【作者】 孙帮勇；

【导师】 周世生；

【作者基本信息】 西安理工大学 ， 制浆造纸工程， 2006， 硕士

【摘要】 为了恢复受损的图像，须对其受损区域进行修描，以满足人们对高质量印刷复制技术的需求，本课题针对该类图像中的结构型破损图像设计了一种基于水平集方法的图像修描技术。该技术可以应用到考古、艺术、军事、广告等领域，它既提高了图像修描的质量，同时也改善了图像的视觉效果，尤其对文物保护和历史资料复原具有十分重要的意义。本文总结了图像处理理论在图像修描中的应用，通过分析人类视觉系统中的心脑认知机制，总结出其对图像修描的一些启发性原则，将一般的图像恢复理论应用到图像修复领域，以最佳猜测原理和贝叶斯结构体系为理论框架，建立了基于偏微分方程和水平集方法的图像修描模型。本文将修描过程视为修描边界曲线的向内演化过程，在修描边界向内收缩的过程中完成对待修描区域的恢复。同时，我们选取待修描区域边界外部的一系列点来修补图像，这样就避免了待修描区域内的点对修描结果的干扰。修描公式中选取了合适的权重，保证了该修描技术与手工修复图画的过程很类似，从而可以达到较好的修描结果。试验结果表明：与现在流行的一些修描算法相比，本文使用的快速步进算法应用比较简单，并且修描效果较好，运行时间很短，特别是在其实用性方面有了很大的提高。在文章的最后，作者总结了本文设计的图像修描技术存在的问题和不足，并对以后的研究方向做出了说明。更多还原

【Abstract】 The damaged area must be inpainted in order to restore the damaged image and meet people’s demand of high-quality printing. In this paper, a kind of technology on image inpainting based on level set is provided, which combined the technology of structure. The fruits of the research on the technology can be applied to such fields as archaeological fields, art, military, and advertisement. It can improve the quality of image inpainting and improve people’s visual effect, and it especially has very important significance on protecting historical relic and restoring historical materials.This paper extends the image processing theory to image inpainting field. After analyzing the heuristic principles of the cognitive mechanism in the human visual systems, we apply the image restoring theory into image inpainting technology and establish the image inpainting model based on PDE and level set method according to principle frame of best guess theory and Bayesian structure system. We treat the missing regions as level sets and use the fast marching method (FMM) to propagate the image information. The boundary of the missing region is seen as 0-level set, and in the course of evolution, the region was inpainted. And we select appropriate weighting function to ensure the course is similar to the inpainting by hand.Although the inpainting basics are straightforward, most inpainting techniques published in the literature are complex to understand and implement. Our algorithm based on fast marching method is very simple to implement, fast, and produces nearly identical results to more complex, and usually slower, known methods. At last, we summary the problems of image inpainting at present and discuss the research direction in the future.更多还原