Multi-Task Consistency-Preserving Adversarial Hashing for Cross-Modal Retrieval.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Cross-modal hashing methods are gaining attention due to low storage costs and high query efficiency.
• Existing methods struggle to bridge the modality gap, limiting semantic consistency exploration and search performance.

Purpose of the Study:

• To propose a novel deep hashing method, Multi-Task Consistency-Preserving Adversarial Hashing (CPAH), for efficient cross-modal retrieval.
• To fully explore semantic consistency and correlation between different modalities.

Main Methods:

• Designed a consistency refined module (CR) to separate modality-common and modality-private representations.
• Presented a multi-task adversarial learning module (MA) to align modality-common representations in feature distribution and semantic consistency.
• Generated compact hash codes from modality-common representations.

Main Results:

• The proposed CPAH method demonstrates superior performance compared to state-of-the-art methods.
• Evaluations on three benchmark datasets confirm the effectiveness of CPAH in cross-modal retrieval.

Conclusions:

• CPAH effectively addresses the limitations of existing cross-modal hashing techniques.
• The method enhances cross-modal retrieval by leveraging semantic consistency and reducing the modality gap.