Abstract: By means of locating key points by Active Appearance Model（AAM） and describing each key point by Scale-Invariant Feature Transform（SIFT）, a new method for facial expression feature extraction is proposed. The gradient direction histogram around these points is used to describe features of expression. Meanwhile, this paper puts the feature points into group and gives them different weights according to their contributions to the expression recognition, and uses the two-level Support Vector Machine（SVM） to classify the category of the fused and weighted features. The experimental results of standard expression databases and multi-pose expression database show that the proposed method not only improves the recognition rates of the frontal face, but also has better robustness for the expression of the non frontal face under some deflection angles.

Key words: active appearance model, Scale Invariant Feature Transform（SIFT）, gradient direction histogram, feature grouping, multi-pose

摘要： 采用AAM定位特征点、尺度不变特征变换（SIFT）描述特征的方式提出一种基于AAM-SIFT的表情特征提取方法。该方法用特征点周围区域梯度方向直方图描述表情特征；同时根据不同子区域对表情的贡献不同，将特征点分组并赋予不同权重，并用两级支持向量机（SVM）对融合的加权特征进行分类识别。在标准表情库和多姿态表情库上的验证结果表明，该方法能有效提高正面人脸表情的识别率，对一定偏转角度的非正面人脸表情也保持较好的鲁棒性。

关键词: 主动外观模型, 尺度不变特征变换, 梯度方向直方图, 特征分组, 多姿态