Feature selection for improved 3D facial expression recognition

Abstract

Automatic recognition of facial movements and expressions with high recognition rates is essential for human computer interaction. In this paper, we propose a feature selection procedure for improved facial expression recognition utilizing 3-Dimensional (3D) geometrical facial feature point positions. The proposed method classifies expressions in six basic emotional categories which are anger, disgust, fear, happiness, sadness and surprise. The most discriminative features are selected by the proposed method based on entropy changes during expression deformations of the face. Developed system uses Support Vector Machine (SVM) classifier organized in two levels. The system performance is evaluated on 3D facial expression database, BU-3DFE. The experimental results on classification performance are superior or comparable with the results of the recent methods available in the literature.