一种改进的Criminisi图像修复算法  

AN IMPROVED CRIMINISI ALGORITHM FOR IMAGE INPAINTING

作　　者：常晨[1] 尹立新[1] 方宝龙[1] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250100

出　　处：《计算机应用与软件》2012年第9期240-242,267,共4页Computer Applications and Software

摘　　要：图像修复的目的是填补有信息缺损的图像,并使观察者无法察觉出图像的填补痕迹。在Criminisi提出的算法基础上进行改进,提出一种图像修复方法。引入曲率来决定目标块的填充次序和最佳匹配块的选择,并改进优先权为各项加权和,通过改变权值可以得到更好的修复效果,同时避免了由于置信度迅速衰减带来的错误填充次序。大量实验结果表明,提出的算法不仅增强了图像结构的连续性,并使得匹配更加准确,对于各种类型图像的修复都取得了令人满意的视觉效果。Image inpainting technique aims at infilling the images with missing or damaged portions in a way that they will be non-detectable for an observer.An improved image inpainting method based on the algorithm brought forward by Criminisi is proposed in the paper.In it the curvature is introduced to determine the filling order of the target patch and the selection of the best matching patch,and the priority is modified to the sum of each weighting.By varying the weight value,better inpainting result can be obtained.Meanwhile,the incorrect filling order caused by rapid decay of confidence is avoided as well.Lots of experimental results show that the proposed method can enhance the continuity of the image structure and make more accurate matching as well as have achieved satisfactory visual effects for different types of image inpainting.

关 键 词：图像修复 纹理合成 优先权 曲率 等照度 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]