Image inpainting based on sparse representation with dictionary pre-clustering

Abstract

This paper proposed a new image inpainting algorithm based on sparse representation. In traditional exemplar-based methods, the image patch is inpainted by the best matched patch from the source region. This greedy search will introduce unwanted objects and has huge time consuming. The proposed algorithm directly employs all the known image patches to form an over-complete dictionary. And then, the overcomplete dictionary is clustered into several sub-dictionaries. Finally, the unrepaired image patches are repaired over their corresponding closest sub-dictionaries through non-negative orthogonal matching pursuit algorithm. Experimental results show that the proposed method achieves superior performance than state-of-the-art methods. In addition, the time complexity is greatly reduced in comparison with the traditional exemplar-based inpainting algorithm.