基于邻域关联因子耦合信息度量规则的图像修复算法  

Image inpainting algorithm based on neighborhood correlation factor coupled with information metrics rule

作　　者：常国锋[1] 许利军[2] CHANG Guofeng;XU Lijun(School of Computer and Information Engineering,Xinxiang College,Xinxiang,Henan 453003,China;School of Computer Science,Wuhan University of Technology,Wuhan,Hubei 430063,China)

机构地区：[1]新乡学院计算机与信息工程学院,河南新乡453003 [2]武汉理工大学计算机学院,湖北武汉430063

出　　处：《山东科技大学学报（自然科学版）》2021年第1期100-108,共9页Journal of Shandong University of Science and Technology(Natural Science)

基　　金：河南省自然科学基金项目(182300410258);河南省科技攻关项目(182102210494);河南省高等学校重点科研项目(15A520094)。

摘　　要：针对当前图像修复算法忽略图像纹理信息的变化度而导致修复结果中存在间断及振铃效应等不足,提出了基于邻域关联因子与信息度量规则的图像修复算法。首先,将待修复块与其邻域块的归一化互相关值引入到优先权的计算中,以构造邻域关联因子,并将其与置信度以及数据项结合计算优先权,从而获取优先修复块。利用图像块对应的均值和方差特征,建立信息度量规则,根据纹理信息的变化度对样本块的大小进行调节。最后,引入误差平方和函数,对待修复块与匹配块的相似度进行测量,获取最优匹配块,利用最优匹配块中的像素点对待修复块进行填充,实现图像的修复。通过实验结果发现,所提算法具备更高的修复质量。In order to overcome the deficiencies such as discontinuity and ringing effect in repair results induced by ignoring the variation of image texture information in current image inpainting algorithm,an image inpainting algorithm based on neighborhood correlation factor and information metrics rule was proposed in this paper.Firstly,the normalized cross-correlation values of the blocks to be repaired and their neighborhood blocks were introduced into the calculation of priority to construct the neighborhood correlation factor.This factor was then combined with confidence and data items to calculate the priority so as to obtain the priority repaired blocks.The information metrics rules were established using the mean and variance characteristics of image blocks,and the size of sample blocks was adjusted according to the variation of texture information.Finally,the error square sum function was introduced to measure the similarity between the repair block and the matching block,and the optimal matching block was obtained.The repair block was filled with the pixels in the optimal matching block to achieve image restoration.The experimental results show that the proposed algorithm has higher repair quality.

关 键 词：图像修复 邻域关联因子 优先权计算 信息度量规则 样本块 误差平方和函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]