Facial expression recognition based on active appearance model & scale-invariant feature transform

The active appearance model (AAM), one of the most effective facial feature localization methods, is widely used in facial expression recognition. However, the results of this model with non-frontal face are not ideal. Thus, we propose a new method for facial expression recognition based on AAM and scale-invariant feature transform (SIFT). The proposed method uses AAM to locate the feature points of a facial expression image. SIFT descriptors are then utilized to describe these feature points, and the gradient direction histogram of the pixels surrounding these points are used to form point feature vectors. The chi-square distance and the nearest neighbor classifier is applied to accomplish the facial expression recognition task. The experimental results from standard expression databases and multi-posture expressions show that the proposed method not only improves the recognition rates of the frontal face but also has better robustness for non-frontal facial expressions under some deflection angles.