Modified Active Appearance Models Features for Facial Expression Recognition

The six basic facial expressions (anger, disgust, fears, happy, sadness, surprise) which have been recognized in psychology are commonly used for facial expressions recognition research. In this work, we propose intuitive and effective facial features to recognize the six facial expressions by modifying the original facial feature points used by the original Active Appearance Model (AAM). The Support Vector Machine (SVM) is employed as the base classifier for facial expression classification. The original facial features are two dimensional vertex point position represented by the xand ycoordinates. However, we observe that every facial expression is just a deviation from the original neutral facial expression, so we try to exploit the differences between any facial expression and the neutral expression for facial expression recognition. The experiment results by applying the proposed differential facial feature, the recognition accuracy is higher than by applying the original facial features for all facial expressions.