基于SSIM相似度的图象修复  

Image inpainting based on SSIM similarity

作　　者：朱弋 刘华明 ZHU Yi;LIU Huaming(School of Computer and Information Engineering,Fuyang Normal Uinversity,Fuyang Anhui 236037,China)

机构地区：[1]阜阳师范大学计算机与信息工程学院,安徽阜阳236037

出　　处：《阜阳师范大学学报（自然科学版）》2020年第4期61-65,共5页Journal of Fuyang Normal University:Natural Science

基　　金：国家自然科学基金项目(61772430)资助。

摘　　要：利用样本块技术修复图象时,存在失误率较高和算法运行速度过慢等弊端,本文提出基于SSIM相似度的图象修复方式。首先以Patchmatch算法为基础,修改相似度的定义,用插值方法自动填补在每一层金字塔的空缺;再引入SSIM相似性度量方式,使误差判断标准更具合理性;最后重新设置传播方式,在保证时间效率的同时提高图象修复质量。结果表明本文提出的方法在修复结果上具备较好的视觉连通性,在运算效率上,修复时长也有所降低。When using sample patch technology to repair images,there are some disadvantages,such as high error rate and slow running speed of the algorithm,this paper proposes an image inpainting method based on SSIM similarity.First of all,based on patchmatch algorithm,the definition of similarity is modified,and the gap in each layer of pyramid is filled automatically by interpolation method;Secondly,SSIM similarity measurement is introduced to make the error judgment standard more reasonable;Finally,the propagation mode is reset to ensure the time efficiency and improve the quality of image inpainting.The results show that the proposed method has good visual connectivity in the restoration results,and the repair time is also reduced.

关 键 词：图象修复 相似度 SSIM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]