Automatic 3-D face model adaptation for model-based coding of videophone sequences

For model-based coding of videophone sequences at very low bit rates, an algorithm for the automatic adaptation of a generic 3-D face model to individual faces in videophone sequences is presented. The face model is adapted to the individual facial features of the person-namely, to the eye centers, the eye corners, the eyelids, the irises, the mouth center, the mouth corners, the lips, the eyebrows, the nose, and the chin and cheek contours. First, these facial features are estimated from the videophone sequence. Then, using these estimated facial features, the 3-D face model is adapted to the individual face. Applying the presented algorithm to the videophone sequences Akiyo, Claire, and Miss America, all required facial features are successfully estimated, and face models are adapted. The distance between manually determined facial feature points in the image and projections of the corresponding vertices of the adapted 3-D face model onto the image plane is between 1.5 and 3.5 pels