Semi Adaptive Appearance Models for lip tracking

Many object tracking methods based on Adaptive Appearance Models (online learning methods) have been developed in recent years. One problem that can be found with these methods is how to learn variations in object appearance without errors in the image sequence. This paper introduces a novel method, in which a solution to remove learning errors by using an offline learning is proposed; in addition, our method can be thought of as a generalization of Active Appearance Models, in which the shape model is built manually and object appearance are modeled sequentially in video sequences. Experimental results on lip tracking show that our proposed tracker is functioning accurately.