基于2D几何与纹理信息分离模型的表情分析与合成

Facial Expression Image Synthesis by Using 2D Model with Uncoupled Geometric and Textural Information

本文提出一种用于特定人逼真表情图像合成的新方法该方法利用二维网格描述脸部呈现不同表情时的几何变化，并使用主分量分析方法将该网格序列用参数化模型表示，从而极大地压缩了数据在重构表情图像时利用参考帧提供的场景纹理信息，结合网格参数模型提供的几何信息，使用图像卷曲与插值技术，生成逼真表情图像．目前使用该方法对人脸下半部表情图像分析及合成进行了初步的实验。

A novel image synthesis method of human face with various expressions is presented. In this method the geometrical and textural information implied in expressions are uncoupled. A 2D grid is used to extract the geometric variation of human face with different expressions. A Principal Component Analysis (PCA) is used so that the grid Sequences can be expressed by a few eigengrids. Therefore a parametric model can be developed to control the expressions effectively. During the image synthesis process, image warping and morphing techniques are employed to combine the textural information, supplied by a few reference image frames with the geometrical information, represented in the grid model, so that vivid facial images with different expressions can be preduced. The method presented in this paper has been investigated with some preliminary experiments. Its effectiveness has been proved.