Particle swarm optimisation based AdaBoost for facial expression classification of still images

This paper describes an application of a particle swarm optimisation based AdaBoost algorithm to classify human facial expressions. The particle swarm is used to choose optimal Haar features for constructing weak classifiers within AdaBoost. This algorithm is trained using the Japanese Female Facial Expression dataset and tested on the Cohn-Kanade AU-Coded Face Expression Database. The results show some improvement in accuracy over AdaBoost as well as a reduction in duplicated weak classifiers. In particular, the time taken for training was dramatically reduced from an average of 106 minutes and 41 seconds to a mere 4 minutes and 39 seconds.