Research On Face Data Privacy Protection Algorithm Based On Conditional Generative Adversarial Nets

Nowadays,the world has stepped into the era of Internet of Everything and artificial intelligence.With the development and popularization of intelligent identification technology,people are confronted with potential threats to privacy security while obtaining convenience.With the wide coverage of video surveillance and social networks,face data has become the easiest biometric privacy information to obtain.Therefore,in the era of big data,the privacy of human face data deserves people's deep thinking.How to hide the face identity information while retaining the face attribute information has important research value and significance.In this paper,we learned the existing face data privacy protection methods.Inspired by supervised learning and Generative Adversarial Networks,we proposed a new face data privacy protection method.After experimental improvement,we realized the protection of face data privacy and the improvement of data availability.The main contributions of this paper are as follows:(1)This paper proposes a face data privacy protection algorithm based on Conditional Generative Adversarial Networks.The method can automatically anonymize the faces in the image while retaining the original data distribution,which is an end-to-end method.This model is based on the CGAN,the generated image considers the original face pose and image background,uses the boundary box annotation to identify the face privacy sensitive area,takes the key points of seven people's faces as the conditional information to constrain model training,uses the improved U-Net to enhance the quality of the generated image.The introduction of face keypoints information enables the model to generate highly realistic anonymous faces and seamlessly transition between the generated face and the existing background.Experiments on Celeb A and MTFL datasets,which are rich in face data and contain key points annotation,show that the proposed model can achieve better privacy protection of face data.(2)In order to further improve the data availability of the anonymous face and make the generated face retain the expression attribute information of the original face better,an improved algorithm of facial data privacy protection based on adaptive learning of facial expressions is proposed.In the face data privacy protection algorithm based on the above,with the introduction of Emotion facial expression recognition model,on the expression of the human face recognition,compared with original facial expression label,optimize the loss function,to improve the generator by the attention of the original facial expression,make its adaptive learning the relationship between facial expression and facial point.The improved model was trained and tested on Ra FD dataset with facial expression tags.The experimental results were compared with the previous model,and it was proved that the anonymous face generated by the improved model could retain facial expression attributes without affecting the performance of privacy protection,thus improving the data availability.