Research On Realistic3D Head Modeling And Facial Expression Synthesis

With the rapid development of the virtual reality technology and the visual technology, three-dimensional virtual human gradually has become a hot research topic in the field of virtual reality.In order to build a personalized virtual human, the three-dimensional head model is very important.In fact,the research on three-dimensional head modeling is widely used in medical plastic, facial morphology, facial expression synthesis and control and other fields.Upon most occasions,we can get a three-dimensional head model from three-dimensional scanning instruments or modeling softwares(e.g. Poser,Maya).But the former needs high costs,and the latter needs professional operators.And those models need a large amount of data to support their realistic structure forms.But onerous data is not good for the control and application.So the emphasis research is how to use less points to build a three-dimensional head model which is easy to control, make the model look realistic, and implement its application.In this thesis, we ues the ASM(Active Shape Mode) method to extract facial feature points information of the input image.Then we build a standard three-dimensional head grid model wehich contains193control data by improving the existing facial three-dimensional information and increasing ears characteristics, teeth characteristics and hindbrain surface model.According to the affine transformation and local positioning adjustment principle, we adjust the feature points and non-feature points of the standard three-dimensional head grid model.Through that we can get the self-adaption three-dimensional grid model which is similar to the input facial image. Then we can build the realistic three-dimensional head model by using texture mapping technology. At last, we synthesize six expressions by combination of AUs(Action Unit) to accomplish the expressional application of the realistic three-dimensional head model.The results show that the created model has a strong sense of reality,and it can create different model for different input facial images. The model has good controllability and wide applicability as well.