The Study On Face Makeup Learning Using Style Transfer Technology

Makeup learning technology is also called makeup transfer technology,which is a special case of style transfer technology.The goal of makeup learning is to transfer the makeup of a reference image another original image.This article focuses on improving the quality of facial makeup learning.After fully studying the style transfer method and face makeup learning method using deep learning,this paper proposes a new double cycle constrained face makeup learning network DCCGAN,which can effectively separate the makeup features and content features of facial image to achieve the style transfer mission of makeup and improve the quality of generated image.This paper also proposes a face makeup learning network DS-DCCGAN based on the fusion of different depth and shallow makeup features.This network is improved on the structure of DCCGAN,which can learn makeup features of different depths from the target image to further improve makeup transfer quality.The work of this article is reflected in the following four aspects:1.After learning the basic concepts of deep learning and convolutional neural networks,several representative and latest style transfer or face makeup learning methods deep learning is introduced and implemented with experiments.Then this article point out the challenges faced by facial makeup learning and its important value in practical application.2.This article proposes a new double-cycle constrained face makeup learning network DCCGAN.This network effectively separates the image makeup style domain and content domain through the dual cycle constrained style transfer module,and exchanges the makeup and content domain from the reference image to the target image.A large number of experimental results show that the DCCGAN network not only has high makeup learning quality,but also has good makeup migration quality and robustness for face makeup images with shadows,posture changes,and face images outside the data set.3.This paper also propose a face makeup learning network DS-DCCGAN based on makeup features from different layers of networks.DS-DCCGAN uses a double cycle consistent constrained style transfer module to learn to obtain makeup features of different depths of the reference image,and merges makeup features of different depths to the target image.A large number of experimental results show that compared to DCCGAN network makeup learning,the DS-DCCGAN network is further improved.It has a low degree of dependence on shadows,posture changes,etc.,and also has good robustness.4.This paper designs an objective evaluation plane for makeup learning quality with structural similarity SSIM and Gram loss as the abscissa and ordinate respectively.The evaluation plane objectively reflects the advantages of the method in this paper compared with other method.