Exemplar-based Image Inpainting by Structured Sparse Representation

Image inpainting is a technique that aims to recovering missing or damaged regions of an image without any manual intervene. Conventional exemplar-based approaches suffer two major defects. One defect is that the inpainting quality heavily depends on the extent of selfsimilarity. If no similar samples exist in the already known region of the image, the missing region may be filled by irrelevant ones, which lead to an unexpected result. The other is that the process of finding the best-match sample is computationally intensive. In this paper, we introduce an exemplar-based image inpainting algorithm by exploiting structured sparse representation techniques to overcome the limitations mentioned above. The filling-in patch is estimated based on a composite dictionary instead of an exhausted search. The experimental results show that the proposed method achieves a good visual quality.