Positive and Negative Expressions Classification Using the Belief Theory

Overview & Synopsis

In this paper, a method based on the transferable belief model is used to classify facial expressions into positive, negative or neutral classes. The classification is based on different visible deformations of facial features. Three kinds of data are extracted from facial deformations: (1) facial distances, (2) nasolabial angle (if any) and (3) facial transient features (if any). For each source of information, a local basic belief assignment is computed. At the end all local belief assignments are fused in order to make a final decision about the classification of the considered expression. Several facial expression databases were used to evaluate the performance of the proposed method. We get more than 95% of correct classification rate to either positive or negative expressions.