Privacy-enhanced facial recognition for IoT based on homomorphic encryption

The combination of face recognition technology and IoT system shows great potential. However, how to ensure the security of personal private information has become an urgent problem. In this paper, a system combining face recognition based on convolutional neural network with homomorphic encryption and edge processing is proposed. Firstly, MTCNN is used to correct the error of photos, then FaceNet model is used to extract the biometric information of users, and finally homomorphic encryption is used to ensure that the face feature information is always in ciphertext state during the whole data transmission process, thus effectively reducing the probability of user information leakage. In scenarios involving multiple faces and multiple targets, the project is upgraded using the FaceNet library and edge computing technology to adapt to multi-face and multi-target detection tasks. The experimental results show that the FaceNet model achieves a detection accuracy of 98.06% on the LFW dataset. Therefore, the system can meet the practical requirements and improve the robustness and accuracy of face recognition to a certain extent.