Image Denoising via Multiple Images Nonlocally Means and Residual Tensor Decomposition

The nonlocal means and iterate filtering techniques have attracted much research effort due to their superior performances. In this paper, the two approaches are combined into a framework to perform denoising based on residual tensor decomposition in loop of iteration. The search of accurate similar patches is essential for image denoising, to reconstruct the damaged part, we utilize multiple images nonlocal means method to exploit the image nonlocal self-similarity and obtain accurate weight, thus eliminate the interference of unsimilar patches. Although the degraded or lost slight structure of the image due to imperfect denoising methods, we propose the use of iterating residual image to compensate the sharpness of image texture via patch-based tensor decomposition, which can describe the intrinsic geometrical structure of there similar data. We use standard test images and multi-frame video test sequences to illustrate that the proposed denoising algorithm outperforms the leading algorithms such as weighted nuclear norm minimization (WNNM), BM3D and NCSR in terms of the quantitative and perceptual evaluation.