Abstract: Although the Total Variation（TV） model is good at maintaining the edge of damaged images and reducing numerical calculation in image inpainting，it needs more improvement in the vision connectivity.A new image multi-level-inpainting method based on TV model and texture synthesis is proposed.The TV model is used to restore the shape information of missing area in an image.The inpainting result visual connectivity is improved by the CDD model.To the large texture area of the inpainted image，the patch based texture restoration model is used to process the obvious trace insulted from two previous levels inpainting.Experimental results show that the inpainting performance is better both in the structure and the texture.Moreover，the approach can be used in the image inpainting with bigger damaged area.

Key words: image inpainting, Total Variation（TV）, Curvature Driven Diffusions（CDD） model, multi-level inpainting, texture synthesis, Mean Square Errors（MSE）

摘要： 整体变分（TV）模型在图像修复中能够保持图像的边缘且数值实现方便，但在图像修复中对于人类视觉连通性的处理还有所不足。根据图像遗失或者损坏的不同类型，针对TV、CDD模型在图像修复中存在的问题，提出了一种结合TV、CDD模型及基于块的纹理合成算法的分层图像修复算法。实验结果表明，这种分层修复的方法在图像的结构修复和纹理修复两方面实现了较好的统一，而且在较大区域图像修复上表现出良好的效果。

关键词: 图像修复, 整体变分（TV）, 曲率驱动扩散（CDD）模型, 分层修复算法, 纹理合成, 最小均方误差（MSE）