Gradient-Based Search Space Reduction for Fast Exemplar-Based Image Inpainting

In this paper a fast exemplar-based image inpainting method is presented by a gradient-based search space reduction. Our method works based on the priority function defined by Criminisi et al. [5]. The priority function encourages pixels with high gradient magnitudes to be selected on image completion path. The behavior of patches which have high gradients at their center are heavily impressed by the gradient vector. In this paper we have employed the gradient vector to reduce the search space and speed up the image inpainting algorithm. By forming lists of pixels with similar gradient vectors at preprocessing stage and selecting the most relevant list for each patch, the search space is reduced in the order of 100 times. As illustrated in the experiments the proposed method speeds-up the original inpainting algorithm significantly.

Image inpainting, texture synthesis, image completion, exemplar, search space reduction