Non-Interactive and secure outsourcing of PCA-Based face recognition

Abstract

In recent years, there have been more and more researches focus on the field of face recognition with the development of artificial intelligence (AI). Principal Component Analysis (PCA) is an important face recognition algorithm which has high accuracy without a large amount of data. Currently, the outsourcing of PCA-based face recognition protocol required three interactions between the clients and the cloud to execute matrix multiplications and eigenvalue decomposition, respectively, which needs very high communicational costs. In this paper, we propose a non-interactive PCA-based face recognition outsourcing protocol, which only needs one encryption and decryption without interactions between the clients and the cloud. That is to say, the client can obtain the final result of face recognition by encrypting the original images and decrypting the outsourcing results only once. The privacy of input and output is protected well by the proposed protocol, and the computational complexity is greatly reduced. In addition, the client can effectively detect the bad behaviors of the cloud and refuse the wrong outsourcing results by a verification algorithm. We prove the feasibility of our protocol from both theoretical and experimental analysis. The theoretical analysis shows that our proposed protocol reduces the computational overheads on the client’s side from O(n3) to O(n2). We simulate the proposed protocol and the experimental results show that when the matrix dimension exceeds 2500×3000, the client can gain more than 16.9825 overhead savings which indicates the efficiency of the proposed protocol.