Latent Semantic Minimal Hashing for Image Retrieval

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Hashing-based similarity search is crucial for large-scale image retrieval, offering speed and efficiency.
• Existing unsupervised hashing methods often fail to fully refine image features or explore latent semantic embeddings concurrently.

Purpose of the Study:

• To propose a novel joint binary codes learning method for image retrieval.
• To enhance feature representation by integrating image features with latent semantic features.

Main Methods:

• Developed a latent semantic minimal hashing approach.
• Employed matrix decomposition to learn refined, discriminative latent semantic features.
• Incorporated a minimum encoding loss to ensure discriminative binary codes.

Main Results:

• The proposed method effectively refines original image features.
• Learned latent semantic features enhance the discriminative power of the data.
• Experimental results show superior performance compared to state-of-the-art hashing methods on large databases.

Conclusions:

• Latent semantic minimal hashing offers a powerful approach for query-by-example image retrieval.
• Simultaneously refining features and learning semantic embeddings leads to improved search performance.
• The method provides a more discriminative and efficient solution for large-scale image retrieval systems.