A wavelet based image denoising using statistical sampler for Bayesian estimator

This paper presents a new wavelet based image denoising method, which includes a Bayesian framework and classical thresholding methods. The main goal here is computing for each wavelet coefficient the probability of being sufficiently clean. The three main novelties of our approach are: (1) estimating local regularity of an image and distinguishing between useful edges and noise; (2) initializing the mask by thresholding the average cone ratio (ACR); and (3) probabilistic shrinkage of wavelet coefficients, using a statistical sampler. The main advantage of this algorithm is improved denoising performance over earlier techniques, which is demonstrated in the results.