Sparsity based image inpainting using optimisation techniques

Digital image inpainting means a technique used for filling in the missing or corrupted area of an image using information from the known region such that the image looks natural and complete after the inpainting process. This paper proposes a patch-wise image inpainting algorithm using various sparse representation over different overcomplete dictionaries. Dictionary generated using patches from source region gives better results as compared to the DCT based dictionary. For sparse coding, various greedy algorithms and relaxation algorithms have been implemented and tested. Relaxation algorithms provide better results compared to greedy algorithms.