基于GAN的人脸图像光照迁移  

Research of Face Relighting Method Based on GAN

作　　者：宁宁 金鑫[1] 张晓昆[1] 李艳楠 NING Ning;JIN Xin;ZHANG Xiaokun;LI Yannan(Beijing Electronic Science and Technology Institute,Beijing 100070,P.R.China)

机构地区：[1]北京电子科技学院,北京市100070

出　　处：《北京电子科技学院学报》2019年第4期33-41,共9页Journal of Beijing Electronic Science And Technology Institute

基　　金：国家自然科学基金面上项目:“无退化的混沌密码标准实现研究”(61772047);中央高校科研业务费项目:“量子计算与人工智能技术在密码分析中的应用研究”(328201903)

摘　　要：获得理想光照的光照迁移是游戏电影制作、图像后期处理等行业的热点问题。此外,它也是计算机视觉领域的研究热点。在以人脸图像为研究对象的多个领域内,光照将极大的影响研究结果,例如光照对比强烈的人脸图像或视频是目前人脸识别领域的难点。在生成对抗网络的启发下,本文设计并实现一种基于生成对抗网络的光照迁移方法,可迁移图像理想的光照效果,也可将图像恢复为均匀光照或者无光照的图像,即去光。迁移结果说明该方法能够迁移不同摄像对象间的光照效果,且迁移结果与真实图像接近。Light migration with ideal illumination is a hot issue in games,film production,and image post-processing.It is also a hotspot in the field of computer vision.In many fields where face images are the subject of research,illumination will affect the research results.For example,face images or videos with strong contrast of illumination are the difficulties in the field of face recognition.Inspired by the GAN,this paper designs and implements a lighting migration method based on GAN,which can migrate the ideal lighting effect of the image,and can also restore the image to uniform or unlit image.The migration result shows that the method can migrate the illumination effect between different camera objects,and the migration result is close to the real image.

关 键 词：人脸图像 光照迁移 卷积神经网络 生成对抗网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]