Multi-View Pose and Facial Expression Recognition

Multi-view facial expression recognition is important in many scenarios, as frontal view images are not always available. In this paper, we investigate facial expression recognition from frontal to profile view. Few works have investigated this issue on live captured data. A recent database, multi-pie, allows empirical investigation of facial expression recognition for different yaw angles. Experiments are carried out on 100 subjects over 7 poses for 6 facial expressions ( neutral, smile, surprise, squint, disgust and scream). Opencv frontal and profile face detectors are used to locate the face region. Head pose classifiers and pose dependent facial expression classifiers are trained using multi-class support vector machines (SVM). We investigate multi-scale local binary patterns (LBPms) as well as local gabor binary patterns (LGBP) as texture descriptors.