Rendering convincing, realistic animations of human faces is a long-standing challenge in computer graphics. Not only are the face and its behavior difficult to model, but because the correct appearance of faces is so familiar, subtle imperfections that might go unnoticed in other contexts are immediately perceptible in facial renderings. This exceptional standard of accuracy, together with the strong interest in creating realistic faces for film, games, telepresence, and other applications, makes successfully modeling and rendering faces an important goal that requires the best available tools for realism.
I will present results from the facial animation project that I am currently working on, in which we approach realism by using measurements wherever possible. In particular, we use image-based techniques to measure the geometry of the face, its directional reflectance, its texture, and the motions it makes during facial expressions and speech. Our final animations are produced by incorporating all of these measurements into a model that is used by a physically-based renderer.
This work was done in collaboration with Brian Guenter and others at Microsoft Research and with Steve Westin, Don Greenberg, and others at Cornell University.
Steve Marschner is a post-doc researcher in the graphics group at Microsoft Research in Redmond, Washington. In 1998, he completed his Ph.D. at the Cornell University Program of Computer Graphics under the supervision of Donald Greenberg. He did his undergraduate work at Brown University, where he earned an Sc.B. in Mathematics/Computer Science. Before moving to his present position at Microsoft, he worked in the Digital Photography group at HP Laboratories in Palo Alto.