Plug-and-Play Prior Based on Gaussian Mixture Model Learning for Image Restoration in Sensor Network

In this paper, we propose a method that use the Gaussian mixture model (GMM) as a plug-and-play prior for image restoration in sensor network. The “plug-and-play" concept as an image prior is extended to image restoration and just mentioned in recent years. Particularly, the plug-and-play alternating direction method of the multiplier is very fit for the GMM framework to regularize the sub-problems. In order to avoid error results caused by general minimum of mean square error criterion, we propose two spatial constraints: one applies the K-nearest-neighbor method based on an Euclidean distance to measure the similarity of image patches in clustering step; the other adopts the Gaussian weight based on the Mahalanobis distance to update the mean vector and covariance matrix. Finally, we compare our method with several recent state-of-the-art methods, and the results show that our proposed method has good performance in preserving details and eliminating the staircase effect.