基于生成对抗网络的人脸妆容自动迁移方法  

Automatic Face Makeup Transfer Method Based on Generative Adversarial Network

作　　者：颜文胜[1] 吕红兵[2] YAN Wensheng;Lü Hongbing(School of Information Technology Engineering,Taizhou Vocationaland Technical College,Taizhou 318000,China;College of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]台州职业技术学院信息技术工程学院,浙江台州318000 [2]浙江大学计算机科学与技术学院,杭州310027

出　　处：《吉林大学学报（信息科学版）》2022年第3期479-487,共9页Journal of Jilin University（Information Science Edition）

基　　金：浙江省高等教育“十三五”教学改革研究基金资助项目(jg20190884);浙江省教育厅科研基金资助项目(Y202044737)。

摘　　要：为有效解决现有人脸妆容迁移方法训练数据缺乏,以及上妆区域错误等问题,提出了一种基于生成对抗网络的人脸妆容自动迁移方法。方法通过构建生成对抗网络目标函数,采用Encoder-Decoder神经网络生成对抗网络生成器,并基于多层卷积神经网络构建鉴别器,训练算法采用交替优化的方式。仿真实验和方法比对结果表明,该方法在保持素颜妆后图像脸部结构不变的同时,尽可能地体现了参考妆容风格,得到了更协调的上妆效果,具有更佳的对比优势和视觉效果,为人脸妆容自动迁移技术提供了新思路。In order to further solve the problems such as lack of training data and the wrong makeup area in the existing facial makeup transfer methods, an automatic face makeup migration method based on the generation countermeasure network is proposed. This method constructs the objective function of generative adversarial networks, and achieves the generator by encoder-decoder neural network. Meanwhile, it constructs the discriminator based on multilayer convolutional neural network. The training algorithm adopts alternating optimization. The results of simulation experiment and method comparison show that this method keeps the facial structure, and reflects the reference makeup style as much as possible, achieves a more harmonious makeup effect, has better comparative advantages and visual effects, and provides a new idea for the automatic facial makeup transfer technology.

关 键 词：人脸妆容迁移 生成对抗网络 图像风格迁移 损失函数 生成器 鉴别器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]