Figure 7.6. Comparison of different algorithms in person-independent recognition test. (a):
Algorithm uses geometric feature only. (b): Algorithm uses both geometric and ratio-image
based appearance feature. (c): Algorithm applies unconstrained adaptation. (d): Algorithm
applies constrained adaptation.
illustrated using one of the training example. We compare the expression recog-
nition rates of the proposed method with geometric-feature-only method. The
overall average recognition rate of our method is 71%, while the rate of the
geometric-only method is 59%. Part of the tracking results is visualized in the
accompanying videos [Wen and Huang, 2004]. In the videos, the upper left of
the frame is the input video frame, the upper right is the geometric feature based
tracking visualized by a yellow mesh. The exemplar is shown on the bottom.
We can observe that our method can still track the texture variations when there
are large 3D motions, or under dramatically different lighting conditions.
In this chapter, a hybrid facial motion analysis scheme is presented. In this
scheme, we propose a novel appearance feature based on our flexible appearance