Abstract: Facial expression feature extraction plays an important role in facial expression recognition. The expression feature extracted by existing methods is the combination of individual facial feature and expression feature. Facial recognition is based on different individual facial feature, but facial expression recognition needs to find out the differences of different expressions. What is more important is individual difference will influence the facial expression recognition, and obstruct the expression reorganization rate. In an optimal situation, the related individual facial feature can be separated during the process of facial expression recognition. This paper presents a method that can eliminate interference of facial features when recognizing the facial expression. Firstly, a three order tensor will be built. Secondly, it uses the tensor analysis method to decompose the face feature and the expression feature into the person subspace and the expression subspace respectively. This method can ensure that parameters of expression and face are not related. The evaluation experiment on JAFFE proves the validity of the method.

Key words: expression feature extraction, expression recognition, affective sate recognition, tensor analysis

摘要： 表情识别的性能依赖于所提取表情特征的有效性，现有方法提取的表情基本上是人脸与表情的融合体，然而不同个体的人脸差异是表情识别的主要干扰因素。在表情识别时，理想情况是将个体相关的人脸特征和与个体无关的表情特征相分离。针对此问题，在三维空间建立人脸张量；然后用张量分析的方法将人脸特征与表情特征进行分离，使获取的表情参数与人脸无关。从而排除不同个体的人脸差异对表情识别的干扰。最后，在JAFFE表情数据库上验证了该方法的有效性。

关键词: 表情特征提取, 表情识别, 情感识别, 张量分析