A fusion process based on belief theory for classification of facial basic emotions

This paper presents a system of facial expressions classification based on a data fusion process using the belief theory. The considered expressions correspond to the six universal emotions (joy, surprise, disgust, sadness, anger, fear) as well as to the neutral expression. Since some of the six basic emotions are difficult to simulate by non-actor people, the performances of the classification system are evaluated only for four expressions (joy, surprise, disgust, and neutral). The proposed algorithm is based on the analysis of characteristic distances measuring the deformations of facial features, which are computed on skeletons of expression. The skeletons are the result of a contour segmentation process of facial permanent features (mouth, eyes and eyebrows). The considered distances are used to develop an expert system for classification. The performances and the limits of the recognition system and its ability to deal with different databases are highlighted thanks to the analysis of a great number of results on three different databases: the Hammal-Caplier database, the Cohn-Kanade database and the Cottrel database.