Geometric Feature-Based Facial Expression Recognition in Image Sequences Using Multi-Class AdaBoost and Support Vector Machines

摘要

Facial expressions are widely used in the behavioral interpretation ofemotions, cognitive science, and social interactions. In this paper, we presenta novel method for fully automatic facial expression recognition in facialimage sequences. As the facial expression evolves over time facial landmarksare automatically tracked in consecutive video frames, using displacementsbased on elastic bunch graph matching displacement estimation. Feature vectorsfrom individual landmarks, as well as pairs of landmarks tracking results areextracted, and normalized, with respect to the first frame in the sequence. Theprototypical expression sequence for each class of facial expression is formed,by taking the median of the landmark tracking results from the training facialexpression sequences. Multi-class AdaBoost with dynamic time warping similaritydistance between the feature vector of input facial expression and prototypicalfacial expression, is used as a weak classifier to select the subset ofdiscriminative feature vectors. Finally, two methods for facial expressionrecognition are presented, either by using multi-class AdaBoost with dynamictime warping, or by using support vector machine on the boosted featurevectors. The results on the Cohn-Kanade (CK+) facial expression database show arecognition accuracy of 95.17% and 97.35% using multi-class AdaBoost andsupport vector machines, respectively.