Abstract:Based on the assumption that wavelet coefficients obey Generalized Gaussian Distribution (GGD), this paper adopts Maximum Likelihood (ML) principle to estimate wavelet coefficients variance of common images in sub-bands. The proposed estimator is product of a sub-band adjustable factor and a power mean factor. Compared to the recently proposed SI-AdaptShr, LAWMAP and other wavelet-based methods, better de-noising results may be obtained for the proposed method. Furthermore, a simplified algorithm is also formed to de-speckle SAR images. It is shown that the new method may remarkably reduce the calculation amount and helpful for the post-processing of large scale SAR images.