Facial feature localization and adaptation of a generic face model for model-based coding

A method for the adaptation of a generic 3-D face model to an actual face in a head-and-shoulders scene is discussed, with application to video-telephony. The adaptation is carried out both on a global scale to reposition and resize the wireframe, as well as on a local scale to mimic individual physiognomy. To this effect a hierarchical scheme is developed to extract the semantic features in the head-and-shoulders scene, such as silhouette, face, eyes and mouth, using a knowledge-based selection mechanism. These algorithms, which are to be an integral part of a general model-based image coder, are tested on typical videophone sequences.