Multilinear Face Model

K. Yano and K. Harada (Japan)


face modeling, multilinear analysis, range image registra tion, expression synthesis, RBF fitting


We construct face models that have full correspondence among different identities by fitting a template mesh to un derlying range and color images. We use RBF interpolation technique to fit the template mesh with range data and make face model by resampling depth value at each vertex posi tion. We generate expressive 2D vector field from ground truth face images and use the result to construct 3D expres sive face models. Generated face models become a large cartesian products of samples (vertex ¢identity¢AU (Ac tion Unit) ). This data model is analyzed by tensor decom position. Our results show that this data model is useful for some application, in particular for facial animation

Important Links:

Go Back