为了提高非均匀纹理图像大区域修复效果，提出了一个分层Criminisi 图像修复算法。首先采用多尺度变分分解模型将图像分解成一系列图层之和，不同图层包含不同尺度的图像特征，而同一图层包含几乎相同的尺度特征。然后在每个图层中分别采用Criminisi 算法进行修补。由于同一图层包含尺度大致相同的图像特征，所以在匹配块的搜索过程中，分层修复能较容易地搜寻到最优的匹配块。最后结合分别修复的各个图层，得到最终的修复结果。实验结果表明，对于不同的非均匀纹理人工图像和自然图像，本文模型都能取得较为满意的修复结果。

To improve the inpainting quality of non-uniform texture images with extensive missing regions, we propose a multi-layer Criminisi image inpainting algorithm. First, a damaged image is decomposed into the sum of a few image layers using a multiscale variational decomposition model. The different image layers contain image features with different scales, and one layer contains features that are under a similar scale. Then, in each image layer, the Criminisi inpainting algorithm is applied. In the matching block search process, because one layer contains image features with roughly the same scale, the multi-layer inpainting algorithm can more easily locate the optimal matching block. Finally, combining the filled image layers, we obtain the final inpainting result. Experimental results on both synthetic and real non-uniform texture images demonstrate the effectiveness of the proposed algorithm.