Fast and Robust Edge-Guided Exemplar-Based Image Inpainting

A fast and robust edge-guide exemplar-based method of image inpainting is proposed in this paper. Unlike traditional exemplar-based methods, we introduce an edge-reconstruction procedure before inpainting textures. The edge reconstruction procedure exploits different properties of edges, such as the curvature similarity, color similarity, and other estimate of how well two edges connect to each other. Guided by the reconstructed edge lines, the improved exemplar-based method is used to restore the textures and remaining structures. Moreover, we redesign the random search strategy to make it more suitable for our framework to solve the time-consuming problem caused by exhaustive search in most exemplar-based methods. After the match patch is chosen, color transfer is used to propagate the match patch information to further improve the visual quality and perceptual reasonability. Comprehensive experiments are performed to compare the proposed method with other well-known methods on synthetic images and natural images. The results show the proposed method can not only greatly reduce the computing time of exemplar-based methods, but also behave better on visual plausibility and continuity.