Home // DIGITAL 2025, Advances on Societal Digital Transformation // View article

On the Protection of Face Recognition Embeddings

Authors:
Gábor György Gulyás

Keywords: Face Recognition; Biometric Templates; Privacy; Security

Abstract:
Face recognition technologies rely on face embeddings, numerical representations of biometric features, which are increasingly used in security and commercial applications. However, the privacy risks associated with storing and processing these embeddings are significant, particularly under strict regulations, such as the GDPR and the AI Act. This paper investigates two main techniques for protecting face embeddings: Locality-Sensitive Hashing (LSH) and Homomorphic Encryption (HE). Through a case study using random projections and the Labeled Faces in the Wild dataset, we show that while LSH allows great reduction in data sizes and offers efficient approximate matching, it provides weak resistance to re-identification attacks. In contrast, HE enables computation directly on encrypted data and offers a more secure, though computationally expensive, alternative. We evaluate recent HE-based approaches and propose optimizations.

Pages: 72 to 75

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-285-2

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025