-------------------------------------------------------------------------------
                                       
                                       
                   B O S T O N   U N I V E R S I T Y
                                       
                      Computer Science Department
                                       
                          C O L L O Q U I U M

                Analysis and Synthesis of Talking Faces

                             Lionel Reveret
			    
                 Computational Perception Laboratory
                    Georgia Institute of Technology


   			 Wednesday, February 7
			       3:00 PM
                       (Coffee served at 2:45PM)

                                       
                         Seminar Room / MCS 135
                                       
Using a face as a medium for human-machine communication has motivated
a lot of efforts in computer animation and computer vision.  Various
models of the face geometry and its deformation have been proposed,
whether for 3D animation or tracking from video sequence by
projection.  Between flexibility and stability, the question of the
appropriate level of control of a face model remains largely dependent
on the context of its application.

In this talk, I present some results on analysis and synthesis of
facial motion in the specific case of speech production. After a
review of some previous results in the perception of audiovisual
speech, it is shown how a functional 3D model of speech motion has
been obtained from a statistical analysis of accurate 3D face
data. The particularity of the model is to be controlled by a set of
only 6 parameters, which represent primary gestures in speech
articulation. After this learning stage, the projection of the 3D
model is used in a video tracking process to extract the 6 control
parameters from speakers' images. This tracking is based on an optimal
alignment between each image of the original sequence and the
configuration of the model textured with an initial image. Being based
on texture, the video tracking makes no explicit assumption on tone
color or the presence of markers.  Finally, an automatic generation of
facial animation from text input is presented.


Host:  Stan Sclaroff (sclaroff@cs.bu.edu)

-------------------------------------------------------------------------------
For colloquium info, including directions, see http://cs-www.bu.edu/colloquium
-------------------------------------------------------------------------------