3d Facial Animation Based On Mpeg-4 Standard

The high development of multimedia technology requires that computers can communicate with human by rich behavior emotion. Although there is a strong resemblance among the facial features of individuals, human faces are actually distinct. Besides, faces survey most of the comprehensive information such as emotion and spirit condition, so human face has been an important research field in computer vision and computer graphics. The dissertation focuses on the following three aspects: MPEG-4-based facial action coding system, text driven facial animation and speech driven facial animation, which have always been challenges in computer vision and computer graphics.First of all, the background and content are given. Then related work, including face modeling, face animation, lip animation and MPEG-4 facial specification, is classified and compared. Issues on the merits and flaws of these approaches are followed.A facial action coding system based on FAPs of MPEG-4 is presented. The system is divided into three layers from top to foot, MPA layer, FC layer, and FAP layer. Supposed that we have already gotten the 68 FAPs, FCs can be calculated from these FAPs. Each MPA is composed of several FCs. So a viseme can be described by a set of MPA parameters. During facial animation, B-spline interpolation is used to generate the immediate frames between key visemes, then the facial motion described by MPA sequences is converted to FAPs by the FACS engine, these FAPs are the ultimate parameters to animate the 3-dimension talking face.In order to obtain the static viseme library as the basis for facial animation, a clustering algorithm of visemes based on parametric lip model is designed. The lip is parameterized by superior and anterior bend for upper and lower lip, width of the mouth, weight of lips' radium. The experiment result shows that parametric lip model is more robust than feather points in clustering mouth visemes.Based on the FACS and dusting algorithm of visemes, we present a system to animate 3-dimension face driven by text. The phoneme stream is obtained from the input text. The key visemes are selected from the static viseme library with the corresponding phonemes. The immediate frames are generated by the coarticulation model based on dominance functions, he silence between words is simulated by an adaptive silence model. The boundary model is employed to lead the dominance function to zero at the border of the current calculating window. The target positions of bilabial stops and fricatives are reached exactly by frequency-conversion sampling.Compared with text-driven facial animation, speech-driven facial animation must do speech recognition. The IFNET based on Chinese mandarin pronunciation rules is presented to achieve speech recognition. The result phoneme stream of speech recognition can be considered as theinput of text-driven facial animation system to synthesize talking face.The algorithms described in this paper, can be exploited in many applications such as video conferencing, games, virtual "showman", online chatting and virtual reality, etc..