基于分层索引的高维数据对象检索  

High-dimensional Data Object Retrieval Based on Hierarchical Index

作　　者：黄颖[1] 张豹 陈伟荣[1] 戴鹏 HUANG Ying;ZHANG Bao;CHEN Weirong;DAI Peng(The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210007,China;Department of Computer Science and Engineering,Southeast University,Nanjing 211189,China)

机构地区：[1]中国电子科技集团公司第二十八研究所,南京210007 [2]东南大学计算机科学与工程学院,南京211189

出　　处：《指挥信息系统与技术》2019年第6期81-85,共5页Command Information System and Technology

基　　金：装备发展部“十三五”预研课题资助项目

摘　　要：随着海量信息检索技术的发展,对文本、图片和视频等高维数据对象的相似性检索要求不断提高。局部敏感哈希(LSH)是解决高维数据近邻检索的主要方法之一,但存在索引存储代价高及查询效率低等问题。提出了一种基于二级混合索引模型构造方法,先利用溢出树(Spill tree)对数据集进行划分,再对每个部分构建基于LSH的哈希表,形成混合索引,支撑高维数据检索。试验表明,该方法缩小了高维数据对象的索引存储空间,提高了查询效率和查询质量。With the development of mass information retrieval technology,requirements for similarity retrieval of high-dimensional data objects such as text,pictures and videos are constantly increasing.The locality sensitive hashing(LSH)is one of the mainstream methods for solving high-dimensional data neighbor queries.However,this method has the problems of high index storage cost and low query efficiency.Aiming at the shortcomings of existing LSH-based algorithm s,a new two-level hybrid index structure is proposed,it first uses the spill tree to divide the data set,and then constructs an LSH-based hash table for each part.It forms a hybrid index to support high-dimensional data retrieval.Experiments show that the method reduces the index storage space of high-dimensional data objects and improves the query efficiency and quality.

关 键 词：高维数据对象 近邻检索 局部敏感哈希 空间划分 

分 类 号：P301[天文地球—地球物理学]