Supervised Matrix Factorization Hashing for Cross-Modal Retrieval.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Cross-modal hashing embeds heterogeneous multimedia data into a shared Hamming space for efficient retrieval.
• Matrix factorization methods have shown promise but struggle to incorporate label and local geometric information effectively.
• Big multimodal data necessitates advanced techniques for effective cross-modal search.

Purpose of the Study:

• To propose a novel cross-modal hashing method that leverages collective matrix factorization.
• To address the limitations of existing methods in utilizing label information and local geometric structure.
• To enhance the discriminative power of latent semantic features for improved cross-modal retrieval.

Main Methods:

• Developed a collective matrix factorization approach incorporating label consistency across modalities.
• Formulated local geometric consistency within each modality as a graph Laplacian term in the objective function.
• Employed an iterative strategy to solve the objective function and learn unified hash codes for different modalities.

Main Results:

• The proposed method significantly improves the discriminative power of latent semantic features.
• Experimental results demonstrate superior performance compared to state-of-the-art cross-modal hashing methods on benchmark datasets.
• Unified hash codes facilitate effective cross-modal search across different data types.

Conclusions:

• The proposed collective matrix factorization-based cross-modal hashing method is effective.
• Integrating label consistency and local geometric consistency enhances feature discriminability and retrieval performance.
• The method offers a robust solution for big multimodal data retrieval challenges.