计算机应用研究2017，Vol.34Issue(12)：3861-3865,3907,6.DOI:10.3969/j.issn.1001-3695.2017.12.077
基于奖励机制和互斥条件的图像对象协同定位
Reward strategy based image object co-localization with mutex constraint
摘要
Abstract
With the increasingly abundant of images shared on the Internet,the localizing and retrieving of generic object become growing demand.In order to apply the Joint philosophy to localization,this paper proposed an reward strategy based object co-localization method with mutex constraint.Co-localization handled the problem of respectively localizing the common class of object in a set of images.To unsupervised locate the objects which had intra-class variation and inter-class diversity,it formulated a sub set of candidates generated by objectness in a graph structure which treated saliency from background and similarity between each other as reward,together with mutex constraints that could prevent selecting the overlapping region.It performed an evaluation of this method compared to previous approaches on the object discovery datasets and MSRC data set,and got excellent result.关键词
协同定位/奖励机制/互斥条件/图结构/无监督Key words
co-localization/reward strategy/mutex constraint/graph structure/unsupervised分类
信息技术与安全科学引用本文复制引用
李劲松,甘朝晖,蒋旻..基于奖励机制和互斥条件的图像对象协同定位[J].计算机应用研究,2017,34(12):3861-3865,3907,6.基金项目
国家自然科学基金资助项目(41571396) （41571396）
国家大学生创新创业训练基金资助项目(201410488017) （201410488017）
