Research And Implementation Of Frontal Face Image Synthesis Based On Deep Learning Method

In many scenes,the frontal face image is the only criterion for judging the identity of a person.However,it is difficult to collect a standard frontal image in an uncontrolled environment.Face frontalization technology is helpful to improve the identification rate.And it make up for the defects of current face recognition methods in multi-poses situation.Due to the non-rigid features of human face,the mapping from profiles to frontal face is nonlinear,which can be learned by deep learning.Autoencoder and generative adversarial network are two effective generation models.Autoencoder learns features by extracting image features and restoring them,but it can generate unclear images.GAN encourages the generator to learn image features by means of game,but the convergence speed is slow.Based on the two generating networks,a face frongtalization model is proposed.This model combines the advantages of autoencoder and GAN,and add pose information by deep learning method to improve the face frontalization effect under multi-poses situation.The main work of this paper includes two aspects:firstly,this paper proposes a Pose-weighted Generative Adversarial Network(PWGAN),which adds a pre-trained pose certification module to learn face pose information.For the single input image,PWGAN combines fusion features with pose features.And for multiple input images,PWGAN uses pose information and the leaping degree of feature map's pixel points to dynamic distribute weights when fusing feature maps.PWGAN makes full use of pose information to make the generation network learn more about facial features.Secondly,based on the PWGAN model,this paper designs and implements a multi-poses face frontalization system.The system mainly includes face frontalization module,model training module,face database module and log module,which can realize face frontalization,face similarity matching,multi-model training and face database storage.The experimental results show that the PWGAN model can effectively generate the images of the face with single or multiple side face images.In the case of multi-poses side face frontalization,the more realistic generation effect is achieved,which better retains the identity characteristics and improves the recognition efficiency of the side face.