.png){kind=icon}
.png){kind=icon}
Return
Deep adversarial multi-label cross-modal hashing algorithm
.png){kind=icon}
PRE
AI
DOI：10.1007/s13735-023-00288-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In recent years, more and more researchers employ the hashing algorithm to improve the large-scale cross-modal retrieval efficiency by mapping the floating-point feature into the compact binary code. However, the cross-modal hashing algorithm usually computes the similarity relationship based on single class labels, while ignoring the multi-label information. To solve the above problem, we propose the deep adversarial multi-label cross-modal hashing algorithm (DAMCH) which takes both multi-label and deep feature into consideration during establishing the cross-modal neighbor matrix. Firstly, we propose the inter- and intra-modal neighbor relationship preserving function to make the Hamming neighbor relationship be consistent with the original neighbor relationship. Secondly, we design linear classification functions to learn binary features' semantic labels and establish the hash semantic preserving loss function to guarantee the binary features have the same semantic information as the original label. Furthermore, we establish the intra-modal adversarial loss function to minimize the information loss during mapping the floating-point feature into the compact binary code, and propose the inter-modal adversarial loss function to ensure different modal features own the same distribution. Finally, we conduct the cross-modal retrieval comparative experiments and the ablation studies on two public datasets MIRFickr and NUS-WIDE. The experimental results show that DAMCH outperforms the current state-of-the-art methods.
Keywords:
Cross-modal retrieval
Image-text retrieval
Cross-modal similarity preserving
Hashing algorithm
Journal
IF:
2.9
Papers:
270
Citations:
866