A Robust and Adaptive Image Inpainting Algorithm Based on a Novel Structure Sparsity

The existing patch sparsity based image inpainting algorithms have some problems in maintaining structure coherence and neighborhood consistence. To address the above problems, a robust and adaptive image inpainting algorithm based on a novel structure sparsity is proposed. The main improvement includes the following three aspects. Firstly, a novel structure sparsity function is defined according to the sparseness of the patch’s nonzero similarities to its neighboring patches to encourage structure propagation preferentially. Secondly, the neighborhood consistence constraint factor is adaptively determined according to the target patch’s structure sparsity value, which aims to reduce block effect and seam effect. Thirdly, to improve computational efficiency, the size of local search region is dynamically determined in accordance with the target patch’s structure sparsity value. Experimental results demonstrate that the proposed algorithm can obtain more pleasurable vision results than that by other similar methods.