高可信度图像修复方法  

Improved exemplar-based image completion algorithm of high reliability

作　　者：汪强[1] 邹北骥[1] 朱建凯[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《计算机应用》2010年第5期1331-1332,1401,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60970098)

摘　　要：在基于样例图像修复思想的基础上,从填充次序与目标块搜索两个方面进行改进,提出了一种图像修复方法。分析了Criminisi等人提出的修复算法,引入混淆系数来确定目标块的填充次序,以图像源区域中"有且只有一个"块与目标块相似来定义可信度,并优先填充可信度高的目标块。混淆系数的计算自然地将寻找高优先级目标块、搜索对应的匹配块两个过程统一,以避免误差累积。通过自然图像移除大面积物体的比较实验,表明算法适用于具有复杂背景的较大区域修复,视觉效果理想。An image completion method based on the exiting exemplar-based image completion idea was proposed by improving the filling order and target patch matching process.The algorithm proposed by Criminisi was analyzed,and a new confusion measuring was used to establish the filling order of the target patch,which considered ＂having one and the only one＂ when computing the similarity within patches and filled the target patch with highest reliability.The process of computing confusion coefficient integrated selecting the highest priority target patch with searching the corresponding matching patch to avoid cumulative error.The experiments for large object removal on real images show that the proposed method can efficiently handle large regions of complex background by producing more plausible effect.

关 键 词：图像修复 纹理合成 优先权计算 混淆系数 样本匹配 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]