Automatic face location detection and tracking for model-based video coding

One major problem in the model-based coding of facial images is the automatic fitting of the initial wireframe model (WFM) to the facial image. In this paper, the solutions for automatic fitting and adaptation of the initial WFM to the facial image are given. The adaptation is performed on a global scale through repositioning and resizing of the WFM. For this purpose, main facial feature contours are extracted using deformable templates and active contour models. The WFM is then altered by affine transforming some of its major control nodes to fit the facial feature points on the facial image, and the rest of the nodes are interpolated relative to the movement of these nodes. The automatically adjusted WFM is found to be accurate when compared to the manually adjusted one.