Image Restoration Based on Improved Patch Clustering in Gaussian Mixture Models

Abstract

In this paper, the restoration of degraded image is studied. Firstly, the unreliable pixel estimation is used to preprocess the noise image to eliminate the unreliable abnormal pixels in the noise image. Secondly, the prior model is used to model the multi-variable Gauss mixture model, and the GMM image restoration method is improved in the clustering step. The distance similarity measurement between patches and the same patch pixel is considered in the clustering. Using kernelized L2-norm not only considers the difference of pixels between image blocks, but also the difference between central pixels and other pixels in the same image block, which can effectively protect image details. Based on the above improvements, a restoration method based on the similarity between each patch and the estimated Gaussian cluster is proposed, which produces a better restoration effect. The experimental results verify the effectiveness of the model proposed in this paper.