Research On Face Aging Method Based On Generative Adversarial Gender Constraint Model

Thanks to the development of deep learning,many computer vision tasks have made great progress,such as cross-age face verification,face editing,face entertainment,etc.The application of these scenes is inseparable from a key technology face aging,and many image processing tasks mostly involve the modification of the age information of face images.In view of the commercial value and scientific research value of this technology,face aging has attracted more and more researchers' attention.Therefore,it is of great significance to carry out research on human face aging technology and explore its scientific and practical value.Face aging technology currently has two major research difficulties.On the one hand,the complex model of the face aging is difficult to train,and it is impossible to make a good simulation of the face aging process;on the other hand,the age information contained in the existing face data set is imperfect,and there is a lack of facial images belonging to the same individual across a wide range of ages.The following is the main research content of this article.(1)Aiming at the problem that the face aging model is very difficult to simulate and difficult to train,the current research tends to use generative models to convert the age information of the input image.At the same time,considering the problem of incomplete data sets that meet the conditions,using the idea of style transfer to achieve face aging is the current research focus.In this paper,based on Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks(CycleGAN),a gender-constrained face aging algorithm(Gender Classified Face Aging with CycleGAN,GFAGAN)is proposed.The algorithm performs constrained training on the image translation model by means of gender constraints,and the resulting face aging model has excellent performance in facing single-sex face aging tasks,and in non-sex-sensitive face aging tasks it also achieved better performance than traditional models.(2)Considering that there are different evaluation criteria for different tasks,this paper proposes an evaluation mechanism based on the participation of subjects,compared with the traditional subjective evaluation method,in studying the influence of the gender-constrained model on the aging process of the face.The evaluation method in this article is more objective and fairer.The experimental results show that the method proposed in this paper has outstanding performance on the face aging task.The model can synthesize more realistic and detailed aging faces,which fully illustrates the superiority of the algorithm proposed in this paper,and verifies excellent performance of gender-constrained model in face aging task.