NURBS-based facial expression modeling for anthropomorphic interface agent

Facial images play an important role in friendly human-computer interactions as well as in human-human communications. To date, more than 500 polygon patches have been commonly used to generate a 3D human face model. It is not easy, however, to control the vertices of these patches in designing the facial forms and their expressions. This paper first discusses algorithms for the manipulation of non-uniform rational B-Spline (NURBS) parameters to yield realistic 3D facial models and their expressions. It then presents a new method which uses a single NURBS patch to construct a 3D facial model with several expressions. This allows one to design and modify the facial model easily. A real human image can be texture-mapped onto this model. A prototype software shows that the quality and the synthesis speed of the image are satisfactory.