Student's-t Mixture Model Based Excepted Patch Log Likelihood Method for Image Denoising

摘要

Recently, patch priors based image denoising method has received much attention in recent years. Expected patch log likelihood (EPLL) is a popular method with the patch priors for image denoising, which achieves image noise removal using the Gaussian mixture priors learned by the Gaussian mixture model (GMM). In this paper, with observation that the student's-t distribution has a heavy tail and is robust to noise compared with the GMM, we attempt to learn image patch priors using the student's-t mixture model (SMM), which is an extension of the GMM. Experiment results demonstrate that our proposed method performs an improvement than the original EPLL.