计算机应用 ›› 2010, Vol. 30 ›› Issue (05): 1331-1332.

• 图形图像处理 • 上一篇    下一篇

高可信度图像修复方法

汪强1,邹北骥2,朱建凯3   

  1. 1. 长沙学院计算机系
    2. 中南大学信息科学与工程学院
    3. 长沙学院
  • 收稿日期:2009-11-12 修回日期:2009-12-25 发布日期:2010-05-04 出版日期:2010-05-01
  • 通讯作者: 汪强
  • 基金资助:
    国家自然科学基金资助项目

Improved exemplar-based image completion algorithm of high reliability

  • Received:2009-11-12 Revised:2009-12-25 Online:2010-05-04 Published:2010-05-01

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。

关键词: 图像修复, 纹理合成, 优先权计算, 混淆系数, 样本匹配

Abstract: 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.

Key words: image completion, texture synthesis, priority computation, confusion coefficient, exemplars match