Multi-Scale Patch Prior Learning for Image Denoising Using Student's-t Mixture Model

Abstract

Patch prior based image regularization technique has drawn much attention recently. The Multi-Scale Expected Patch Log Likelihood (MSEPLL) algorithm as a popular method for learning multi-scale prior of image patches has shown competitive results. However, the current algorithm learns patch prior with the Gaussian Mixture Model that is sensitive to outliers commonly. In this paper, we extend the MSEPLL method and attempt to employ the student's-t mixture model (SMM) to learn multi-scale image patch prior in a more robust way. Experiment results demonstrate that our proposed method performs well both in visual effect and quantitative evaluation.