Applications - 3D Face Modeling, Analysis and Recognition - page 190

Table 5.11 Comparison of 3D Dynamic Facial expression Recognition Approachs

Authors Features Classification RR 3 (%)

Sun and Yin (2008) 12 Motion Units HMM 70.31

Sun and Yin (2008) Tracking model HMM 80.04

Sun and Yin (2008) Curvature+Tracking model HMM 82.19

Sun and Yin (2008) Curvature+Tracking model 2D-HMM 90.44

Sandbach et al. (2011) FFD+Quad-tree GentleBoost+HMM 73.61, 81.93

Le et al. (2011) Level curve-based HMM 92.22

Fang et al. (2011) LBP-TOP SVM-RBF 74.63

Drira et al. (2012) Geometric Mean Deformation LDA-Random Forest 93.21

It can be noted that the lower recognition was obtained for the fear (Fe) expression (91.24%),
which is mainly confused with the anger (An) and disgust (Di) expressions. Interestingly, these
three expressions capture negative emotive states of the subjects so that similar facial muscles
can be activated. The best classified expressions are happiness (Ha) and surprise (Su) with
recognition accuracy of 95.47% and 94.53%, respectively.

5.5.3 Discussion and Comparative Evaluation
To the best of our knowledge, the only four works reporting results on expression recognition
from dynamic sequences of 3D scans are Sun and Yin (2008), Sandbach et al. (2011), Le et al.
(2011), and Fang et al. (2011). These works have been evaluated on the BU-4DFE data set,
but the testing protocols used in the experiments are sometimes different, so that a direct com-
parison of the results reported in these papers is difficult. In the following, we present in Table
5.11 the last results published on 3D facial expression recognition from dynamic 3D scans.

Exercises
1. What are the main limitations of 2D facial expression recognition systems?

2. We consider the following ROC Figure 5.21 curves of two 3D face authentication
approaches evaluated using the same experimental protocol:
1. How do you interpret these curves?
2. You have been asked as an expert in Biometrics to choose between these two approaches
to design two different access control systems. Complete the table below. Explain your
choices?

Approach 1 Approach 2

Access Control system for Canteen for children
Access Control system for safe of a bank

3. Give one application of facial expression recognition in medical application.