Realistic Facial Animation Based On Contour Change-Wrinkle Synthesis

Facial animation has been a topic in computer graphics during the latest 30 years . Realistic facial animation exhibits not only contour change (e.g. wink, mouth opening), but also wrinkle (e.g. forehead furrow, fossette). In addition, the re-targeting of realistic facial animation is also a problem that desires attention.This paper analyzes the present research of realistic facial animation, and sums up those open problems. To solve them, this paper propose realistic facial animation based on contour change-wrinkle synthesis. To model realistic facial animation, we firstly complete the physical character quantization and extraction. The muscle actuation is quantized as FAP while wrinkle is quantized as EigenERI. Then we apply optical flow to extract FAP and PCA to extract EigenERI. After physical character quantization and extraction, we can acquire an example set comprising the correspondence between FAP and EigenERI.The synthesis in this paper is divided into two parallel flows: geometry-controlled contour change, SVM-driven wrinkle prediction. Based on geometry-controlled image warping, geometry-controlled contour change realize contour warping. Given a set of muscle actuation parameters, SVM-driven wrinkle prediction outputs corresponding wrinkle parameters with SVM and applies Wrinkle Enhancement proposed in the paper to reduce noise. After the combination of the two flows, the synthesis system will output realistic facial animation include wrinkle.Due to the incorporation of Wrinkle Enhancement, the synthesis in the paper is robust and can effective restrain the disturbance of the error from SVM prediction on wrinkle. In addition, because the realistic facial animation synthesis is divided intotwo parallel flows which are independent on each other, the synthesis system is easy to be re-targetted to unknown object.Finally, this paper states the shortcomings of the synthesis and propose future research direction.