Virtual Human Face Synthesis Based On Video Frame Sequence

Virtual human face synthesis is one of the key techniques that combine 2D human face images and video surveillance system with 3D human face recognition and human face modeling. It is an extremely challenging project worthy of theoretical research.The current methods are almost based on 3D face and texture databases. For building database is difficult and the algorithm is comparatively complicated with huge computation, there is still a big gap between the research and application. It is necessary to find an effective synthesizing method.First, a specified human facial model is created by modifying the built-in general model depending on the individual facial characteristics, which are extracted from human face images appeared in video sequences. Then, the viewpoint-irrelevant texture is obtained through applying multi-resolution technology to the transition regions, which are determined upon the matching result of the model and image. Finally, the individual virtual human face is synthesized by texture mapping on adjusted model.In addition, based on the MPEG-4 facial animation parameters standard, this paper actualizes some typical virtual facial expression by modifying the parameters of the corresponding parts in human face according to the characteristics during producing expression.In order to solve the problem of unnatural performance of virtual human face caused by the rough model, the paper adopts grid subdivision technique to the facial model which can smooth the virtual human face, and make the virtual human face more lifelike in premise of keeping the topology of the original triangle,The experiments results show the algorithm proposed in this paper can build 3D facial model from human face video sequence with arbitrary pose, and obtain vivid virtual face with typical expressions which can be observed in different views.