基于边界特征耦合惩罚因子的图像修复算法  

Image inpainting algorithm based on boundary characteristic coupling penalty factor

作　　者：王春华 韩栋[1] 

机构地区：[1]黄淮学院信息工程学院

出　　处：《电子测量与仪器学报》2018年第1期96-102,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(60073057);河南省科技攻关计划(172102210117);驻马店市科技计划(17135)资助项目

摘　　要：为了解决当前较多图像修复算法主要是通过全局搜索的方法来完成图像修复,导致其效率不高,以及修复图像中易出现块效应等不足,提出了基于边界特征耦合惩罚因子的图像修复算法。首先,利用待修复块的边界特征来建立边界因子,通过该因子构造优先权模型,以度量待修复区中的像素点的优先权,确定优先修复块。然后,利用像素点的R、G、B分量,通过C(p)均值聚类方法对图像进行分割聚类,以待修复块所覆盖的子块为依据,确定最优匹配块的搜索范围。最后,以像素点的调用次数为依据,构造惩罚因子,并将该因子与像素点对应的欧式距离进行联合,构造匹配度模型,完成最优匹配块的搜索,从而实现图像修复。实验结果显示,与当前图像修复算法相比,所提算法具有更高的修复质量与效率,能够较好的克服修复过程中产生的块效应等不良现象。所提算法具有较好的图像复原能力,可适用于大面积损坏图像的重构。In order to solve the defects as low efficiency and block effects induced by using the global search method to complete the image restoration in current image inpainting algorithms,image inpainting algorithm based on boundary characteristic coupling penalty factor is proposed in this paper. Firstly,the boundary factor is established by using the boundary feature of the patch to be repaired,and the priority model is constructed by this factor to measure the priority of the pixel in the inpainting area for determining priority repair blocks. Then,the image was segmented and clustered by using the R,G and B components and K-clustering method,and the search range of the optimal matching block is determined according to the sub-blocks. Finally,the penalty factor is constructed on the basis of the number of calls to the pixels,and the matching degree model is constructed combining the penalty factor and the Euclidean distance corresponding to the pixel point to finish image restoration. The experimental results show that the proposed algorithm has higher repair quality and efficiency which can better overcome the artifacts such as block effect in the repair process compared with the current image inpainting algorithm. The proposed algorithm has better image restoration ability and can be applied to reconstruct large area damaged images.

关 键 词：图像修复 边界特征 K均值聚类方法 惩罚因子 欧氏距离 匹配度模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]