Image Denoising With Gaussian Mixture Model

Abstract

In this paper, we present a Gaussian mixture model (GMM) based method for image denoising. The method partitions an image into a set of overlapping patches, and assumes that the image patches are random variables described by a GMM. The distribution parameters of the noise free image patches are estimated from the noisy parameters which are calculated by expectation maximization (EM). Minimum mean square error (MMSE) estimation technique is used to estimate the clean image patches. The experimental results show that new method can effectively suppress additive noise and preserve details of image signal.