Abstract
Model-based coding is usually used in virtual conference system, in which the facial expressions are the focus of all users. In model-based coding, by receiving the facial animation parameters, a 3-D facial model can be animated with some expressions, but the visual perception is usually not good enough. We intend to develop a facial expression cloning system which can make a 3-D facial model lifelike.We transfer a person’s subtle facial change to another’s by using the expression ratio image (ERI). The expression ratio image is the result of expression image divided by neutral image. Controlling the lighting condition strictly, expression ratio image can be well transferred to another. Finally, an expressive 3-D facial model will be shown. On the other hand, we intend to achieve an intelligent system which can make a 3-D facial model lifelike and doesn’t need to transmit any more information, i.e., we intend to model the FAP to ratio image conversion. In order that each FAP follows a unique ratio image, the FAP and the expression ratio image have to be considered as jointly distributed. We use the parametric coefficients of FAP and ratio image, which is dimension-reduced by PCA, and use Gaussian mixture model (GMM) to model the FAP to ratio image conversion. Thus, an automated system which achieves the desired lifelike 3-D facial model is developed.