Abstract: In order to solve the problems that the priority function of Criminisi algorithm is unreliable, the matching block traversal search spends a lot of time and when selecting matching block the similarity function is inaccurate, this paper presents an improved Criminisi image restoration algorithm. The algorithm improves the priority formula, introduces variance difference term to enhance the distinguish ability on image structure section. When searching for a matching block area, it introduces a parameter, it can shorten the seeking time of matching block effectively. And a new similarity function of repaired block and matching block is defined. The experimental results show that compared with the original algorithm, this improved Criminisi algorithm is more reliable and effective in image restoration.

Key words: image inpainting, texture synthesis, priority function, similarity function

摘要： 针对Criminisi算法存在的优先权函数不可靠，匹配块遍历搜索时间复杂度大，以及选取匹配块时相似度函数不准确的问题，提出了一种改进的Criminisi图像修复算法。该算法改进了优先权计算公式，引入了方差差值项，增强了对图像结构部分的辨别能力。在搜寻最佳匹配块时，引入一个比例系数，从而有效缩短了匹配块寻求时间，降低时间复杂度。且定义了新的待修复块与匹配块之间的相似度函数。通过实际修复实验证明，改进后的Criminisi算法的修复结果更加可靠和有效。

关键词: 图像修复, 纹理合成, 优先权函数, 相似度函数