An image inpainting algorithm based on sparse modeling with double constraints

An exemplar-based inpainting algorithm via sparse representation is presented to focus on the problem of removing large objects from digital image. Exemplars that are similar with the target patch are searched in known regions. These exemplars are all neighbors of the target patch in a high-dimensional data space by regarding each of them as a high-dimensional vector. The unknown region is estimated through locally linear embedding method by supposing these neighbors in a same manifold, and then, sparse representation is applied to enforce compatibility and sharpness with the surrounding region. Experimental results and comparison with the traditional exemplar-based inpainting algorithm show that the proposed algorithm can effectively repair the texture and structure information in the damaged image.