Research On Facial Expression Translation Based On Conditional Generative Adversarial Networks

With the rapid development of artificial intelligence,natural human-computer interaction has received extensive attention.Facial expression is the main form of emotional behavior in people's daily communication.In recent years,due to the extensive application of deep learning in the field of computer vision,face image generation has achieved good results,but it is still difficult to achieve natural human-computer interaction.It is the emotion that plays an important role in natural human-computer interaction,and the machine cannot express the emotional state of communication by simulating facial expressions.Therefore,letting computers have the ability to simulate facial expressions has important research significance for the development of natural human-computer interaction.This research has attracted the attention of more and more researchers at home and abroad,and has become a hot topic in the field of computer vision.Facial expression translation refers to transplanting an expression on an expression face image to a neutral expression face image based on the construction of the translation model.In the study of facial expression translation,facial expressions are generally divided into seven categories.The main task of facial expression translation is to use a single network to simultaneously translate and generate seven facial expressions while ensuring the same identity characteristics.The existing facial expression translation model has been achieved.Although there have been some achievements in the existing facial expression translation model,there are still some problems in that it is necessary to adopt a paired expression data set or only to convert between two expression fields,it is difficult to maintain the consistency of facial identity features,and it is difficult to simultaneously implement different types of facial expression translation in a single model.In order to solve these problems,this paper designs and constructs a facial expression translation model based on conditional generative adversarial network,and studies the problems of solving paired datasets,maintaining facial identity features and simplifying network models,and puts forward the following research results.(1)A new facial expression translation model based on conditional generation confrontation network is constructed by assigning expression domain label information to the generation network,learning the mapping of seven expression domain images,and finally generating seven types of facial expressions.(2)The reconstruction loss function is introduced into the generative network,and the difference between the reconstructed expression image and the original expression image is calculated by calculation to ensure the consistency between the generator reconstructed expression image and the original expression image identity feature.(3)Adding the domain classification loss function of the pseudo image in the generative network,the generative network tries to minimize the loss,ensuring that the generated fake expression image can be correctly classified into the target expression domain;adding the domain classification loss function of the real image in the adversarial network,by minimizing this loss,the network correctly classifies the real image to the corresponding original expression domain;achieving simultaneous generation of seven facial expressions in a single generative network.Through the experiments on CelebA dataset.CK+ dataset and FERG-DB dataset,the facial expression translation model based on conditional generative adversarial network is designed to be robust and effective for seven facial expression translation.