基于显著结构的图像修复  

Image Inpainting Using Salient Structure

作　　者：赵洋 ZHAO Yang(College of Computer Science,Sichuan University,Chengdu 610065)

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《现代计算机》2020年第9期49-53,共5页Modern Computer

摘　　要：针对基于样本块的经典算法Criminisi算法在实际计算中有置信项减小,数据项逐渐失效的问题,提出一种改进方案。即改进置信项,增添显著结构项;将已经修复好的像素块作为候选块与待修复块匹配;采用非局部纹理信息匹配候选块与待修复块。实验表明,相较于改进前的算法,提出的算法效果有较大的提升。An improved scheme is proposed to solve the problem that the Exemplar-based classical Criminisi algorithm reduces the confidence term and the data term gradually fails in the actual calculation;the confidence term is improved and the salient structure term is added;the re paired pixel blocks are regarded as the candidate block during the repairation;Non-local texture information is used to match the candidate block with the block to be repaired.The experimental results show that the proposed algorithm is more effective than the previous one.

关 键 词：图像修复 样本块 显著边缘 优先级 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]