Facial expression recognition using enhanced local binary patterns

Abstract

Facial expression, a non-verbal communication, is a means through which humans convey their inner emotional state, thus playing an important role in social interaction and interpersonal relations. Facial expression recognition plays a significant role in human-computer interaction as well as various fields of behavioral science. There are six known classes of emotional state which are anger, disgust, fear, happiness, sadness and surprise, associated with their respective facial expressions, according to Ekman's studies. Humans recognize facial expressions almost effortlessly and without delay, but this is quite challenging for digital computers. The paper presents facial expression recognition using local binary patterns. The main contribution of the paper is the feature selection applied, in which the high variance LBP pixels are selected to represent faces. By selecting the high variance pixels based on LBPs, the recognition rates were improved significantly. The tests are completed on the BU-3DFE database. The experiments show that after applying feature selection, the recognition rates are improved by 11%.