Joint Specifics and Consistency Hash Learning for Large-Scale Cross-Modal Retrieval.

Area of Science:

• Computer Science
• Information Retrieval
• Machine Learning

Background:

• Multimedia data is rapidly increasing, making cross-modal similarity retrieval a significant challenge.
• Hashing is a key technique for efficient large-scale cross-modal data searching by mapping data to a low-dimensional Hamming space.
• Existing methods often fail to fully preserve and fuse discriminative modal-specific features and heterogeneous similarity.

Purpose of the Study:

• To propose a novel joint specifics and consistency hash learning method for cross-modal retrieval.
• To develop a framework that preserves both modal-specific features and cross-modal semantic consistency.
• To enable scalable and effective cross-modal similarity searching.

Main Methods:

• An asymmetric learning framework is introduced to exploit label information for discriminative hash code learning.
• The method learns specific information within individual modality subspaces and consistent information across multiple subspaces.
• An alternatively iterative optimization strategy is employed for scalable learning.

Main Results:

• The proposed method effectively preserves and fuses discriminative modal-specific features and heterogeneous similarity.
• The joint specifics and consistency hashing learning is scalable for large-scale cross-modal retrieval.
• Experiments on five benchmark databases demonstrate significant effectiveness and efficiency.

Conclusions:

• The proposed joint specifics and consistency hash learning method significantly improves cross-modal retrieval performance.
• The asymmetric learning framework and iterative optimization offer a scalable solution for multimedia data searching.
• This approach advances the field of cross-modal similarity retrieval by better integrating diverse data features.