SAR image despeckling based on improved Directionlet domain Gaussian Mixture Model

In this paper, a new SAR image despeckling method based on the improved Directionlet domain Gaussian Mixture Model (GMM) is proposed. Firstly, the cartoon texture model is used to decompose the SAR image to a cartoon part and a texture part. Secondly, the cartoon part is kept unchanged, the coefficients of the texture part in the improved Directionlet domain are modeled by the Gaussian Mixture Model. Thirdly, the Bayesian minimum mean square error estimation is used to evaluate each of coefficients. Finally, the two parts are added to obtain the despeckled image. Experimental results show that the proposed method outperforms the spatial filters and other methods based on wavelets, stationary wavelet and non-subsampled contourlets in terms of speckle reduction as well as detail and edge preservation.