RF04-05 Facial identification system based on deep embeddings and storage in NoSQL databases Details

Start Date: 2026-03-10

End Date: 2026-03-10

Recent advances in deep learning have significantly improved facial recognition systems through the use of deep embeddings, which represent facial features as high-dimensional numerical vectors. These embeddings enable facial identification by comparing vector similarities between different images. However, the storage and efficient management of large volumes of high-dimensional embeddings present important challenges, particularly when using traditional relational databases designed primarily for structured data. This project proposes the design and implementation of a prototype facial identification system based on deep embeddings generated from facial images using the DeepFace library in Python. The embeddings will be stored and managed using a NoSQL database, specifically MongoDB, which offers a flexible document-oriented model suitable for handling semi-structured and high-dimensional data. The system will include mechanisms for comparing embeddings using similarity metrics in order to identify individuals. Experimental tests will be conducted to analyze embedding distance behavior and evaluate the accuracy of the identification process.