Abstract
Realistic facial animation transfer from one individual to others
has been a persistent challenge. In this paper, we present an
effective method that transfers facial animation from 2D videos onto
3D faces in a visually pleasing manner. Our method is based
on a Laplacian deformation framework. We represent the facial
animation with the displacements of a set of feature points. By the
assumption that the feature points move only in the X-Y
directions, we can map the displacements of the feature points from
a 2D video to a 3D face. These displacements are used to drive the
Laplacian deformation and calculate the deformed positions of the
non-feature points on the 3D face.
The approach produces accurate, realistic and smooth transfer.
Furthermore, the method is efficient and practical, and the
interface is intuitive. The proposed technique outperforms previous
methods based on machine learning and anatomy in terms of speed and
applicability. Our method is useful for a wide range of
applications, such as, avatars,
character animation for 3D films, computer games, and online
chatting. The versatility of our approach is demonstrated by some
special effects, such as expression exaggeration and expression
imitation.
PDF
|
|
|