Abstract: For the image inpainting algorithm based on examplar，proposed by Criminisi and his partner，has some deficiencies in determining matching block size，this paper presents an improved algorithm.To obtain a more accurate match，this algorithm determines the matching block size according to the size of texel sample，and uses multi-level synthesis method to improve the repair result，it makes up the deficiency of structural fracture during Criminisi algorithm.The experiment results show this method is effective in repair structured texture images.

Key words: sample, texture synthesis, image inpainting, texel, multi-level synthesis

摘要： 针对Criminisi等人提出的基于样本的图像修复算法在确定匹配块大小时存在的一些不足，提出了一种改进的基于样本的图像修复算法。该算法根据样本的纹元大小确定其匹配块大小，使得匹配更为准确，并利用多级合成提高修复效果，弥补了Criminisi算法修复时出现的结构断裂缺陷。实验结果表明，该方法在修复强结构纹理图像时有很好的效果。

关键词: 样图, 纹理合成, 图像修复, 纹元, 多级合成