Non-Local Self-Similarity Tensor Ring Completion: Tensor Ring Completion using non-local self-similarity for Image Recovery

Tenor Ring (TR) is a cutting-edge tensor decomposition algorithm, and the tensor completion model based on TR decomposition is widely used in image recovery, which is able to estimate the complete image with only a portion of observable pixel points and get good results. However, the existing completion models do not consider the non-local self-similarity of images, which makes it difficult to achieve better accuracy. To address this problem, this paper proposes a TR decomposition tensor completion model NLTRC combining the non-local self-similarity denoising module, which can not only exploit the image spectrum and spatial correlation, but also exploit the non-local self-similarity to approximate the original tensor at low rank, and then recover the image. Extensive experiments show that our algorithm achieves better results than several state-of-the-art baselines in color and multispectral images.