Real-time Face Structure Reproduction Model for Generative Adversarial Networks

In this paper, the image to image translation algorithm based on generative countermeasure network is used to capture and extract the face features in the source image and target image, obtain the mapping function after training the target image model, and transfer the features of the original face image to the face model of the target image. The conditional generative adversarial networks is selected as the solution to the face to face mapping problem. At the same time, a novel method for real-time face reconstruction of single target video sequence is proposed. When studying the selected loss function, we find that the system has wide applicability and strong ease of use, and there is no need to adjust the parameters. The results show that in addition to face capture and feature migration, in many tasks that need to process highly structured images, the pix2pix algorithm based on cGAN is also a highly expected method. This means that this system can not only be more perfect in face feature migration, but also have the possibility to expand the function in other fields of image processing.