Exploring Auxiliary Context: Discrete Semantic Transfer Hashing for Scalable Image Retrieval

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Unsupervised hashing is crucial for scalable content-based image retrieval due to its efficiency.
• Current methods suffer from limited semantic discrimination in learned hash codes.
• Image representation limitations hinder the effectiveness of traditional hashing techniques.

Purpose of the Study:

• To propose a novel hashing approach, Discrete Semantic Transfer Hashing (DSTH), for enhancing semantic representation in image retrieval.
• To augment discrete image hash codes by exploring auxiliary contextual modalities.
• To develop a unified framework that preserves visual similarities and facilitates semantic transfer.

Main Methods:

• Formulated a unified hashing framework for simultaneous visual similarity preservation and semantic transfer.
• Imposed discrete, bit-uncorrelation, and bit-balance constraints on hash codes for direct semantic transfer.
• Developed a novel discrete optimization method using augmented Lagrangian multipliers for iterative problem-solving.

Main Results:

• The proposed DSTH method effectively augments the semantics of discrete image hash codes.
• The learning process exhibits linear computation complexity and desirable scalability.
• Experiments on benchmark datasets show DSTH outperforms existing state-of-the-art approaches.

Conclusions:

• DSTH offers a significant advancement in unsupervised hashing for content-based image retrieval.
• The method successfully addresses the limited semantic discrimination issue in traditional hashing.
• DSTH provides a scalable and efficient solution for enhancing image retrieval systems.