Based On StarGAN Realization Style Of Face Image Motivation Research

Image,as the most intuitive information carrier,has always been the hot research object of many researchers.Image sharpening is a crucial and difficult problem in the field of image processing.In the past 20 years,obtaining clear images with high resolution has been the research direction of many researchers.Generating clear stylized images will further enhance the value of style images.StarGAN model is a kind of image transformation model based on generative adversarial network,which can realize the migration of face images in multiple domains,so as to produce a variety of face images.But the StarGAN model produces images with poor sharpness,in order to get a high resolution image.In this paper,by improving the network structure of StarGAN model,a sharpening model C-StarGAN based on StarGAN is designed,and the resolution of the generated face stylized image is further improved compared with that of the StarGAN model.The research content of this paper is summarized as follows:Firstly,the network structure and implementation principle of StarGAN model are studied,and the stylized images generated by StarGAN are analyzed to find that there are some problems such as lack of clarity,lack of denoising,and not obvious detail performance.In order to optimize the problems existing in StarGAN model,a clear model C-StarGAN based on Stargan is proposed.In this model,the U-Net network model is introduced to improve the fusion,and the jump connection in U-Net network structure can greatly improve the accuracy of image segmentation.Secondly,degradation removal module is introduced in the later stage to remove the blur,noise and other bad factors in the low resolution image.Add Facial Priors to provide variety and rich facial details;Then the loss is reconstructed to enhance the reality and fidelity of the stylized image generated by the model.After fusion,the C-StarGAN model can generate clear style images.Finally,a variety of technologies are integrated to effectively solve the problems such as lack of clarity and obvious details in StarGAN model,and generate more real and diversified facial style images.The model was trained and tested on CelebA-HQ data set,and tested on FFHQ data set.After subjective evaluation test and objective evaluation test,the experimental results show that the model can not only effectively improve the clarity and detail features of the generated images,but also ensure the diversity of the generated image styles,which verifies the effectiveness of the model.