A unified iterative denoising algorithm based on natural image statistical models: derivation and examples
Shan Tan, Licheng Jiao · Optics Express · 2008
Lots of prior models for natural image wavelet coefficients have been proposed in the last two decades. Although most of them belong to the Scale Mixture of Gaussian (GSM) models, they are of obviously different analytical forms. As a result, Bayesian image denoising algorithms based on these prior models are also very different from each other. In this paper, we develop a novel image denoising algorithm by combining the Expectation Maximization (EM) scheme and the properties of the GSM models. The developed algorithm is of a simple iterative form and can converge quickly. It only uses the derivative information of a probability density function and is suitable for all GSM-type prior models that have an analytical probability density function. The developed algorithm can be viewed as a unified Bayesian image denoising framework. As examples, several classical and recently-proposed prior models for natural image wavelet coefficients are tested and some new results are obtained.