Denoising using Laplacian mixture model with local parameters in shearlet domain

Abstract

An adaptive Bayesian estimator for image denoising in shearlet domain is presented, where a mixture of Laplace distributions are used as the prior model of shearlet coefficients of images. The mixture of Laplacian probability density function has a large peak at zero and its tails fall significantly slowly than a single Laplacian pdf and the Laplacian mixture model can model shearlet coefficients distribution better. Under this prior, a Bayesian shearlet estimator is derived by using the maximum a posterior (MAP) rule. Simulations with images contaminated by additive white Gaussian noise are carried out to show that the performance in shearlet domain substantially surpasses that in wavelet domain, both visual effect and peak signal-to-noise ratio (PSNR). © 2012 Springer Science+Business Media B.V.