基于SEAL库的密文人脸识别系统

Face Ciphertext Recognition System Based on SEAL

随着机器学习与深度学习技术的发展,人脸识别技术在过去几年中取得了巨大的进步。当广泛应用这项技术时,必须考虑人脸识别的安全性。为了保证用户的隐私,人脸特征向量不能以明文的形式进行传输。针对此问题我们提出采用全同态加密设计一个基于全同态加密的密文人脸识别系统,实现部分借助了谷歌的Face Net人脸识别库和微软的SEAL同态加密库。整个系统可在不对人脸特征模板解密的情况下完成人脸识别操作,且服务器中数据库保存的是密文人脸特征向量,无需担心人脸特征向量泄露的危险性。我们还对系统进行了优化,使得密文特征向量比对效率大幅提升,基本已达到应用场景的需求。

With the development of machine learning and deep learning technology, face recognition technology has made great progress in the past few years.The security of face recognition must be considered when we apply this technology widely.To ensure user privacy, face feature vectors cannot be transmitted in plain text.Aiming at this problem, we propose to design and build a ciphertext face recognition system based on fully homomorphic encryption by adopting FV scheme. The implementation part relies on Microsoft’s SEAL(Simple Encrypted Arithmetic Library) Library and Google’s FaceNet Library.The whole system can complete the face recognition operation without decrypting the face feature template,and the face feature template in the ciphertext state is saved in the database, so there is no need to worry about the leakage of the face feature template.In addition, the system is optimized to greatly improve the efficiency of ciphertext feature vector ratio, which basically meets the requirements of application scenarios.