Compact Structure Hashing via Sparse and Similarity Preserving Embedding.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Fast approximate nearest neighbor (ANN) search is crucial for large-scale databases.
• Hashing-based ANN techniques offer low memory cost and good computational complexity.
• Existing methods often neglect the sparse reconstructive relationship of data.

Purpose of the Study:

• To address the limitation of neglected sparse reconstructive relationships in hashing-based ANN methods.
• To propose a novel hashing method that learns compact and effective hash codes.
• To improve the retrieval performance in large databases.

Main Methods:

• Proposes Special Structure-based Hashing (SSBH).
• Preserves underlying geometric information among data.
• Exploits sparse reconstructive relationships within the data.

Main Results:

• SSBH learns compact and effective hash codes.
• Experimental results demonstrate SSBH's robustness and effectiveness.
• SSBH outperforms state-of-the-art hashing methods.

Conclusions:

• SSBH effectively addresses limitations in current hashing-based ANN techniques.
• The method enhances retrieval performance by considering data structure.
• SSBH offers a more robust and effective solution for ANN search.