基于稀疏自编码的无监督哈希算法  

Unsupervised Hashing algorithm based on sparse autoencoder

作　　者：张丽萍[1] 孟卫平[1] 谭家海[2] ZHANG Li-ping;MENG Wei-ping;TAN Jia-hai(Shaanxi International Business College,Xi’an 712046,China;Xi′an Institute of Optics and Precision Mechanics,Chinese Academy of Sciences, Xi’an 710119,China)

机构地区：[1]陕西国际商贸学院,陕西西安712046 [2]中国科学院西安光学精密机械研究所,陕西西安710119

出　　处：《液晶与显示》2018年第11期950-957,共8页Chinese Journal of Liquid Crystals and Displays

基　　金：陕西省科技厅农业科技公关项目(2015NY061)~~

摘　　要：最近邻搜索在大规模图像检索中变得越来越重要。在最近邻搜索中,许多哈希方法因为快速查询和低内存被提出。然而,现有方法在哈希函数构造过程中对数据稀疏结构研究的不足,本文提出了一种无监督的稀疏自编码的图像哈希方法。基于稀疏自编码的图像哈希方法将稀疏构造过程引入哈希函数的学习过程中,即通过利用稀疏自编码器的KL距离对哈希码进行稀疏约束以增强局部保持映射过程中的判别性,同时利用L2范数来哈希编码的量化误差。实验中用两个公共图像检索数据集CIFAR-10和YouTube Faces验证了本文算法相比其他无监督哈希算法的优越性。Nearest neighbor search is becoming more and more important in large scale image retrieval.Many hash methods are proposed in nearest neighbor search owing to fast query and low memory.However,there is a lack of research on the sparse structure of data in the process of hash function construction.The proposed hashing method introduces the sparse construction process into the learning process of hash function and uses KL distance of sparse autoencoder on hash code sparse constraints to enhance locality preserving discriminant mapping process.The proposed method leverages L2 norm to control quantization error in hash encoding.The experimental results on two common image retrieval datasets CIFAR-10 and YouTube Faces show that the proposed algorithm is superior to other unsupervised hashing algorithms.

关 键 词：图像哈希 稀疏自编码 KL距离 量化误差 无监督算法 

分 类 号：O431.2[机械工程—光学工程] O431.1[理学—光学]