Local Facial Asymmetry for Expression Classification

Abstract

We explore a novel application of facial asymmetry: expression classification. Using 2D facial expression images, we show the effectiveness of automatically selected local facial asymmetry for expression recognition. Quantitative evaluations of expression classification using local asymmetry demonstrate statistically significant improvements over expression classification results on the same data set without explicit representation of facial asymmetry. A comparison of discriminative local facial asymmetry features for expression classification versus human identification is given.