Segmentation of the facial area for videophone applications

A new segmentation scheme for videophone application is described. In model-based image coding, the facial area must be separated from the head-and-shoulder scene in order to fit a parameterised face model to the face of the speaker. Based on the observation that the head and shoulder of the speaker usually undergo two different motions, a dedicated segmentation scheme can be signed. The experimental results on a real image sequence are presented.