引入标记分布的人脸表情图像生成  

Facial Expression Image Synthesis with Label Distribution

作　　者：杨静波 赵启军[1] 吕泽均[1] YANG Jingbo;ZHAO Qijun;LV Zejun(College of Computer Science,Sichuan University,Chengdu 610065)

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《现代计算机》2021年第12期78-82,共5页Modern Computer

基　　金：国家重点研发计划项目(No.2017YFB0802300);国家自然科学面上基金项目(No.61773270、61971005)。

摘　　要：随着生成对抗网络在图像生成领域的发展,人脸表情图像生成效果有了显著提升。然而,目前方法往往基于传统表情分类,忽略表情的复杂多样性。然而针对生成多样表情的数据库数据规模较小。为了解决这一问题,提出引入标记分布的人脸表情生成方法。方法在数据量较少的情况下,以标签分布对表情标签进行处理,基于生成对抗网络实现人脸表情图像生成,并在Oulu-CASIA数据库和CFEED数据库上对该方法进行验证。With the development of generative adversarial network in the field of image synthesis,facial expression image synthesis has shown remarkable improvements.However,current methods are often based on traditional facial expression classification and ignore the complex diversity of facial expressions.In order to solve this problem,a facial expression image synthesis method with label distribution is proposed.Under limited training data,the method processes the facial expression tags with label distribution,and realizes the facial expression image synthesis based on the generative adversative network.Moreover,this method was verified on Oulu-CASIA database and CFEED database.

关 键 词：标记分布 图像生成 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]