为了克服传统数字壁画修复方法主要依赖图像先验信息且难以修复大面积损坏的局限性，提出一种新颖的壁画修复框架。通过构建一种轮廓先验信息引导的扩散概率模型，来实现全面修复。修复框架利用轮廓和破损信息作为引导信号，逐步扩散和重构缺失部分，同时为了解决颜色失调的问题，使用一种潜变量扩散概率模型，该模型通过整合全局颜色信息，在修复过程中动态调整和校正颜色，使修复结果既保留了原有风格，又避免了局部颜色偏差。该修复框架不仅能够实现对大面积破损区域的精准修复，还能有效恢复壁画的色彩和纹理细节。使用莫高窟的壁画数据集评估了修复框架的有效性，对不同程度破损的壁画修复结果进行定性和定量的实验评估，结果表明该方法在壁画修复任务中具有显著优势，可见该方法为壁画的保护和数字修复提供了一种有效的解决方案。

In order to overcome the limitations of traditional digital mural restoration methods that primarily rely on image prior information and are inadequate for restoring large areas of damage, a novel mural restoration framework was proposed. Comprehensive restoration was achieved by constructing a diffusion probabilistic model guided by contour prior information. In the framework, contour and damage information were employed as guiding signals, and the missing portions were gradually diffused and reconstructed. In order to address the issue of color misalignment, a latent variable diffusion probabilistic model was employed, in which global color information was integrated to dynamically adjust and calibrate colors during the restoration process. As a result, the restoration outcomes were ensured to retain the original style while avoiding local color deviations. This restoration framework was capable of precisely restoring large damaged areas and effectively recovering the mural's color and texture details. The effectiveness of the restoration framework was evaluated using a mural dataset from the Mogao Caves, with qualitative and quantitative experimental assessments of restoration results for murals with varying degrees of damage. The results show that significant advantages in mural restoration tasks are demonstrated by the proposed method. It is concluded that this approach provides an effective solution for mural preservation and digital restoration.