基于样本块的图像修复方法的改进  

Improvement of image inpainting based on sample patch

作　　者：陈龙[1,2] 熊辉[1,2] 汪继文[1] 

机构地区：[1]安徽大学计算机科学与技术学院,合肥230601 [2]安徽大学计算机科学与技术创新实验室,合肥230601

出　　处：《计算机应用》2011年第A01期47-49,共3页journal of Computer Applications

基　　金：安徽省自然科学基金资助项目(090412045);2009年国家大学生创新实验项目(091035716)

摘　　要：基于纹理合成的图像修复算法适合大面积信息缺损区域的修复。基于Criminisi算法,通过对数据项的重新定义,从结构和纹理上对图像进行修复。改进算法在搜索最佳样本块时,考虑到样本块所含已知像素点的信息量对寻找最佳样本块的影响,构造了包含信息量的最优搜索式;为了平滑置信因子更新导致的误差传播,定义了基于曲率的置信因子更新方程。基于上述构造,经多次实验,改进算法得到合适参数,避免了明显冗余像素点的产生并使得纹理连接处连贯。Image inpainting algorithm based on texture synthesis is suitable for large area image restoration.Based on Criminisi＇s algorithm,this paper redefined the data item,combined the structure and synthesis to restore a picture.The information that sample patch contains was considered when searching the best sample patch and thus the optimal search-type formula was obtained.In order to smooth the propagation of error when updating the confidence term,a new formula based on curve rate was defined.Based on previous mentioned structure and after many experiments,best fit parameters were got,with which the visible redundant pixels could be avoided and a coherent connection was achieved.

关 键 词：样本块 纹理信息 等照度线 冗余 曲率 峰值信噪比 

分 类 号：TP391.413[自动化与计算机技术—计算机应用技术]