SAR Image Despeckling Based On Nonlocal Means And Lee Filter

Synthetic aperture radar (SAR) system has the ability of all-weather, all-time, high resolution and strong penetrating, therefore, the SAR image has been widely applied in national security and national economy. Due to the special SAR imaging system, the image is polluted by speckle noise. The speckle noise makes the signal-to-noise ratio decreased and the characteristics of target weakened, which cause difficulties in the research of target recognition and feature extraction. Consequently, SAR image despeckling is an essential process.In order to reduce the speckle and restore true radar reflection coefficient, this paper purposes to use the non-local mean algorithm framework and combines the advantages of the Lee filter to SAR images despeckling, the major works includes the following three aspects:(1) An algorithm based on fast non-local Lee filter for SAR image despeckling is presented. The algorithm designs five kinds of direction template for the different types of target in the SAR images, which improves the speed of block matching and makes the calculation of block similarity rapidly, then, combine the advantages of non-local mean with the Lee filter. Thus, a new kind of non-local Lee filter is designed to improve the performance of Lee filter.(2) A refined sigma filter for SAR image despeckling is presented, based on a novel reference image which is constructed according to the first filtering results and the method noise, and then uses the reference image to replace the original SAR image for the selection of homogeneous pixels. The reference image has the advantages of containing almost the same amount of details as the original SAR image but with less speckle noise, and thus enables a better sigma selection, so the performance of sigma filter is improved.(3) An algorithm based on hybrid similarity of the non-local Lee filter is proposed, which combines the Lee filter with the non-local mean frame, and integrates with the structure similarity and homogeneity similarity. In contrast with the great power of smoothing speckle based non-local mean filter, our algorithm preserves the details. The algorithm also uses a fast algorithm to greatly reduce the complexity.This work was supported by the Fund for Foreign Scholars in University Research and Teaching Programs (the111Project)(No. B07048), and the Program for Cheung Kong Scholars and Innovative Research Team in University(No. IRT1170).