Secure Outsourcing of Laplacian-Eigenmap-Based Face Recognition in IoT

Face recognition is widely used in Internet of Things (IoT) applications such as time and attendance systems, security systems, etc. Due to the limited storage and computing power of IoT devices, feature dimensionality reduction algorithms with high computational complexity in face recognition are usually outsourced to powerful clouds. However, less research has been conducted on outsourcing the computation of nonlinear dimensionality reduction algorithms that are more effective in recognizing high-dimensional data. For the first time, this paper proposes a Laplacian eigenmaps (LE)-based outsourcing computation scheme for face recognition. As a nonlinear dimensionality reduction method based on spectral graph theory, LE is commonly used in face recognition and can be solved by the approximation of generalized eigen-decomposition (GED). In this paper, a GED outsourcing algorithm is designed to prevent the leakage of the client’s input and output privacy to the cloud by using matrix expansion and double random perturbation techniques. The proposed outsourcing scheme reduces the computational complexity of the client, and the client can detect malicious behaviors from the cloud with non-negligible probability. Besides, the proposed outsourcing scheme has the same recognition accuracy as the original LE-based face recognition algorithm. Additionally, the correctness, privacy, verifiability, and efficiency of the proposed outsourcing algorithm are theoretically analyzed and experimentally verified, and the results indicate that without losing recognition accuracy, the proposed outsourcing algorithm achieves better security and performance compared to current dimensionality reduction outsourcing algorithms.