Research On Face De-identification GAN-Based Model

With the rapid development of face identification technology,the abuse of face identification technology in various scenes has aroused people's concern about the privacy leakage of their own identities and other issues,and researches on face identification have emerged.The purpose of face identification is to save information irrelevant to human identity,such as some important biological features,and eliminate information related to human identity privacy as much as possible.In this paper,a face deidentification model based on GAN is proposed,which USES VGG network to improve CycleGAN for face deidentification.As the initial stage of model training and the training of semantic loss of reconstructed image.VGG is used to retain the original semantic content information of input data to the maximum extent.In order to make the model produce higher quality images,a new fuzzy confrontation loss is added to the training model to generate clearer images of contour and edge.At the same time,the scaling convolution is used to replace the original deconvolution operation in the up-sampling process of the model,which solves the checkerboard effect in the image generation problem.Finally,the model combines style transformation and counterattack to achieve the purpose of face identification.In this paper,the effectiveness and stability of the model are verified by comparing the improved independent tests with some existing models.