Automated facial conformation for model-based videophone coding

In order for model-based coding to be used on a head-and-shoulders video sequence, such as would be encountered in a videophone application, it is first necessary to form a model of the head. This model must be correctly conformed to the subject's face as it appears at the beginning of the video sequence transmission, so the subsequent deformations of the model accurately depict changes in the face and head from frame to frame. This paper proposes a method which allows this conformation to be performed entirely automatically. The method presented utilizes flexible template matching and a steepest-gradient search method to determine the optimal conformation of a generic facial model to represent a given single frame of a face. Results illustrating the success of the method are given and the limitations of the algorithm are discussed.