基于Gist特征的场景图像修复算法  

Scene image inpainting algorithm based on Gist feature

作　　者：李策[1] 任辈杰 潘峥嵘[1] 朱翔[1] LI Ce;REN Bei-jie;PAN Zheng-rong;ZHU Xiang(College of Electrical and Information Engineering,Lanzhou Univ.of Tech.,Lanzhou 730050,China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050

出　　处：《兰州理工大学学报》2018年第5期102-107,共6页Journal of Lanzhou University of Technology

基　　金：国家自然科学基金(61866022);甘肃省基础研究创新群体项目(1506RJIA031)

摘　　要：传统图像修复算法中存在修复信息仅依赖图像自身,缺乏多样性与有效性,很难对区域进行合理可靠的修复,以及待修复区域选择出现的标定不可更改或者人为标定随机性等问题.针对上述问题提出一种基于Gist特征的场景图像修复算法.首先,采用一步点击的操作过程,快速获得待修复区域,增加了待修复区域的选择性,减少了人工标定区域时的随机性;其次,采用包含Gist特征在内的多特征约束进行分步筛选,从数据库中筛选出与修复区域具有相似信息的备选场景图像,在筛选出的备选场景图像中利用全局滑动窗获得与待修复区域最相似的图像块,该筛选过程进一步提高了修复结果的可靠性;最后,优化过程中,采用泊松融合消除备选场景图像块与待修复场景图像合成过程中的边缘效应,获得最终修复结果.实验对比结果表明,本文场景图像修复算法可以较好地对场景图像进行修复,而且使得修复结果更符合人类视觉感知效果.There are some obvious artfacts in traditional image inpainting results due to several aspects the algorithms which some depend on the image itself,some lacks in diversity and effectiveness to process rational and reliable results,even conduct unchangeable or random selection by people.Aimed at the problems mentioned above,a scene image inpainting algorithm is put forward based on Gist feature.Firstly,one-click procedure is used to obtain quickly the inpainting area,so that the selectivity of the area is increased and the randomness of man-made area calibration decreases.Secondly,a multi-feature constrain which Gist feature included is adopted to conduct stepwise sieving of scene image candidates from the image database with similar information to the inpainting area and then a global sliding window is used to get the matched image block with most similarity from the sieved scene image candidates,so that the reliability of the inpainting result is further improved with this sieving process.Finally,a Poisson blending is used to eliminate the edge effect between the scene image block of the image candidate and the surrounding area of the target region.The result of comparison experiment shows that the proposed algorithm can be used to conduct better result and makes it more consistent with human visual perception.

关 键 词：图像修复 Gist特征 场景图像 全局滑动窗 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]