A combined approach to feature extraction for mouth characterization and tracking

A statistical active appearance model (AAM) is developed to model and track the lips. A major challenge in modelling the mouth area, considered the most deformable facial feature, is the weak contrast between lips and skin colour. In order to perform accurately, AAM requires a strong initialization point. In our approach, key information is firstly extracted using chrominance analysis. Then optimal parameters of the model are determined using AAM fitting procedure. A detailed description of our model and its corresponding fitting algorithm are given and preliminary results on different databases are summarised.