摘要
在研究Criminisi等人的基于样例的修复算法基础上,提出了一种快速修复算法。该算法采用梯度数据项连同置信度确定填充顺序,使得优先权的计算更加准确;并采用局部窗口搜索的策略,加快了修复速度;最后利用颜色和梯度共同决定相似性,使得修复后的图像具有更好的视觉效果。大量实验结果表明,该算法不仅能够快速有效地修复受损图像,而且能够很好地保持受损区域的纹理和结构。
The exemplar based method by Criminisi et al.is studied.After that,a fast image completion algorithm is proposed.This algorithm firstly combines gradient data with confidence to determine the filling order,which can make a more accurate calculation of priority;and then use a local window searching strategies to accelerate the repair rate;finally,apply the color and gradient value to determine the similarity,which can make the visual effect of restored image better.Experimental results show that the proposed algorithm can not only repair damaged images quickly and effectively,but also maintain a good texture and structure of the damaged region.
出处
《微型机与应用》
2010年第22期34-36,39,共4页
Microcomputer & Its Applications
基金
国家自然科学基金青年基金项目(60802040)
四川省教育厅青年基金项目(07Zd1114)
关键词
图像修复
目标去除
纹理
结构
image completion
object removal
texture
structure
