Feature Extraction based on Local Directional Pattern with SVM Decision-level Fusion for Facial Expression Recognition

Facial expression recognition, as one of the important topics in pattern recognition and computer vision, has broad applications in fields of human-computer interaction, psychological behavior analysis, image understanding. This paper presents a novel facial expression recognition method based on global and local features extraction and facial recognition using decision-level fusion. We first extract Local Directional Pattern (LDP) global features of the whole face which can guarantee basic expression difference and decrease the influence of no-facial region meanwhile, and then the Local Directional Pattern Variance (LDPv) descriptor is used to extract local features of regions of eyes and mouth to extrude their contribution on expression changes. After feature extraction, PCA technique is utilized to reduce dimension of input feature space. Finally, in order to avoid redundant feature repeat we don't use feature fusion with simple concatenation, a decision-level fusion for global LDP feature and local LDPv feature by Support Vector Machine (SVM) is selected to recognition respectively. Furthermore, we also research the optimal parameters for regions-dividing and weight of LDPv. The proposed method is investigated on two standard databases Cohn-Kanade and JAFFE, and extensive experimental results indicate the effectiveness.