Graphics Reference
In-Depth Information
Chapter 7
In this chapter, we discuss the face motion analysis using flexible appear-
ance model described in Chapter 6. Our goal is to utilize appearance cues to
improve the detection and classification of subtle motions which exhibit simi-
lar geometric features. For this purpose, we design novel appearance features
for the analysis of these detailed motions. Compared with most existing ap-
pearance features, our appearance features are less illumination dependent and
less person dependent. In our facial expression classification experiment, we
show that this appearance features improve the classification performance under
variations in lighting, 3D poses and person.
In Section 1, we first describe the proposed novel appearance features and
explain how the dependencies on illumination and person are reduced. Then we
describe an online appearance model adaptation scheme to further improve the
performance in changing conditions. Next, in Section 2, experimental results
on facial expression recognition are presented to demonstrate the efficacy of
the novel appearance features.
1. Model-based 3D Face Motion Analysis Using Both
Geometry and Appearance
Given a face video, we can employ both geometric and appearance features to
analyze the facial motions. The system diagram for the hybrid motion analysis
is illustrated in figure 7.1. First, we use a geometric-based method described in
Chapter 4 to estimate 3D geometric deformation. The geometric deformation
features are extracted as the coefficients of MUs. Figure 7.2(b) shows a snapshot
of the geometric tracking system, where a yellow mesh is used to visualize the
geometric motions of the face. The input video frame is shown in Figure 7.2(a).
Search WWH ::

Custom Search