Non-local mean despeckling algorithm based on edge strength map and region division for SAR image

Abstract

In order to reduce the multiplicative speckle noise of synthetic aperture radar(SAR) image, an improved non-local mean (NLM) algorithm is proposed in this paper. Firstly, the coefficient of variation (CV) was used to estimate the SAR image region division threshold, and estimate area division factor based on threshold. Thus, the SAR image wasdivided into texture area and homogeneous area by area division factor. Then, similarity windows of different scales were used to estimate the ESM corresponding to pixels in different regions, and the ESM matrix of each pixel in the search window under the similarity window and the ESM matrix of the center pixel were selected further. Moreover, the two-norm of the ratio of the two corresponding elements were used to construct the similarity measurement parameter that adapts to the multiplicative noise, filter value was obtained by combining the similarity parameter with the adaptive decay factor dominated by CV. Finally, the non-local average weighted filtering of the SAR image was realized by different regions with different filtering values. Compared with NLM filtering algorithm based on the mean ratio and the coefficient of variation, the proposed algorithm performs better in that the equivalent numbers of looks increased by more than 0.3% and edge preservation index increased by more than 0.1%.

Keywords

• synthetic aperture radar image
• speckle noise
• non-local mean filtering
• edge strength map
• region division