Automatic landmarking and alignment for facial expression analysis

In this paper we develop a fully automatic emotional expression recognition system. We propose improvements on a recent facial landmark localization method. This method is then used to bootstrap a Bezie´r volume tracker for tracking facial features in video. The motion of the facial features is used in classifying the expression of the face. We present extensive cross-database results for the landmarking and expression recognition modules, and show that automatic landmarking significantly improves the subsequent expression recognition.