Despeckling of SAR images using discrete shearlet transform

In this paper, a new SAR image despeckling method is presented. Discrete shearlet transform based on multiresolution analysis is used to obtain the transformed coefficients. Shearlet transform is good in processing multidimensional data, and have the property of high directional sensitivity and optimal sparse representation. Shearlet coefficients of noise free image are modeled using two sided generalized Gamma distribution (GrD) because of heavy tailed distribution of clean image transformed coefficients and the speckle noise is modeled using zero mean Gaussian distribution. The distribution parameters are estimated using method of log cumulants (MoLC). A Bayesian MAP estimator is used to estimate the noise free shearlet coefficients. The performance of the proposed method and some other existing methods are evaluated and compared using parameters peak signal to noise ratio (PSNR) and equivalent number of looks (ENL). Results are taken on synthetic image as well as real SAR image. The obtained results show the dominance of proposed technique over other existing techniques.