Automated Assessment of Emotions - Recognising Facial Expressions with SEEM

KEYWORDS:

ABSTRACT:
    A way to assess facial expressions is to extract thirty vectors which are defined on the facial features eyes, eyebrows and mouth. The distances between these vectors are input to an artificial neural network, that will classify into the basic emotions. This method is semi-automatic. In this report a description is given of a fully automated method to extract the distances between the vectors. This fully automated method, that uses the facial feature shapes, has been implemented for the mouth. Two attempts were made to extract the mouth shape. First, an contour follower was used. Since no results were achieved, a second attempt was made. This resulted on a automatic facial feature extraction system SEEM that is based on vision techniques. The system uses a priori knowledge (anthropometric facial dimensions) and integral projection to locate the face features and to extract the mouth contour at real-time speed.
    A small image sequence was generated to test the extraction of distances between the mouth vectors. These distances were used to train a backpropagation neural network for emotion classification. It can be concluded that the SEEM system is able to extract the distances and that it is possible to use this data to recognise the facial expressions.