Speckle Noise Reduction in SAR image based on K-SVD

This paper presents a method based on K-SVD (Singular Value Decomposition) to despeckle Synthetic Aperture Radar (SAR) image. First, we get sparse representation of a noisy image f over a fixed complete dictionary D by using Orthogonal Matching Pursuit (OMP). Then, train D on the noisy image to get a updated dictionary D ' with K-SVD algorithm. Finally, reconstruct image with the new dictionary D'. Compared with traditional methods, this method is better in despeckling effect and image fidelity