Fast discrete cross-modal hashing with semantic consistency.

Area of Science:

• Computer Science
• Machine Learning
• Information Retrieval

Background:

• Supervised cross-modal hashing is crucial for large-scale retrieval but often suffers from class information loss and high computational costs.
• Existing methods frequently use pairwise similarity matrices, leading to complexity and potential quantization errors.

Purpose of the Study:

• To develop a Fast Discrete Cross-modal Hashing (FDCH) method that addresses limitations of existing approaches.
• To improve semantic consistency and retrieval accuracy in cross-modal hashing tasks.

Main Methods:

• FDCH utilizes both class labels and pairwise similarity matrices to learn a shared Hamming space.
• An asymmetric hash code learning model is proposed to avoid complex symmetric matrix factorization.
• An efficient discrete optimization scheme directly generates discrete hash codes, reducing complexity from O(n^2) to O(n).

Main Results:

• FDCH effectively preserves semantic consistency by incorporating class labels.
• The method significantly reduces computational complexity and memory costs.
• Experiments demonstrate FDCH's superiority over existing cross-modal hashing methods on real-world datasets.

Conclusions:

• FDCH offers an effective and efficient solution for supervised cross-modal hashing.
• The proposed method enhances retrieval performance and reduces computational burden.
• FDCH provides a robust approach for large-scale cross-modal retrieval tasks.