Wavelet denoising with generalized bivariate prior model

Abstract

This paper presents a novel wavelet method based on a prior model of wavelet coefficient with a generalized bivariate distribution function. The generalized bivariate prior model extends the isotropic bivariate prior model and anisotropic prior model, and can well characterize the joint parent-child statistical properties of many categories of images. Owing to the fact that the variance of wavelet coefficients are quite different for different scales and positions, we propose a local adaptive marginal variance estimation method based on this newly generalized prior model to improve the accuracy of marginal variance by considering the distribution of wavelet coefficients as the combination of Gaussian distribution and Laplacian distribution. By simulation experiments from three aspects, the results of the proposed method outperform the mainstream isotropic wavelet methods and anisotropic wavelet methods, and the practicality of the proposed method is verified by real noise image experiments.