基于表情活动单元的人脸模型自动适配

Automatic adapation of human face model based on countenance action unit

提出一种算法，使用变形模板估计眼睛和嘴巴的特征，利用这些特征用Ｃａｎｄｉｄｅ模型对人脸进行适配，它对人脸模型的全局适配没有旋转限制，而且表情活动单元的初值决定于模仿适配。算法在视频序列Ａｋｉｙｏ和Ｍｉｓｓ Ａｍｅｒｉｃａ（ＣＩＦ，１０ Ｈｚ）上实验，证明其正确性和有效性。

In the proposed algorithm eye and mouth features are firstly estimated using deformable templates. Then the face model Candide is adapted to these estimated features. The algorithm differs from previous works in the following aspects a) there is no restriction on the rotation for the global adaptation and b) initial values of action units are determined due to the mimic adaptation. The algorithm has been applied to the test sequences Akiyo and Miss America (CIF 10 Hz) with high accuracy and efficiency.