RF04-06 Optimizing Facial Recognition with Vector Databases for Deep Embeddings Details

Start Date: 2026-03-10

End Date: 2026-03-10

Deep learning–based facial recognition generates high-dimensional vector representations, or deep embeddings, which encode unique facial features. Efficient storage and retrieval of these embeddings is critical for large-scale systems. This project proposes a prototype system that uses vector databases to store and query facial embeddings. The study also includes a comparison with NoSQL databases to evaluate their performance in storing and retrieving embeddings. Metrics for evaluation include retrieval time, precision, recall, and identity separability. The methodology involves embedding extraction from a facial dataset, storage in both database types, implementation of similarity queries, and performance analysis. The expected outcome is a comprehensive understanding of the advantages and limitations of vector versus NoSQL databases in large-scale facial recognition.