Research on Face Recognition System Based on RLWE Homomorphic Encryption

Research on Face Recognition System Based on RLWE Homomorphic Encryption

연구 분야: Cryptography

학회: International Conference on Security and Privacy in New Computing Environments

With the development of artificial intelligence face recognition technology is being widely used in various fields. However, the traditional face recognition system has serious security risks, face data is easy to leak, and privacy protection mechanism is insufficient. In order to solve this problem, a secure face recognition system based on RLWE homomorphic encryption algorithm in lattice cryptography is constructed. This method can directly process the face image in the encrypted domain, and realize the recognition function while protecting the security of biometric data. Experiments on CASIA-WebFace data set show that RLWE encryption algorithm is successfully embedded in the face recognition model based on ResNet-34, and the overall recognition accuracy of homomorphic encryption system reaches 97.4%, which is only 2% lower than that of plain text domain recognition. This proves the validity and application prospect of homomorphic encryption technology based on lattice cryptography in protecting the security of biometric information. However, the operation efficiency still needs to be improved. This study provides a valuable exploration path for homomorphic encryption technology to be applied to security biometric identification.