Translation Of Heterogeneous Face Image Based On Generative Adversarial Networks

The translation of heterogeneous face images refers to inputting a face image of one type of visual domain and then converting it into a face image of another type of visual domain.With the continuous development of computer vision and deep learning algorithms,the translation of heterogeneous face images has been used more and more in the fields of public security systems,intelligent Makeup,social entertainment,and image editing,etc.In view of the rapid development and advantages of deep learning algorithms in the field of computer vision,this paper proposes two translation methods based on generative adversarial networks in the application direction of two heterogeneous face image translation.The main work done in this paper is as follows:1.The research in verification of human face issue has impelled the demand and interest of law enforcement agencies and digital entertainment industry in transferring sketches to photo-realistic images.However,sketch-photo synthesis remains a significant challenging problem despite the rapid development of neural networks in image-to-image generation tasks.So far,existing approaches still have inextricable limitations due to the lack of paired data in the training stage and the fact that the striking differences between sketch and photo.In this work,we present a new framework for translating face sketches to photo-realistic images in an unsupervised fashion.Compared with current unsupervised image-to-image translation methods,our framework leverages an additional semantic consistency loss to keep the input semantic information in the output,and replaces the pixel-wise cycle-consistency with perceptual loss to generate sharper images for our task.We employ the generator architecture proposed by PGGAN and train it with a GAN loss for realistic output,a cycle consistency loss for driving the same input and output to remain constant.Extensive experiments on multiple baseline sketch-photo datasets demonstrate that our network achieves a significant improvement both in qualitatively and quantitatively.2.The animation of face images as a computer art form is widely used in daily life.However,drawing realistic scene images in anime style by hand is very time-consuming and laborious,and requires a lot of superb artistic skills.This paper proposes an unsupervised learning generative adversarial network framework with attention mechanism,and uses face analysis algorithms to reconstruct key local information.It also solves the problem that there is no publicly matched data set of the real face image and the anime-style face image.We train the semantic consistency loss and the original adversarial loss together to ensure the consistency of edge features and content information,and combined with the cyclic consistency assumption and the perceptual loss function,calculate the loss between the source domain image and the generated source domain image to ensure the consistency of the overall feature information of the input image.In addition,the use of a local cyclic consistency loss function in the attention adversarial network enables the generator to focus on the generation of facial detail features and generate more refined facial images.