基于伪逆的局部保留迭代哈希  

Pseudo-inverse Locality Preserving Iterative Hashing

作　　者：杜仲舒 王永利[1] 赵亮[1] DU Zhongshu;WANG Yongli;ZHAO Liang(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2018年第8期1585-1588,1642,共5页Computer & Digital Engineering

基　　金：国家自然科学基金(编号:61170035);江苏省"六大人才高峰"项目(编号:2014-WLW-004);中央高校基本科研基金(编号:30920130112006);江苏省科技成果转化专项资金项目(编号:BA2013047)资助

摘　　要：高维数据的近似最近邻检索是许多应用研究的一个基础问题,同时大数据带来的维度灾难对传统的哈希算法产生了挑战。为此论文提出了基于伪逆的局部保留迭代哈希算法(pseudo-inverse locality preserving iterative hashing,PLIH),不仅有效保持数据的近邻关系,同时解决了哈希过程中的矩阵奇异和量化损失较大的问题。在该算法中,构建邻接图并最小化近邻在低维空间的距离,保持投影后矩阵的高维近邻关系;采用伪逆替代逆矩阵解决了矩阵奇异的情况下求解投影矩阵失效的问题。最后通过迭代量化使得投影矩阵在量化过程中的损失降至最小。论文通过与其他哈希算法在公开数据集上的比较,发现正确率和召回率都有5%到10%的提升,证明了该算法的可行性。The approximate nearest neighbor search for high-dimension data is a fundamental problem in many application domains. Dimension disaster brought by Big Data also challenges traditional hashing algorithm. In this paper,pseudo-inverse locality preserving iterative hashing（PLIH）is designed to maintain the high-dimensional neighborhood relation,solving the problem of matrix matrices and minimizing the loss in the quantization process.In this algorithm,the adjacent graph is constructed and the high-dimensional neighborhood relation of the matrix is preserved. The pseudo-inverse of the matrix is used to solve the problem of matrix matrices. Finally,the iterative quantization minimizes the loss of the projection matrix during the quantization process. In this paper,precision and recall have an increase of 5% to 10% compared with other hashing algorithms on public data sets,which proves the feasibility of the algorithm.

关 键 词：近似最近邻 伪逆局部保留投影 局部敏感哈希 迭代量化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]