Research On Facial Expression Generation Based On Candide-3Model

Non-linguistic information of facial expressions’ transfer occupies an importantposition in people’s daily exchanges. Along with the computer technology increasinglydevelops and the portable devices become widely popular, two-dimensional animation stilloccupies a large part of the entertainment market share, which leads the results thattwo-dimensional animation techniques have broad market prospects like thethree-dimensional animation. As a result, paper focuses the research on image generationof facial expressions based on the subject background of facial image expressiongeneration which faces the field of digital entertainment. Paper realizes the quick changeand deformation of animation characters by using Candide-3model. The main work of thispaper is shown as follows:(1)Select the locational key points in Candide-3model, and improve the method offacial matching similar ratio. Enhance the matching accuracy of the model and the realfacial image.(2)According to the movement of the face, divide facial movement area hierarchicallyin Candide-3model, pertinently and secondarily refine the different part of the regionshierarchically. Then generate the real and nature facial expression animation.(3)Develope the expression generation system in WPF platform, and verify thefeasibility of the theory.The innovation of this paper are as follows:(1)A model-to-image matching algorithm based on two coordinate axes proposed.Improve the Candide-3model, and increase the accuracy of the positioning by adding twopupil points. At the same time, combine the thought of FAPU in MPEG-4and existingsimilar ratio matching algorithm, and extend the similar ratio method from one coordinateaxis to two coordinate axes. Increase the matching accuracy and realize the quick changeand deformation of animation characters by using Candide-3model.(2) Divide the face into high frequency region and low frequency region. Stratify thegrid of Candide-3model correspondingly. According to the descriptions of AU in FacialAction Coding System(FACS), summarize the movement of human face throughquantitative and qualitative analysis. On this basis, divide the movement area of Candide-3 model and secondarily refine the different areas, which has increased the pertinence ofgrid’s refining and ensured the effect of human facial deformation.On the basis of the theories above, a single-image expression generating system isdesigned to achieved in WPF platform. Experiments show that the method is simple andsmart, has a high efficiency, and ensure the effort of human facial deformation.