Selection of regularization parameter in GMM based image denoising method

摘要

Currently, the image denoising methods using Gaussian mixture model to learn image prior have received much attention. Among these methods, expected patch log likelihood based image denoising approach has been shown to be surprisingly competitive in image restoration. However, recent related works generally utilize global regularization parameter that influences the performance of denoising algorithm. In this paper, with the consideration that the Gaussian mixture model has the capability of clustering, we propose an adaptive estimation method of regularization parameter for expected patch log likelihood based image denoising. Our method jointly employs the Lagrange multiplier technique and entropy concept to select regularization parameter for each underlying cluster. Experimental results illustrate the relatively good performance of our image denoising method in terms of visual improvement and peak signal to noise ratio.