Locality sensitive hashing with bit selection

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 52; no. 13; pp. 14724 - 14738
• Main Authors: Zhou, Wenhua, Liu, Huawen, Lou, Jungang, Chen, Xin
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2022; Springer Nature B.V
• Subjects: Artificial Intelligence; Binary codes; Computer Science; Criteria; Image retrieval; Machines; Manufacturing; Mechanical Engineering; Processes; Similarity; Special Issue on Multi-view Learning; Nearest neighbor search; Locality sensitive hashing; Hash bit; Bit selection; Binary code
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-022-03546-9

Locality sensitive hashing (LSH), one of the most popular hashing techniques, has attracted considerable attention for nearest neighbor search in the field of image retrieval. It can achieve promising performance only if the number of the generated hash bits is large enough. However, more hash bits assembled to the binary codes contain massive redundant information and require more time cost and storage spaces. To alleviate this limitation, we propose a novel bit selection framework to pick important bits out of the hash bits generated by hashing techniques. Within the bit selection framework, we further exploit eleven evaluation criteria to measure the importance and similarity of each bit generated by LSH, so that the bits with high importance and less similarity are selected to assemble new binary codes. To demonstrate the effectiveness of the proposed framework of bit selection, we evaluated the proposed framework with the evaluation criteria on five commonly used data sets. Experimental results show the proposed bit selection framework works effectively in different cases, and the performance of LSH has not been degraded significantly after redundant hash bits reduced by the evaluation criteria.