Game Development Reference
In-Depth Information
the 3D information with the optical flow constraint leads to a linear algorithm that
estimates the facial animation parameters. Each synthesized image reproducing
face motion from frame t is utilized to analyze the image of frame t +1. Since
natural and synthetic frames are compared at the image level, it is necessary for
the lighting conditions of the video scene to be under control. This implies, for
example, standard, well distributed light.
Pighin, Szeliski and Salesin (1999) maximize this approach by customizing
animation and analysis on a person-by-person basis. They use new techniques
to automatically recover the face position and the facial expression from each
frame in a video sequence. For the construction of the model, several views of
the person are used. For the animation, studying how to linearly combine 3D face
models, each corresponding to a particular facial expression of the individual,
ensures realism. Their mesh morphing approach is detailed in Pighin, Hecker,
Lischinski, Szeliski and Salesin (1998). Their face motion and expression
analysis system fits the 3D model on each frame using a continuous optimization
technique. During the fitting process, the parameters are tuned to achieve the
most accurate model shape. Video image and synthesis are compared to find the
degree of similarity of the animated model. They have developed an optimization
method whose goal is to compute the model parameters yielding a rendering of
the model that best resembles the target image. Although a very slow procedure,
the animated results are very impressive because they are highly realistic and
very close to what we would expect from face cloning. (See Figure 9.)
Figure 9. Tracking example of Pighin's system. The bottom row shows the
result of fitting their model to the target images on the top row. Images
courtesy of the Computer Science Department at the University of
Washington.
Search WWH ::




Custom Search