Research And Implementation Of Secure Face Recognition Technology In Cloud Environment

With the advent of the mobile Internet era and the development of cloud computing technology,more and more face recognition calculations are completed in the cloud.In the process of face recognition calculation,the client's private data is completely exposed on the cloud server,which brings great Insecure.On the other hand,with the development and maturity of deep neural network technology in classification tasks in recent years,it has become the mainstream technology of face recognition.However,many scholars have further studied the deep neural network,and proposed a number of methods for generating adversarial examples that can attack deep neural network models,which achieved good effect on the mainstream deep neural network model in recent years.Therefore,face recognition technology that relies on deep neural network models is no longer absolutely safe.In summary,face recognition in cloud environment is insecure in terms of privacy leakage between participants and adversarial attack.How to solve the above two problems and realize secure face recognition in cloud environment Is the starting point of this study.The work of this thesis includes:1.Homomorphic encryption and oblivious transfer protocol are to design a secure face recognition scheme based on homomorphic encryption.This scheme can protect the privacy information of all participants in face recognition and realize secure face recognition on the face recognition scheme level.Moreover,the deep neural network facenet is used in the face feature data extraction part,which reduces the dimension of the extracted feature data,so as to reduce the calculation amount when combined with the safety calculation algorithm.The whole scheme also utilizes the efficient secret Hamming distance calculation formula and the parallel calculation method to further improve the calculation efficiency.At the same time,the introduction of facenet also improves the recognition accuracy and makes the whole application more valuable.2.Analyze and verify the defense capability of the deep neural network face recognition model used in the secure face recognition scheme based on homomorphic encryption for the mainstream adversarial examples attacks and improve the defense ability by fine-tuning.The secure face recognition model is realized which realize secure face recognition on the face recognition model level and prevent third party from using the adversarial examples attack the face recognition model.In the process of fine-tuning,the early termination method is adopted to prevent over-fitting of the model,and the secure face recognition model is generated.Compared with the original model,it has the advantages of higher recognition accuracy,better defense ability and stronger security.3.The secure face recognition model is applied to the secure face recognition scheme based on homomorphic encryption.The prototype of the secure face recognition system in cloud environment is designed and implemented in engineering.The experimental results show that compared with the general face recognition system,the system realizes secure face recognition at two levels of face recognition scheme and face recognition models,which prevents privacy data leakage between participants in the face recognition process and can also prevent third parties from using the confrontation sample to attack the face recognition model in the scheme.The entire system is more secure,with higher recognition accuracy and computational efficiency.Compared with the existing secure face recognition technology,this paper implements secure face recognition at two levels of face recognition scheme and face recognition model,which not only protects the privacy information between participants but also prevents third parties from using adversarial examples attack the face recognition model in the scheme and makes the recognition accuracy,calculation efficiency and application value higher.