摘要
表情动画作为语音驱动人脸动画的一部分,在增加人脸动画逼真性方面起着重要的作用,但已有的工作没有定量分析人脸表情动画与语音之间的关系.文中通过研究人脸表情动画与语音的相关性,采用典型相关性分析方法(CCA)定量分析两者之间的内在联系,得出这些关系直观的量化的结论.首先计算人脸表情动画与语音的典型相关性系数,衡量两者的相关程度;然后分析人脸表情动画与语音的典型负荷、典型交叉负荷等数据,并挖掘两者内部各分量之间的联系,由此得出直观的量化的结论.最后验证了结论的稳定性.分析结果表明两者具有强相关性,并揭示了人脸表情动画各成分与语音声学特征之间的具体内在联系.文中成果可为语音驱动人脸动画技术提供理论参考及结果评价依据.
The facial expression animation,as a component of the speech driven facial animation,plays a very important role in enhancing the verisimilitude of facial animation.However,previous works haven＇t quantitatively analyzed the correlations between facial expression motion and speech.In this paper,we adopt canonical correlation analysis（CCA） to quantitatively analyze the correlations between facial expression motion and speech,and reach the intuitive and quantitative conclusions.First,we calculate the CCA between facial expression motion and speech to measure their degree of correlation.Then we analyze the Canonical Loadings,Canonical Cross Loadings and other analysis data between facial expression motion and speech,find out the specific internal relations and draw the intuitive and quantitative conclusions.Finally,we verify the stability of the conclusions.The analysis result shows that the two are strongly correlated and reveals the specific internal relations between the components of facial expression motion and the acoustic features.This article can be used as theoretical reference and judging criterion for speech driven facial animation technique.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2011年第5期805-812,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60970086)
国家自然科学基金委广东联合基金重点项目(U0935003)
关键词
人脸表情动画
典型相关性分析
facial expression animation
canonical correlation analysis
