An adaptive SAR image speckle reduction algorithm based on undecimated wavelet transform and non-local means

Speckle noise is an inherent problem in synthetic aperture radar (SAR) images. Recently, non-local (NL) means performs well in speckle reduction. However, there exist two issues in traditional NL means. Speckle noise and strong targets would affect the correctness, thus resulting in over smoothing and speckles unchanged. In this paper, an improved algorithm combined with wavelet transform and adaptive NL means is proposed. First, linear minimum mean-square error (LMMSE) filtering in undecimated wavelet domain is operated and we get a “clean” image. Second, we eliminate the influence of strong targets adaptively when similarity is derived. At last, under the help of similarity, original image rather than the “clean” image is filtered. Proposed method performs better in both speckle reduction and strong targets preserving compared with the traditional method.