Facial Expression Recognition with 3D Deformable Models

We present a fully automated three-dimensional model-based, real-time capable approach to distinguish six universal facial expressions from visual images of human faces. The face model is fitted to the images of a publicly available data base. From the model parameters two sets of features are computed, person-specific and non-person-specific to estimate the facial expression visible in the current image sequence. We integrate the complex, state-of-the-art Candide-3 face model, which is specifically appropriate for facial expression recognition and is also publicly available, which further increases the comparability of our approach. Our experimental evaluation presents recognition rates of the six basic facial expressions defined by Ekman et al.