PROJECT TITLE :

High-quality Image Restoration Using Low-Rank Patch Regularization and Global Structure Sparsity

ABSTRACT:

In recent years, picture restoration has improved significantly thanks to techniques based on nonlocal self-similarity and global structural regularisation. Repetitiveness of small picture patches can be used as a powerful prior in the reconstruction process through nonlocal self-similarity. A similar approach underpins global structure regularisation, in which pixels are used to indicate the structure of objects in the image. Keeping this structural information to a minimum reduces the likelihood of artefacts in the reconstruction process. Most image restoration methods have only evaluated one of these two options up to now, and not both at the same time. Nonlocal self-similarity and global structure sparsity are combined in a single efficient model in this article. A weighted nuclear norm-based adaptive regularisation technique is used to recreate a group of similar patches at the same time. A novel technique that decomposes the image into smooth and sparse residual components, the latter of which is regularised using the l1norm, preserves the image's global structure. For efficient picture recovery, an algorithm that uses the alternating direction approach of multipliers algorithm is used. Image completion and super-resolution tasks are used to test the suggested method's performance. Our solution outperforms current state-of-the-art approaches for these tasks, regardless of the extent of image corruption, according to experimental results.