一种基于Poisson方程的分离型图像修复方法  

An isolated image inpainting approach based on the Poisson equation

作　　者：邵肖伟[1] 刘政凯[1] 李厚强[1] 

机构地区：[1]中国科学技术大学信息处理中心,安徽合肥230027

出　　处：《电路与系统学报》2008年第6期1-6,共6页Journal of Circuits and Systems

基　　金：国家自然科学基金重点项目(60333020)

摘　　要：图像修复是一种恢复图像中损坏部分的技术,具有广泛的应用。基于采样复制的修复方法对纹理图像有较好的效果,但容易产生块效应,而且对结构信息的修复可能产生较大的偏差。本文提出了一种基于Poisson方程的分离型修复算法,首先将原图分解为结构图像和纹理图像两部分,然后根据其特性分别进行修复,叠加后得到最终的修复结果。对结构图像使用Laplacian算子强化结构信息,然后对Laplacian场进行修复并使用Poisson方程重建,可以同时保持锐利的区域边界以及平滑的区域背景。实验表明该方法可以有效改善修复的视觉效果,对大区域修复也有良好的表现。Image inpainting is a technique used to recover damaged portions of an image and has a variety of applications. The exemplar based inpainting algorithm has a good performance on texture inpainting, but it may produce block effects and lead to artificial inpainting results when processing structure images. In this paper, we propose an isolated inpainting algorithm based on the Poisson equation. First the image to be repaired is decomposed into two components： the structure and texture images; then the two components are repaired separately by different methods according to their image characteristics; finally, the repaired components are combined back. For the structure image, the Laplacian operator is employed to enhance the structure information, and the Laplacian field is inpainted by the exemplar based algorithm, followed by a reconstruction based on the Poisson equation. In this way, the repaired structure image maintains both sharp edges and smooth interiors. Experiments show the proposed algorithm can improve the visual effect significantly and it has a good performance on large area inpainting.

关 键 词：图像修复 分离型 POISSON方程 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]