Subdivision Feedback Based 3D Facial Modeling for E-learning

Online talking with facial animation is an alternative way of face-to-face communication for e-learning. In online talking faces field, facial modeling is very important. Available personalized 3D facial modeling systems cannot refine initial reconstructed 3D facial model automatically; therefore the output model is not very photorealistic. This paper proposes a subdivision feedback based personalized 3D facial modeling algorithm. After conventional facial modeling, in the feedback phase, quality of synthesized faces is estimated by facets’ similarity metric between original face images and synthesized ones. Through subdividing the coarse facets locally and adaptively, we gain new personalized facial mesh. Using cylindrical projection and texture mapping in the forward phase once more, we obtain new facial model. Experimental results show that the model becomes more photorealistic after subdivision feedback.