
软件学报 ISSN 1000-9825, CODEN RUXUEW E-mail: jos@iscas.ac.cn
Journal of Software, 2022,33(4):1516−1526 [doi: 10.13328/j.cnki.jos.006482] http://www.jos.org.cn
©中国科学院软件研究所版权所有. Tel: +86-10-62562563
基于 StarGAN 和类别编码器的图像风格转换
∗
许新征
1,2
,
常建英
1
,
丁世飞
1,2
1
(中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116)
2
(矿山数字化教育部工程研究中心(中国矿业大学), 江苏 徐州 221116)
通信作者: 许新征, E-mail: xuxinzh@163.com
摘 要: 图像风格转换技术已经融入到人们的生活中, 并被广泛应用于图像艺术化、卡通化、图像着色、滤镜处
理和去遮挡等实际场景中, 因此, 图像风格转换具有重要的研究意义与应用价值. StarGAN 是近年来用于多域图
像风格转换的生成对抗网络框架. StarGAN 通过简单地下采样提取特征, 然后通过上采样生成图片, 但是生成图
片的背景颜色信息、人物脸部的细节特征会与输入图像有较大差异. 对 StarGAN 的网络结构进行改进, 通过引入
U-Net 和边缘损失函数, 提出了用于图像风格转换的 UE-StarGAN 模型. 同时, 将类别编码器引入到 UE-StarGAN
模型的生成器中, 构建了融合类别编码器的小样本图像风格转换模型, 实现了小样本的图像风格转换. 实验结果
表明: 该模型可以提取到更精细的特征, 在小样本的情况下具有一定的优势, 以此进行图像风格转换后的图片无
论是定性分析还是定量分析都有一定的提升, 验证了所提模型的有效性.
关键词: 图像风格转换; 生成对抗网络; StarGAN; U-Net; 类别编码器
中图法分类号: TP391
中文引用格式: 许新征, 常建英, 丁世飞. 基于 StarGA N 和类别编码器的图像风格转换. 软件学报, 2022, 33(4): 1516–1526.
http://www.jos.org.cn/1000-9825/6482.htm
英文引用格式: Xu XZ, Chang JY, Ding SF. Image Style Transfering Based on StarGAN and Class Encoder. Ruan Jian Xue Bao/
Journal of Software, 2022, 33(4): 1516−1526 (in Chinese). http://www.jos.org.cn/1000-9825/6482.htm
Image Style Transfering Based on StarGAN and Class Encoder
XU Xin-Zheng
1,2
, CHANG Jia n- Ying
1
, DING Shi- Fei
1,2
1
(School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116 , China)
2
(Engineering Research Center of Mining Digitalization of Ministry of Education (China University of Mining and Technology), Xuzhou
221116, China)
Abstra ct : The image style transferring technology has been widely integrated into people’s life, and it is widely used in image artistry,
cartoon, picture coloring, filter processing, and occlusion removal of the practical scenarios, so image style transfering has an important
research significance and application value. StarGAN is a generative adversarial network framework for multi-domain image style
transfering in recent years. StarGAN extracts features through simple down-sampling, and then generates images through up-sampling.
Nevertheless, the background color information and detailed features of people’s faces in the generated images are quite different from
those in the input images. In this study, by improving the network structure of StarGAN, after analyzing the existing problems of the
StarGAN, a UE-StarGAN model for image style transfering is proposed by introducing U-Net and edge-promoting adversarial loss
function. At the same time, the class encod er is introduced into the generator of UE-StarGAN, and a small sample image style trans fer in g
model is designed to realize the small sample image style transfer. The results of this experiment show that the model can extract more
detailed features, have some advantages in th e case of small sample size, and to a certain extent, th e qualitative and quantitative analysis
resul ts of th e i mag es c an b e i mpro v ed aft er the ima g e st y le tr an s f eri ng , whi c h verifi e s th e effe ct i ven es s of t h e p ro p o sed mod el .
∗ 基金项目: 国家自然科学基金(61976 217, 61976216)
本文由“面向开放场景的鲁棒机器学习”专刊特约编辑陈恩红教授、李宇峰副教授、邹权教授推荐.
收稿时间: 2021-06-01; 修改时间: 2021-07-16; 采用时间: 2021-08-27; jos 在线出版时间: 2021-10-26
评论