Abstract: Imagebased facial animation is useful in practical applications because of its realistic texture and easy implementation. Facial animation driven by complex mesh can improve the sense of reality but reduce the efficiency of animation. In this paper, modelphoto matching was made easy because of CANDIDE model. Local optimization of the mesh after the sketchy matching not only improves the representative of the model but also reserves the matching efficiency. Mesh deformation based on thinplate splines interpolation instead of the traditional nodemoving can effectively improve the reality. The experimental results demonstrate that this method can generate vivid facial animation and the replacement of character becomes easier and faster.