Quadruplet-Based Deep Cross-Modal Hashing

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep cross-modal hashing retrieval combines hashing efficiency with deep neural networks for feature extraction.
• Existing methods often use pairwise or triplet loss, potentially limiting exploration of cross-modal semantic similarities.

Purpose of the Study:

• To propose a novel deep cross-modal hashing method that better preserves semantic similarities across modalities.
• To introduce and evaluate a quadruplet loss function for improved cross-modal retrieval.

Main Methods:

• Developed a quadruplet-based deep cross-modal hashing (QDCMH) method.
• Utilized deep neural networks for feature extraction and a novel quadruplet loss for hash mapping.
• Conducted experiments on two benchmark cross-modal retrieval datasets.

Main Results:

• The proposed QDCMH method achieved state-of-the-art performance on benchmark datasets.
• Demonstrated the effectiveness of the quadruplet loss in capturing complex semantic relationships across modalities.
• Showcased improved accuracy and efficiency in cross-modal retrieval tasks.

Conclusions:

• The quadruplet loss is a more effective approach for preserving semantic similarities in deep cross-modal hashing.
• QDCMH offers a significant advancement in cross-modal retrieval accuracy and efficiency.
• The proposed method provides a strong baseline for future research in deep cross-modal hashing.