3D HMM-based Facial Expression Recognition using Histogram of Oriented Optical Flow

In this paper, we propose a 3D HMM (Three-dimensional Hidden Markov Models) approach to recognizing human facial expressions and associated emotions. Human emotion is usually classified by psychologists into six categories: Happiness, Sadness, Anger, Fear, Disgust and Surprise. Further, psychologists categorize facial movements based on the muscles that produce those movements using a Facial Action Coding System (FACS).  We look beyond pure muscle movements and investigate facial features – brow, mouth, nose, eye height and facial shape – as a means of determining associated emotions. Histogram of Optical Flow is used as the descriptor for extracting and describing the key features, while training and testing are performed on 3D Hidden Markov Models. Experiments on datasets show our approach is promising and robust.