Face recognition has important applications
in psychology and biometric-based authentication, which increases the
need for developing automatic face identification systems.
Psychologists have long been studying the link between
symmetry and attractiveness of the human face, but based on
qualitative human judgment alone. The use of objective facial asymmetry
information in automatic face recognition tasks is relatively new. The
current paper presents a statistical
analysis of the role of facial asymmetry measures in face
recognition, under expression variation. We first describe a baseline
classification method and show that the results are comparable with those
based on certain popular (non-asymmetry
based) classes of features used in computer vision. We find that facial
asymmetry further improves upon the classification
performance of these popular features by providing complementary
information. Next, we consider two resampling methods to improve
upon the baseline method used in previous work, and present a detailed
comparison study. We demonstrate that resampling
methods succeed in obtaining near perfect classification results on a
database of 55 individuals, a statistically significant improvement over the
baseline method. Results regarding the role of asymmetry of different
parts of the face in distinguishing between individuals, expressions
and between males and females are also reported as additional aspects
of the study.
Keywords: asymmetry, bagging, classification, feature
subset, random subspace, resampling