Privacy-Preserving and verifiable [formula omitted]-based face recognition with cloud/edge server assistance

Abstract

In the current cloud computing era, outsourcing overloaded computations to cloud servers has become an increasingly popular computing paradigm. Meanwhile, face recognition (FR), as a typical and extensively deployed biometric authentication technique in the real world, always involves time-consuming large-scale matrix operations or complex optimization problems. Therefore, it is a naturally actual demand to study the cloud/edge server-assisted FR algorithm. Nevertheless, the sensitivity of the FR data and the incredibility of the cloud/edge server make this intriguing computing paradigm suffer from many security challenges. In this paper, we focus on the popular SRC-based FR and, for the first time, present a high-efficiency and secure outsourcing design for this algorithm. The key technique ingredient that emerged in our design is a new norm-preserving matrix transformation which is utilized to outsource the heavy ℓ1-minimization problem in SRC-based FR. This novel technique makes our design well protect the critical privacy information in SRC-based FR and discern the dishonest server with an optimal probability of 1. Simultaneously, compared with existing linear programming outsourcing algorithms, our design is tailored for SRC-based FR and enables the client to gain considerable computational savings without significantly increasing the cloud/edge server’s computational load. Also, we corroborate our theoretical claims by conducting extensive experiments on the publicly available database.