FaceGroup: Continual Face Authentication via Partially Homomorphic Encryption & Group Testing

The Zero Trust (ZT) paradigm has recently emerged. The core idea of ZT is never to trust but always authenticate. By incorporating ZT into network architectures, neither users nor service providers need to trust each other, which significantly enhances the security level of these architectures. Cloud-based facial authentication is one of the plausible access control solutions to bring ZT into network architectures. Many of the state-of-art works on encrypted cloud-based authentication use Fully Homomorphic Encryption (FHE) to protect users’ private data. FHE enables a cloud server to perform arithmetic operations on encrypted inputs without decrypting them but with a significant computing overhead. In this work, we introduce novel approaches to incorporate Partially Homomorphic Encryption (PHE) into cloud-based facial authentication by changing the distance metric from Euclidean distance to Manhattan distance. As a result, we reduce the computational overhead by a factor ranging from 20 to 55. In addition, we propose a novel two-stage architecture for group facial recognition, which can further reduce the total computation cost of authentications required to identify an individual from a crowd. Compared with conventional facial recognition methods, to find people of interest, group facial verification can cut the cost of calculating facial recognition by 55%. With such a lightweight design, FaceGroup is scalable and can be deployed on resource-constrained devices.