Research on Face Privacy Protection Scheme Based on Improved StyleGAN and Conditional GAN

With the rapid development of deep learning technology, face generation technology has been widely used in many fields. However, this also brings the problem of face privacy protection. Aiming at the security problem of face photo information uploaded to cloud server, this paper proposes a face privacy protection scheme based on improved StyleGAN and conditional generation Adversarial network (cGAN). By improving StyleGAN network structure and introducing self-attention mechanism, realistic simulated faces are generated, and the mixed recognition test of real and false face images is carried out by combining conditional GAN. The experimental results show that the generated artificial face can deceive the mainstream face detection system, and the discriminator obtained by training GAN can accurately distinguish the real face from the simulated face. The scheme can not only effectively deceive the attacker, but also significantly improve the level of face privacy protection, providing security for related applications.