Learning combined features for automatic facial expression recognition

Abstract

Facial expressions are one of the most natural and powerful means for the human being in his social communications, whether to share his internal emotional states or to display his moods or intentions, which, in fact, may be true or simply played in a theatrical way. Given the numerous and variety of applications that can be easily planned, building a system able to automatically recognising facial expressions from images has been an intense field of study in recent years. In this paper, we propose a new framework for automatic facial expression recognition based on combined features and deep learning method. Before the feature extraction, we use Haar feature-based cascade classifier in order to detect then crop the face in the images. Next, we extract pyramid of histogram of gradients (PHOG) as shape descriptors and local binary patterns (LBP) as appearance features to form hybrid feature vectors. Finally, we use those vectors for training deep learning algorithm called deep belief network (DBN). The experimental results on publicly available datasets show promising accuracy in recognising all expression classes, even for experiments which are evaluated on more than seven basic expressions.