A Sparse Representation-Based Label Pruning for Image Inpainting Using Global Optimization

This paper presents a new label pruning based on sparse representation in image inpainting. In this literature, the label indicates a small rectangular patch to fill the missing regions. Global optimization-based image inpainting requires heavy computational cost due to a large number of labels. Therefore, it is necessary to effectively prune redundant labels. Also, inappropriate label pruning could degrade the inpainting quality. In this paper, we adopt the sparse representation of label to obtain a few reliable labels. The sparse representation of label is used to prune the redundant labels. Sparsely represented labels as well as non-zero sparse labels with high similarity to the target region are used as reliable labels in global optimization based image inpainting. Experimental results show that the proposed method can achieve the computational efficiency and structurally consistency.