关键词: Criminisi算法, 基于样本, 图像修复, 边界, 区域搜索, 自适应模板

Abstract: The exemplar-based image inpainting algorithm proposed by Criminisi and his partner, which uses exhaustive searching to get the best-matching patch and adopts fixed-size patch to fill the image, will cause error matching and information extension. For this problem, based on the decisive role of a pixel’s neighboring block in the pixel’s value and the importance of structure information, this paper presents an image inpainting algorithm based on regional searching and adaptive template to enhance the local information harmony and boundary recovery ability, so it can improve the whole inpainting effect. A number of experiment results demonstrate that the improved method not only reduces the inpainting time and keeps the structure better, but also has a better visual effect.

Key words: Criminisi, exemplar-based, image inpainting, boundary, regional searching, adaptive template