What perceptible information can be implemented in talking head animations

In this paper, we describe our talking head animation system. We scan and analyze high resolution static 3D faces from multiple subjects and extract deformation characteristics by principal component analysis. The extracted components have similarities across subjects and are used to reconstruct arbitrary facial postures for face animation driven by 3D kinematic data. We also build a simple lip model to estimate lip surfaces not visible to our 3D scanner. We present some results of 3D face analysis, including the effects of varying the number of principal components used to animate a talking head.