Performance of Feature Extraction Methods used for Facial Expression Recognition System

Facial expression recognition is challenging task in computer vision. If machines will be able to understand facial expression then human machine interaction will be possible. Happy, angry, fear, surprise, disgust, sad and neutral are the expressions considered for classification. Texture and shape features are extracted using LBP and HOG. In proposed research work performance of LBP and HOG features is compared with reference to seven performance matrices. By combining LBP and HOG features it is possible to get advantage of both features. The features are classified using SVM. The performance is tested with two posed and two wild datasets. Combination of LBP and HOG features gives superior results than single feature method. It enhances the robustness and accuracy of FER system. 97.89% recognition accuracy is achieved by the proposed method. Normalized LBP and HOG features are robust to variation in lighting and contrast. The proposed work will be useful to design robust FER system for wild images.