Structure-aware contrastive hashing for unsupervised cross-modal retrieval.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Cross-modal hashing is vital for large-scale similarity retrieval due to efficiency and low storage needs.
• Unsupervised cross-modal hashing faces challenges in accurately constructing sample similarity without manual annotation.
• Existing methods often yield suboptimal hash codes due to inaccurate similarity matrices derived from basic representations.

Purpose of the Study:

• To propose a novel unsupervised hashing model, Structure-aware Contrastive Hashing (SACH), for improved cross-modal retrieval.
• To address the limitations of existing methods in constructing accurate similarity relationships in unsupervised cross-modal hashing.
• To enhance the quality of hash codes and retrieval performance in cross-modal applications.

Main Methods:

• Employs both high-dimensional and discriminative representations to build an informative semantic correlative matrix across modalities.
• Introduces a multimodal structure-aware alignment network to minimize the heterogeneous gap in high-order semantic spaces.
• Reduces disparities within heterogeneous data sources and enhances semantic information consistency across modalities.

Main Results:

• Demonstrates superior performance of the SACH method in cross-modal retrieval tasks.
• Achieves better results compared to existing state-of-the-art methods on widely used datasets.
• Validates the effectiveness of the proposed approach in generating accurate similarity matrices and high-quality hash codes.

Conclusions:

• The proposed SACH method significantly advances unsupervised cross-modal retrieval.
• SACH effectively overcomes the limitations of previous unsupervised hashing techniques.
• The model enhances semantic consistency and reduces data disparities for improved retrieval accuracy.