Dynamics analysis of facial expression changes for person identification

We propose a new method for analyzing dynamics of facial expression changes to identify persons. Several methods have been proposed to identify persons using facial images. In most methods, variations in facial expressions are one trouble factor because an input face image does not always contain the same facial expression as training images. However, the dynamics of facial expression changes are one measure of personal characteristics. In the proposed method, facial feature points are extracted using Active Appearance Models (AAMs) in the first frame of each video. They are then tracked using the Lucas-Kanade (LK) based feature point tracking method. Next, the starting and ending frames of facial expression changes are extracted by differences in the facial feature points' position between two successive frames. Finally, a feature vector is obtained as the sequence of the 2D coordinate variations of facial feature points. In the identification phase, an input feature vector is classified by calculating the distance between the input vector and the training vectors using dynamic programming matching (DP matching). We show the effectiveness of the proposed method using facial expression videos of the Facial Expressions and Emotions Database from Technical University of Munich (FEEDTUM database).