| dc.description.abstract | In the applications of virtual reality, virtual actors (avatars) are commonly used. The key part of virtual actors is just the virtual face. The principal function of virtual faces is facial expression; however, to render expressions on virtual faces is tedious and time-consuming. Thus we expect to develop an automatic system to “reuse” the existed facial expressions.
In this study, we propose an approach of facial motion cloning, which is used to transfer pre-existed facial motions from one face to another. The face models have different characteristics, shapes, the scales of facial features, and so on, but expressions can still be accurately duplicated after precise computation of the scales of facial motions.
The face that provides the original motions is called the “source face,” and the face that will be added on the copied motions is called the “target face.” Facial motions are represented by sets of “morph targets,” which record the displacement vectors of all face vertices between the neutral state and a particular motion.
There are two major steps in the proposed system. The first step is to correspond the two face models according to their facial features. In the step, we must use the facial feature points. We project the face models onto a 2D plane, and then re-triangulate the model according to the feature points. The corresponding relationship of the vertices of the two face models is acquired by calculating the barycentric coordinate. The second step is to clone the facial motions. We duplicate the motions from the source face to the target face, and calculate the scale of facial features between the two faces to get the correct motion scale. The facial motion animation is expected to be demonstrated in real-time; thus we also consider the fast algorithms to develop the cloning system. | en_US |