Two-step variance-adaptive image denoising

In this paper, we describe a two-step variance-adaptive method for image denoising based on a statistical model of the coefficients of balanced multiwavelet transform. The model is derived in a statistical framework from a recent successful scheme developed in the seemingly unrelated front of lossy image compression. Clusters of multiwavelet coefficients are modeled as zero-mean Gaussian random variables with high local correlation. In the adopted framework, we use marginal prior distribution on the variances of the multiwavelet coefficients. Then, estimates of the local variances are used to restore the noisy multiwavelet coefficients based on a minimum mean square error estimation (MMSE) procedure. Experimental results, using images contaminated with additive white Gaussian noise, show that the proposed method outperforms most of the denoising schemes reported in the literature. In this paper, the performance comparison is restricted to non-redundant multiresolution representations.