Unsupervised cross-modal similarity via Latent Structure Discrete Hashing Factorization
作者：
Highlights：
• We propose a latent structure discrete hashing factorization framework to distill the shared discrete hash codes from the multiple sources inherent geometric similarity which is defined by the minimum and maximum singular values.
• An eigenvalue queuing strategy for the intrinsic geometry structure is designed to generate a discrete hash dictionary with the aid of the Hadamard matrix, thus reducing the quantitative loss.
• For binary optimization, we design a discrete iterative algorithm which does not depend on the hash encoding length, i.e., it does not need to be learned bit by bit.
摘要
•We propose a latent structure discrete hashing factorization framework to distill the shared discrete hash codes from the multiple sources inherent geometric similarity which is defined by the minimum and maximum singular values.•An eigenvalue queuing strategy for the intrinsic geometry structure is designed to generate a discrete hash dictionary with the aid of the Hadamard matrix, thus reducing the quantitative loss.•For binary optimization, we design a discrete iterative algorithm which does not depend on the hash encoding length, i.e., it does not need to be learned bit by bit.
论文关键词：Cross-modal hashing,Discrete hashing,Latent similar structure
论文评审过程：Received 26 November 2020, Revised 22 January 2021, Accepted 9 February 2021, Available online 10 February 2021, Version of Record 15 February 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.106857
