Total variation regularized low-rank tensor decomposition with nonlocal for single image denoising

Various noises in the image will reduce the quality of the image and seriously affect the processing of subsequent computer tasks. The recovery of single images is a more challenging problem than recovery of spectral images due to the lack of spectral information. In order to solve this problem, in this paper, we propose a method combining non-local self-similar priors and tensor decomposition to fully explore the inherent low-rank structure of a single image. Specifically, we use tucker decomposition to characterize the global self-similar patch of a single image. At the same time, we introduce anisotropic spatial-spectral total variation regularization to describe the segmented smooth structure in the image. In order to deal with the complex noise situation in the real scene. We model the noise in two parts, one part is sparse spot noise, and the other part is ubiquitous noise. Then we use the augmented Lagrange multiplier method to solve it. Experiments have proved that the introduction of non-local self-similar priors is crucial to the denoising problem of a single image. The proposed method is superior to all comparison methods