Accelerating Homomorphic Encryption-based Facial Recognition Systems through Clustering
Sei Nakanishi, Yoshiaki Narusue, Hiroyuki Morikawa · 2024
Homomorphic encryption is gaining attention for enhancing privacy and data security in facial recognition systems. However, utilizing homomorphic encryption poses a significant challenge in that the computational complexity reduces the processing speed. In this paper, we propose a novel approach to leverage the K-means algorithm to pre-cluster facial feature data. In this approach, clustering is performed on the cloud server, and authentication information in clusters is compared in the order of clusters that are most likely to contain the authentication target. This speeds up the recognition process and significantly increases the overall processing speed. By splitting a database of 500 registered individuals into 6 clusters, our system completed the recognition in an average of 2.64 sec., achieving a 365% improvement in speed. This result facilitates faster facial recognition while preserving data security, thereby enhancing the practicality of homomorphic encryption-based facial recognition systems.