Abstract: The method of facial expression recognition based on static gray image which is simple and facile can get very high recognition ratio in person-dependent expression recognition，but the result of this method is failed in person-independent expression recognition because it is easy to be affected by some disturbances as complexion and illumination.The motion features extracted from dynamic expression sequence can effectively reflect the movement of expression and obtain better result in person-independent expression recognition.This paper expounds the researches of extracting two kinds of dynamic sequence features by optical flow and gray difference between frames and each kind of feature is combined with Support Vector Machines（SVM） and Hide Markov Models（HMM） in person-independent expression recognition.The characteristics and performances of the two features are also analyzed and the results show the validity of dynamic sequence features in facial expression recognition.

摘要： 基于静态灰度图特征识别表情的方法简单、快捷，在进行特定人表情识别时可以取得很好的识别结果，但在进行非特定人表情识别时却容易受到肤色、光照等因素的影响，识别效果较差。通过动态序列提取的运动特征能有效地反映表情运动的形变过程，用于非特定人表情识别时可以取得较好的识别结果。研究了通过光流和帧间灰度差两类方法提取表情序列动态特征，再与支持向量机（SVM）和隐马尔柯夫模型（HMM）两种分类器组合，进行非特定人表情识别，并分析比较了两类方法的特点与优劣，说明了利用运动特征识别人脸表情的有效性。