Facial image reconstruction and manipulation from measurements obtained using a structured lighting technique

Computer generation and animation of facial images such as that described by Park (1982) rely on manual measurements of facial profiles and image reconstruction using solid modelling techniques. The resultant images tend to have a rather artificial cartoonlike quality which lacks realism when compared with the originals. Our method automates the measurement of facial dimensions and uses a new texture mapping technique to reproduce the appearance of complex surface detail. Facial dirnensions are obtained by scanning the face of a live subject with a line of light from a low power laser. Tile method uses simple special purpose hardware to process raw data and allows a facial scan to be made in under five seconds, obviating the need for the subject to remain still for an extended period. A realistic reconstruction of the original image is made by first generating a 3-D polygon model of the original subject and then using a texture mapping technique to paste a real image over the surface of the model. Computed directional illumination and geometric transforms are applied to allow the reconstructed image to be scaled, translated and rotated. Animation is achieved by blending in sequences of subpictures which are obtained from an original training set. A key feature of the method is that the image and subpictures relate to a static frontal view of the original subject and are invariant to the application of scaling, rotation and translation transforms.