Facial Makeup Transfer With GAN For Different Aging Faces

Facial aging technology is widely used in scenes such as finding missing children and tracking criminals.If certain interferences such as facial makeup and facial occlusion are performed on the aging face,it is likely to affect the recognition of the tracking system.Therefore,the research on the makeup of different aging faces is extremely important.Existing research on the transfer of facial makeup has been able to achieve the transfer from the non-makeup domain to the makeup domain.However,few studies involve the transfer of facial makeup in different age groups.Moreover,the existing face datasets rarely contain age attribute tags and makeup attribute tags at the same time,which also makes facial makeup transfer on faces of different ages full of challenges.To solve the above problems,we design a scalable learning framework AB-Net,which can realize the transfer of facial makeup of different ages while protecting the identity of the characters.AB-Net is composed of two sub-network modules,Aging-Net and Makeup-Net.First,AB-Net learns the aging mechanism of facial features through Aging-Net,and feeds back the learned aging pattern to Makeup-Net,and then trains Makeup-Net to realize the mapping relationship between non-makeup domain and makeup domain.And transfer the makeup style to the nonmakeup face.In the training process of the entire network model,we use multiple losses to ensure that AB-Net preserve information about the identity,background,etc.Simultaneously conduct experiments on the CACD face dataset and the Morph face dataset to verify the effectiveness and robustness of AB-Net on facial makeup transfer tasks of different ages.Specifically,the main contributions of the article are as follows:(1)A novel network model framework AB-Net is proposed to solve the problem of facial makeup transfer of different age groups while maintaining the identity of the characters.(2)A new autoencoder Patch Auto Encoder is designed,which can learn the subtle aging characteristics of facial aging.The proposed loss function Texture loss is used to capture the texture features of the face,which in turn encourages the generator Patch Decoder to make the generated face more natural and real.(3)Use global domain-level loss and instance-level loss calculated by pixel-level histogram loss on individual local facial regions to solve the problem of facial makeup transfer.