Keith Waters
Compaq Cambridge Research Laboratory
A Position Statement for Panel 3: 3D Modeling and VR
The 1998 International Workshop on Very Low Bitrate Video Coding
High Fidelity Facial Animation

The ability to create highly realistic computer generated faces has significantly advanced over the past few years. This has been due in part, to the advent of 3D scanners and photometric techniques capable of creating highly detailed images of the face. In addition algorithms can now emulate muscle and skin to mimic facial expressions, and accurate lip-synchronization can be achieved with both synthetic and real speech. Concurrently very low bit rate coding schemas, such as the MPEG-4 Face and Body Animation (FBA) specification, demands high fidelity facial animation to be controlled at very low bit rates. Despite these significant advances some simple issues humble facial animation. As the realism of the face increases, in particular making the face look more like real people, we become much less forgiving of modeling and animation imperfections. Put simply: if it looks like a person it should behave like a person. This is due to the fact that we are extremely competent at reading small and very subtle facial characteristics in everyday life. In fact there is strong evidence suggesting that our brains are even "wet-wired" to interpret images of faces.

The creation of non-human avatars such as dogs, cats and cartoons, overcome these limitations because we have no experience of such talking characters. Essentially we are desensitized to imperfections in the modeling and animation because we have no experience of talking dogs and cats. Taking this one step further, two dots and an upward curving line can convey as much information about the emotion of happiness as a complex 3D facial model that contracts simulated skin and muscles at the corner of the mouth. Consequently, a challenge for facial animation is not only to create cartoon characters, but highly realistic, high fidelity faces.

The production of high fidelity facial animation remains elusive; animation capable of deceiving the eye -- a Turing test for the face -- has yet to be demonstrated. Is it possible? Will it be possible one day, for a computer-generated face to be passed-off as the real thing?