Gaussian Mixture Model Based Image Denoising with Adaptive Regularization Parameters

Abstract

Recently, Gaussian mixture model have been studied extensively in image denoising, for the reason that it can better represent image prior. However, the current Gaussian mixture model based image denoising approach commonly employs global regularization parameter, therefore leading to limited denoising performance. To further enhance the performance this method, we exploit a new scheme for spatially adaptive regularization parameter selection, which utilizes scale space technique and residual image statistics to set regularization parameter value according to image details. The experiment results show that our proposed image denoising method can obtain relatively well results both in vision and the value of peak signal to noise ratio.