Local motion tracking in semantic-based coding of videophone sequences

Reliable and fast tracking of facial features in 'head and shoulders' scenes is of paramount importance for analysis of motion in model-based coding techniques. A new method of tracking the position of important facial features is presented here. The algorithm enables tracking of local motion of important facial features (eyes open-close, mouth open-close). It is therefore possible to determine not only the spatial location of the investigated facial feature, but also its shape. The method is based on eigenvalue decomposition of the sub-images extracted from the video sequence. The motion and shape of each facial feature is tracked separately. The algorithm can be used for tracking objects in scenes other than 'head-and-shoulders'.