APPLICATION EXAMPLES OF THE FACE PROCESSING FRAMEWORK - 3D Face Processing: Modeling, Analysis and Synthesis - page 106

Figure 9.3. (a) The interface for the student. (b) The interface for the instructor.

child. Preliminary results suggest that the synthetic avatar helps the children
more patient in the learning session.

2.3 Future work
In this proactive HCI environment, the facial motions of the children are very
challenging to analyze compared to database collected in controlled laboratory
conditions. One reason is that real facial motions tend to be very fast at certain
occasions. These fast motions can cause the current face tracker lose track. In
the future, we need to future improve the speed of the face tracker so that it can
capture fast facial motions in real-life conditions. Another reason is that real
facial expressions observed in real-life environment are more subtle. Therefore,
the performance of the face expression recognition can be affected. We plan to
carry out extensive studies of facial expression classification using face video
data with spontaneous expressions. In this process, we could be able to improve
the expression classifier.
Another future direction of improvements is to make the synthetic face avatar
more active so that it can better engage the children in the exploration. For
example, we plan to synthesize head movements and facial expressions in the
context of task states and user states. In this way, the avatar can be more lifelike
and responsive to users' actions in the interaction. One possible direction is
explore the correlation of speech and head movements as the work by Graf et
al. [Graf et al., 2002].

3. Summary
In this chapter, we have described two applications of our 3D face processing
framework. One application is to use face processing techniques to encode face