algorithm for SAR images based on nonsubsampled contourlet transform and correlation of subband coefficient is realized. The algorithm makes distinctions between texture and noise with NSCT. While remaining the coefficients of the low frequency image information, the coefficients of the high frequency image information are filtered by correlation criterion. Then, a new algorithm of speckle reduction of SAR local image is implemented which is due to Krawchouk moment. We select a sample set first, which can describe the local characteristics of the SAR image. The characteristic parameter vector of each sample is extracted by Krawtchouk moment. The source SAR image can be segmented into groups according to the local features. Speckle noise of each kind can be filtered well. The results show that both two algorithms have achieved good effects. Compared with the former algorithm, the latter has strong pertinence and good performance.Next, a rapid multi-threshold segmentation algorithm based on particle swarm optimization is introduced in spatial domain. The algorithm is making use of SAR image gray information and gradient information. In view of the rough edge and low computation speed of SAR image segmentation algorithm based on wavelet analysis in transformed domain, a fast SAR image segmentation algorithm based on chaotic particle swarm algorithm and maximum Tsallis entropy in NSCT domain is proposed. The Tent map chaotic particle swarm algorithm is introduced to search for the optimal threshold. The repetitive computations of fitness function in iteration are reduced significantly using recursive mode. Experimental results show that the second algorithm can achieve better segmentation effect.Finally, Texture segmentation based on Krawtchouk moment and SVM is researched. Texture features are extracted from SAR images by using Krawtchouk moment. The results show that, compared with the segmentation results based on Zernike moment, this algorithm can achieve a higher accuracy rate of texture segmentation.