Abstract: When computing the order of target patch priority, Criminisi algorithm has defect, to solve this problem, an improved inpainting algorithm is proposed. To establish a new function about priority, the proposed algorithm takes into account local feature of image—curvature and gradient, and uses the information of curvature and gradient as a data item. Thus, the order of filling inpainting is more reliably. The experimental results show that the proposed algorithm overcomes the problem of excessive diffusion from high texture area to low area, and obtains better visual appearance.

Key words: image inpainting, priority, local feature, curvature, gradient

摘要： 针对Criminisi算法计算目标块填充优先权等级时存在缺陷的问题，提出了一种改进的修复算法，方法在确立新的优先等级函数时，充分考虑图像的局部特征信息——曲率和梯度，将曲率及梯度信息作为优先权值的数据项，从而获得更加可靠的填充修复顺序。实验结果表明，和Criminisi算法相比，该方法克服了修复过程中高纹理区域向低纹理区域过度扩散的问题，并取得了更加理想的视觉修复效果。

关键词: 图像修复, 优先权, 局部特征, 曲率, 梯度