Evidential reasoning for facial gestures recognition from cartoon images

Facial gesture analysis is a challenging subject for automatic recognition. There are uncertainties of the possible errors on feature extraction or of more than one gesture having similar facial features. To tackle this problem, a new approach for facial gesture analysis is proposed using evidential reasoning. Before any recognition, different expressions must be identified. The six universal expressions are used and tested. They are: anger, disgust, fear, happiness, sadness and surprise. In order to computerize them, a set of parameters is used to encode these expressions. A subset of FACS (the Facial Action Coding System) is selected and modified for this purpose. Cartoon drawings are designed and constructed for different expressions. Several facial marks are identified, and the relationships between them are measured. They are used as feature vectors for recognition purpose. In the recognition phase, the Dempster-Shafer theory of evidential reasoning is applied to combine the evidence as represented by the facial features. This approach generates hypotheses, given the set of facial evidence. It accumulates the corresponding confidence, and calculates the support and the plausibility of the resulting hypothesis. This study demonstrates the feasibility of applying Dempster-Shafer theory to facial gesture recognition. The performance is greatly affected by the number of FACS codes used to describe an expression