有监督鉴别哈希跨模态检索  

SUPERVISED DISCRIMINATIVE CROSS-MODAL HASHING

作　　者：朱治兰 荆晓远[1] 董西伟[1,2] 吴飞[1] Zhu Zhilan;Jing Xiaoyuan;Dong Xiwei;Wu Fei(College of Automation , Nanjing University of Posts and Telecommunications, Nanjing 210023 , Jiangsu, China;School of Information Science and Technology , Jiujiang University, Jiiijiang 332005, Jiangxi, China)

机构地区：[1]南京邮电大学自动化学院,江苏南京210023 [2]九江学院信息科学与技术学院,江西九江332005

出　　处：《计算机应用与软件》2019年第4期223-229,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61702280);江苏省研究生创新工程项目(CXLX11_0418)

摘　　要：随着大数据时代的到来,利用哈希方法实现对异质多模态数据的快速跨模态检索受到越来越多的关注。为了获取更好的跨模态检索性能,提出有监督鉴别跨模态哈希算法。利用对象的标签信息对所要生成的哈希码进行约束。算法中的线性分类项和图拉普拉斯算子项分别用于提升哈希码鉴别能力和保留模态间相似性。对算法的目标函数利用迭代法进行求解。该算法在两个基准数据集的实验结果展现出优于目前最前沿的跨模态哈希检索方法。With the advent of the era of big data, applying Hash methods for heterogeneous multimodal data to achieve fast cross-modal retrieval is receiving more and more attention. In order to obtain better cross-modal retrieval performance, we proposed supervised discriminative cross-modal hashing (SDCH). We used label information of objects to constrain the hash code to be generated. The linear classification term and the graph Laplacian term in the algorithm were used to improve the discriminating ability of hash codes and preserve inter-modal similarity, respectively. The objective function of the algorithm was solved in an iterative method. Related experiments have been conducted in two benchmark databases. Experimental results show that the algorithm is superior to state-of-the-art cross-modal Hash methods.

关 键 词：跨模态检索 哈希 标签信息 线性分类项 图拉普拉斯项 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]