Abstract: Using exemplar-based inpainting method to search the source patch globally may lead to error match, and such error accumulates and propagates to other areas quickly, even makes the results completely unacceptable. This paper introduces an algorithm for digital inpainting based on color region segmentation. Patch matching is restricted within the relevant regions where the current patch overlaps. Experimental results show that the method can efficiently handle different cases especially large regions in complex background by producing visually more plausible effect.

Key words: image inpainting, exemplars match, color region segmentation

摘要： 现有的基于样本块纹理合成的数字图像修复技术在全局式搜索样本匹配块时可能选择错误的匹配块，导致繁衍错误信息而使修复结果不理想。该文提出一种基于颜色区域分割的图像修复方法，其目标匹配块的搜索限定在源样本块所覆盖的颜色区域中。实验结果表明这是一种更健壮、视觉效果更理想的修复算法，可以有效修复各种破损情况，尤其是具有复杂背景的较大区域填充。

关键词: 图像修复, 样本匹配, 颜色区域分割

CLC Number: 

• TP312