SAR Image Despeckling Based on the Nonlocally Centralized Sparse Representation Model

Abstract

In this paper, a novel SAR image despeckling method based on the nonlocal centralized sparse representation (NCSR) is presented. NCSR model is initially proposed for image restoration, exhibiting excellent denoising performance for nature images corrupted by additive Gaussian noise. However, SAR images, whose speckle noise is multiplicative and follows non-Gaussian distribution, are very different from the nature images. Considering the properties of speckle, SAR images are pre-processed, and then based on the generalized likelihood ratio (GLR) criteria, a new metric function is developed for acquiring a better estimates of the sparse coding coefficients of each corresponding noiseless SAR image patch. Experimental results show that the proposed method has a good capability of speckle smoothing in homogeneous region, as well as edge and texture preservation.