Bit selection hashing for large scale image retrieval
TIAN Xing
LU Xiao-yi
WING W Y NG
HUANG Jia-jian
Abstract：Hashing methods have been widely used for solving large scale image retrieval problems. Among existing hashing methods, unsupervised hashing methods are most widely used because they do not require semantic information of images in the database. The shift-invariant kernelized locality-sensitive hashing (SKLSH) is a representative unsupervised hashing method which generates hash functions randomly without considering the performance of each projection. There-fore, weak hash functions yielding low retrieval performances may be generated by the SKLSH. In this work, we propose the bit selection hashing (BSH) which selects hash bits for the SKLSH based on the performance of hash bit projections in three aspects:similarity fitness, information capacity, and code independence. Then, a greedy selection method is applied to find the optimal combination of hash bits for the BSH. Two real world image databases are used to compare the perfor-mance of the proposed BSH with other representative hashing methods. Experimental results show that the BSH yields a significant improvement in comparison to the original SKLSH and other hashing methods.
Keywords：unsupervised hashing methodbit selectionlarge scale image retrieval
Publication Date：2017-01-01
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 769-775 )