Action unit recognition based on motion templates and GentleBoost

To identify people´s facial expressions, we need to accurately identify the Action Unit´s motion. Generally, we locate facial feature points in images and then track them to identify Action Unites´ motion. However, some Action Unites have not a clear outline and they can not be located and tracked actually. So we proposed an approach which can quickly and automatically identify Action Unites using Motion History Image feature based boosted classifiers. The detected face region is then divided into several relevant regions of interest, each of which contains an Action Unit. Through scanning the Motion History Image of the regions, we can get some motion segmentations. We classify these motion segmentations to identify whether there is an Action Unit. We tested our method with several videos, and the method has achieved an identifying rate of 97%.