一种基于区域搜索的快速图像修复算法  

A Fast Image Inpainting Algorithm Based on Local Search

作　　者：张巧焕[1] 唐向宏[1] 任澍[1] 

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2011年第5期139-142,共4页Journal of Hangzhou Dianzi University：Natural Sciences

摘　　要：基于纹理合成的图像修复技术在修复大面积破损区域时,普遍存在时间复杂度高的问题。针对Criminisi等提出的基于样本图像修复算法中的匹配技术存在的问题。该文提出了一种根据图像破损区域尺寸进行快速匹配的算法。同时,为了强调前次修复对以后修复产生误差累积的影响,提出了新的置信度更新方法。实验结果表明,改进算法对图像结构边缘的修复有明显的改善,也实现了图像修复的实时性。Image inpainting techniques based on texture synthesis can be used to fill large region with high time complexity.For the problem of matching technology of exemplar-based image inpainting proposed by Criminisi and his partner,this paper proposes a fast matching algorithm according to the damaged area.Meanwhile,to emphasize the inpainting result of last time influences the later inpainting and will cause error accumulation,this paper also presents a novel method that updates confidence.Experiments results show that the improved approach improves the result in the front and obtains real-time performance.

关 键 词：纹理合成 图像修复 匹配技术 置信度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]