Paper 2021/1559

Facial Template Protection via Lattice-based Fuzzy Extractors

Kaiyi Zhang, Hongrui Cui, and Yu Yu

Abstract

With the growing adoption of facial recognition worldwide as a popular authentication method, there is increasing concern about the invasion of personal privacy due to the lifetime irrevocability of facial features. In principle, {\it Fuzzy Extractors} enable biometric-based authentication while preserving the privacy of biometric templates. Nevertheless, to our best knowledge, most existing fuzzy extractors handle binary vectors with Hamming distance, and no explicit construction is known for facial recognition applications where ${\ell }_2$-distance of real vectors is considered. In this paper, we utilize the dense packing feature of certain lattices (e.g., $$ and Leech) to design a family of {\it lattice-based} fuzzy extractors that docks well with existing neural network-based biometric identification schemes. We instantiate and implement the generic construction and conduct experiments on publicly available datasets. Our result confirms the feasibility of facial template protection via fuzzy extractors.