J4 ›› 2016, Vol. 43 ›› Issue (1): 94-98.doi: 10.3969/j.issn.1001-2400.2016.01.017

• 研究论文 • 上一篇    下一篇

一种相似性保持的线性嵌入哈希方法

王秀美;丁利杰;高新波   

  1. (西安电子科技大学 电子工程学院,陕西 西安  710071)
  • 收稿日期:2014-09-26 出版日期:2016-02-20 发布日期:2016-04-06
  • 通讯作者: 王秀美
  • 作者简介:王秀美(1978-),女,副教授,博士,E-mail:wangxm@xidian.edu.cn.
  • 基金资助:

    国家杰出青年科学基金资助项目(61125204); 国家自然科学基金重点资助项目(61432014);国家自然科学基金资助项目(6147230);国家自然科学基金青年基金资助项目(61100158)

Linear embedding Hashing method in preserving similarity

WANG Xiumei;DING Lijie;GAO Xinbo   

  1. (School of Electronic Engineering, Xidian Univ., Xi'an  710071, China)
  • Received:2014-09-26 Online:2016-02-20 Published:2016-04-06
  • Contact: WANG Xiumei

摘要:

在图像检索技术中,针对高维特性海量的图像数据检索速度慢、数据存储容量大及图像和其哈希编码之间相关性差的缺点,将相关性预测函数引入到哈希算法中,提出了一种相似性保持的线性嵌入哈希方法.该方法利用相关性预测函数保持高维数据与其编码之间的邻近关系,使边界损失代价最小化,构建线性哈希映射矩阵,获得紧致的哈希编码,提高了图像与编码间的相关性,实现了高精度的图像检索.通过与现存经典的哈希算法相对比,实验结果验证了线性嵌入哈希方法在查全率和查准率上的有效性.

关键词: 相似最近邻搜索, 哈希, 相关性预测函数, 查准率, 查全率

Abstract:

In order to implement quick and effective search, save the storage space and improve the poor performance of affinity relationshaps between high dimensional data and its codes in image retrieval, a new linear embedding hashing is proposed by introducing the preserving similarity. First, the whole data set is clustered into several classes, and then the similarity predicted function is used to maintain affinity relationships between high dimensional data and its codes so as to establish the objective function. By minimizing the margin loss function, the optimal embedded matrix can be obtained. Compared with the existing classic hashing algorithm, experimental results show that the performance of the linear embedding hash algorithm is superior to the other binary encoding strategy on precision and recall.

Key words: approximate nearest neighbor search, hashing, similarity predicted function, precision, recall

中图分类号: 

  • TP391