Facial Action Unit Detection Using Active Learning and an Efficient Non-linear Kernel Approximation

This paper presents large-scale naturalistic and spontaneous facial expression classification on uncontrolled webcam data. We describe an active learning approach that helped us efficiently acquire and hand-label hundreds of thousands of non-neutral spontaneous and natural expressions from thousands of different individuals. With the increased numbers of training samples a classic RBF SVM classifier, widely used in facial expression recognition, starts to become computationally limiting for training and real-time performance. We propose combining two techniques: 1) smart selection of a subset of the training data and 2) the Nystrom kernel approximation method to train a classifier that performs at high-speed (300fps). We compare performance (accuracy and classification time) with respect to the size of the training dataset and the SVM kernel, using either an RBF kernel, a linear kernel or the Nystrom approximation method. We present facial action unit classifiers that perform extremely well on spontaneous and naturalistic webcam videos from around the world recorded over the Internet. When evaluated on a large public dataset (AM-FED) our method performed better than the previously published baseline. Our approach generalizes to many problems that exhibit large individual variability.