Related Experiment Video
Updated: Nov 27, 2025
13:51
Cross-Modal Multivariate Pattern Analysis
Published on: November 9, 2011
20.3K
Deep Semantic-Preserving Reconstruction Hashing for Unsupervised Cross-Modal Retrieval
Shuli Cheng1, Liejun Wang1,2, Anyu Du1
1College of Information Science and Engineering, Xinjiang University, Urumqi 830046, China.
Entropy (Basel, Switzerland)
|December 8, 2020
Summary
Deep hashing improves cross-modal retrieval but struggles with semantic reconstruction. Our novel Deep Semantic-Preserving Reconstruction Hashing (DSPRH) method enhances unsupervised retrieval by preserving semantic information.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Deep hashing is crucial for large-scale cross-modal retrieval due to its efficiency.
- Reconstructing semantic information in cross-modal retrieval remains a significant challenge.
Purpose of the Study:
- To address the challenge of unsupervised cross-modal retrieval semantic reconstruction.
- To propose a novel deep semantic-preserving reconstruction hashing (DSPRH) algorithm.
Main Methods:
- Introduced a spatial pooling network module using tensor regular-polymorphic decomposition for high-order context semantics.
- Employed global covariance pooling for channel semantic information and accelerated convergence.
- Utilized a dual bottleneck auto-encoding structure for visual-text modal interaction and a novel loss function for metric learning.
Main Results:
- The DSPRH algorithm demonstrated superior performance in retrieval tasks.
- Achieved better performance on the MIRFlickr-25K and NUS-WIDE datasets.
Conclusions:
- DSPRH effectively tackles unsupervised cross-modal retrieval semantic reconstruction.
- The proposed method enhances the preservation of semantic information for improved retrieval accuracy.
