基于三网络对抗学习的素描人脸合成方法  

Sketch face synthesis method based on three-network adversarial learning

作　　者：杜康宁 李凯旋 曹林[1,2] DU Kang-ning;LI Kai-xuan;CAO Lin(Key Laboratory of the Ministry of Education for Optoelectronic Measurement Technology and Instrument,Beijing Information Science and Technology University,Beijing 100101,China;School of Information and Communication Engineering,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学光电测试技术及仪器教育部重点实验室,北京100101 [2]北京信息科技大学信息与通信工程学院,北京100101

出　　处：《计算机工程与设计》2021年第6期1691-1698,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61671069);北京信息科技大学“勤信人才”培育计划基金项目(QXTCPA201902);北京市教委面上基金项目(KM202011232021)。

摘　　要：为解决现有素描人脸合成方法中素描人脸图像细节缺失、清晰度低及可适用性差的问题,提出一种三网络对抗学习的模型。由面部特征提取网络、生成网络及判别网络组成,引入面部细节损失与对抗损失相结合的复合损失函数,提高合成素描人脸图像的质量。在公共素描人脸数据集中与现有方法的定量与定性对比实验验证了该方法能够生成更加逼真、清晰的素描人脸图像。To solve the problems of lack of detail,low definition,and poor applicability of sketch face images in the existing sketch face synthesis methods,a three-network adversarial learning model was proposed.The model consisted of a facial feature extraction network,a generation network,and a discrimination network.A composite loss function was introduced,in which facial details loss and the generative adversarial loss were combined to improve the synthetic face image quality.Through the quantitative and qualitative comparison experiments with the existing methods on the public sketch face data set,it is verified that the method proposed can generate more realistic and clear sketch face images.

关 键 词：素描人脸合成 生成对抗网络 深度学习 对抗学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]