Lipless Tracking and Emotion Estimation

Automatic human lip tracking is one of the key components tomany facial image analysis tasks, such as, lip-reading and emotion from lips. It has been a classical hard image analysis problem over decades. In this paper, we propose an indirect lip tracking strategy: `lipless tracking´. It is based on the observation that many of us don´t have clear lips and some even don´t have visible lips. The strategy is to select and localize stable lip features around the mouth for tracking. For this purpose deformable contour-segments are modelled based on lip features and tracking is done using dynamic programming and Viterbi Algorithm. The strength of proposed algorithm is demonstrated in emotion estimation domain. Finally, real-time video experiments performed on private and publicity available data set (MMI face database) have shown the robustness of our proposed lipless tracking technique.