DP-Face: Privacy-Preserving Face Recognition Using Siamese Network

With the rapid development of deep learning, face recognition technology based on deep learning has been widely developed in recent years. However, during the training of the deep learning model, there is a risk of privacy leakage. If an attacker obtains private data, such as tags of the training data, the face images may be restored, and private information is leaked. To pro-tect the private information of the face recognition model, we introduce differential privacy technology to propose a privacy - preserving face recognition scheme using the Siamese Network framework called DP-Face. Unlike other privacy-preserving face recognition methods, we can adjust the balance between privacy and utility through privacy budgets according to actual needs. Experimental results show that the effectiveness and privacy of the proposed DP-Face can be well guaranteed.