Abstract: The restoration of Tang Dynasty tomb mural is a two-step process, one is crack detection and the other is image inpainting. While the first step offers the prior image model to the second one, the area which marked is what will be wanted to be filled in. According to the standard of diseases to mural, this paper uses the morphological transformation which includes top-hat and below-hat to detect the cracks of mural by multi-scale morphological transformation, and to get the mask of checking, fissures, and fractures. At the part of inpainting, it improves the order of filling of CDD algorithm, replaces orthogonalizable diffusion with cross diffusion. This method is applicable to Tang Dynasty tomb murals.

Key words: morphological transformation, Curvature-Driven Diffusions（CDD）, multi-scale detection, inpainting

摘要： 古代壁画虚拟修复包括病害标记与缺损信息填充两部分。病害标记结果为虚拟修复建立先验模型，标记区域为图像信息缺损部分，未标记区域为有效信息源，修复过程为有效信息向缺损区域扩散的过程。依据唐墓室壁画的病害分类标准，采取多尺度病害标记，利用形态学高帽、低帽算子设计多尺度结构元素对病害进行标记，形成龟裂、裂隙、裂缝三种病害的分布图，并产生相应的掩膜。以此掩膜图像为先验模型，改进CDD算法的信息扩散方式，用交叉扩散代替正交扩散以适应带有确定方向性的信息缺失区域被合理填充，该方法能实现唐墓室壁画的多尺度修复。

关键词: 形态学变换, 曲率扩散, 多尺度标记, 图像修复