【作者】 胡同森； 刘玉彪； 田贤忠； 龚婷；

【Author】 HU Tong-sen, LIU Yu-biao, TIAN Xian-zhong, GONG Ting School of Information Engineering , Zhejiang University of Technology, Hangzhou 310032, China

【机构】 浙江工业大学信息工程学院；

【摘要】 本文提出了一种PCA在表情识别方面的新方法——类内PCA。首先,将所有的训练样本按表情类别分好;再对分好类的训练样本进行Gabor滤波,得到每一类表情的Gabor特征;然后对滤波后的表情样本按类别进行PCA特征选择并得到每类表情的最主要特征,构造每类表情的特征空间,我们把构造这些特征空间的特征向量称为特征表情;最后在特征表情的基础上进行分类识别。本方法在日本女性表情数据库JAFFE上的实验结果表明,与传统的PCA方法相比,该方法的识别率有较大的提高。更多还原

【Abstract】 In this paper, a new PCA method is proposed for facial expression. We call it within-class PCA. Firstly, all the train samples are classified by facial expression class. Secondly, we use Gabor filter to filter these samples and get the Gabor features of every type of expression. Thirdly, the principal features of filtered samples are selected by the PCA according to the different class and we use these features to construct the feature space for every class of facial expression. These feature vectors of the feature spaces were called feature expression. Finally, the test samples, on the basis of the feature expression of the different type of expression, are classified. Experiments on JAFFE database show that the method is more accurate than the conventional PCA.更多还原