基于改进深度学习的动画人物面部表情生成方法的研究  

Research on Facial Expression Generation Method of Animated Characters Based on Improved Deep Learning

作　　者：崔婷婷[1] 于海霞 CUI Tingting;YU Haixia(School of Information Engineering and Media,Hefei Technology College,Hefei,Anhui 231000,China)

机构地区：[1]合肥职业技术学院信息工程与传媒学院,安徽合肥231000

出　　处：《九江学院学报（自然科学版）》2021年第4期68-72,108,共6页Journal of Jiujiang University：Natural Science Edition

基　　金：安徽省人文社科重点项目(编号SK2020A0757);安徽省重点科技项目(编号KJ2020A0990);合肥职业技术学院校级人文项目(编号2021SKB13)的成果之一。

摘　　要：在动画人物面部表情设计与生成的研究中,常规的表情生成方法因图像细节特征不明显,导致生成的面部表情真实度不足。为了解决这一问题,提出一种基于改进学习的动画人物面部表情生成方法。以真实的人脸面部表情图像作为依据,利用改进深度学习设计级联分类器,从真实图像中提取出人脸表情特征图像,柔化图像边缘,增强特征细节。为了保证输入图像与生成表情图像之间内容和风格的统一性,从内容约束和风格约束两方面设计损失函数,优化判断网络,并在损失函数的约束下融合特征信息,生成动画人物面部表情。实验结果表明,设计的基于改进深度学习的表情生成方法的特征点定位准确,输入图像与生成图像之间的皮尔逊相关系数高,均方根误差小,生成的面部表情真实感得到了增强。In the research on the design and generation of facial expressions of animated characters,the conventional method of facial expression generation was not obvious because of the image details,resulting in the lack of authenticity of generated facial expressions.In order to solve this problem,an improved learning based facial expression generation method was proposed.Based on the real facial expression image,the improved deep learning was used to design a cascade classifier to extract facial expression feature image from the real image,soften image edges and enhance feature details.In order to ensure the unity of content and style between input image and generated expression image,the loss function was designed from both content constraints and style constraints to optimize the judgment network,and the feature information was fused under the loss function constraints to generate facial expressions of animated characters.The experimental results showed that the proposed method was accurate in locating feature points,had high Pearson correlation coefficient and small root mean square error between input images and generated images,and enhanced the reality of generated facial expressions.

关 键 词：改进深度学习 动画人物 面部表情生成 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]