Image Denoising Using Gaussian Scale Mixture Model in the Nonsubsampled Contourlet Domain

Abstract

A new method which using Gaussian scale mixtures model in the nonsubsampled Contourlet domain is proposed for image denoising. First, a Gaussian scale mixture model is introduced in order to capturing the correlation of nonsubsampled Contourlet locally coefficients. Then the coefficients are estimated by Bayes least squares estimator based on the model. Finally, the inverse nonsubsampled Contourlet transform is applied to the modified coefficients. This arithmetic combines the character of nonsubsampled Contourlet for image edge representation, shift-invariance and the effective of Gaussian scale mixture model for capturing correlation of locally coefficients. The numerical experimental results show the validity of the proposed method.