基于非固定长度散列表的无监督式海明距离搜索  

Unsupervised Hamming Distance Search Based on Unfixed Length Hash List

作　　者：李明兰 LI Minglan(College of Mathematics and Statistics,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学数学与统计学院

出　　处：《重庆理工大学学报（自然科学）》2020年第1期71-76,134,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：山东省杰出青年科学基金项目(JQ201419);山东省自然科学英才基金项目(ZR2015FM023)

摘　　要：为提高大规模数据的检索效率,提出一种无监督式的海明距离搜索方法。首先,为了提高散列表处理效率,通过放宽尺寸约束得到具有可变长度散列键的多个散列表;然后,基于多散列表检索概率的理论分析,通过一个搜索算法,得到适合的散列键长度(HKL)的集合,并由此得到检索机制。实验采用了SIFT、ANN和Flickr三个大规模数据集,利用最小召回率和检索阈值验证方法的性能。实验结果表明:所提方法可高效处理大规模数据集,实现了1~3个数量级的加速度。To improve the retrieval efficiency of large-scale data,an unsupervised Hamming distance search method is proposed.Firstly,by relaxing size constraints,multiple hashing table of variable hash key length can be obtained to improve the efficiency of hash table processing.Then,based on theory probability analysis of multiple scattered list retrieval,an appropriate set of hashing key length(HKL)is calculated through a search algorithm.Three large-scale data sets,SIFT,ANN and Flickr,are used in the experiment,and the performance of the method is verified by using the minimum recall rate and the retrieval threshold.The experimental results show that the proposed method can efficiently handle large scale data sets and achieve 1 to 3 orders of magnitude of acceleration.

关 键 词：数据检索 大规模数据集 多散列表 海明空间 散列键长度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]