A Secure and Efficient Face-Recognition Scheme Based on Deep Neural Network and Homomorphic Encryption

Abstract

In recent years, with the maturity of face recognition technology, face recognition has been widely used in real life, raising concerns about the accuracy of face recognition results, the efficiency of face recognition and the safety of data. So we proposed a secure and efficient face-recognition scheme based on deep neural network and homomorphic encryption. The entire scheme is divided into two parts: the client and the server. The client obtains the face images. The server performs recognition. Face features are extracted using deep neural networks and then encrypted with the Paillier algorithm. The data of face features is transferred from the client to the server with encrypted mode and does not need to be decrypted in the entire recognition process. In the recognition process, we adopt a highly efficient secretive Hamming distance calculation method and introduce a parallel computing scheme to encrypt feature data and calculate the ciphertext Hamming distance, which greatly improves the recognition efficiency of the entire program. No messages are leaked between the client and the server on the entire scheme, which achieves the purpose of protecting privacy and security. Compared with the previous secure face recognition scheme, the experimental results show that we improve the accuracy of and the efficiency of recognition while ensuring security.