基于Gabor、Fisher脸多特征提取及集成SVM的人脸表情识别

摘要

针对静态的灰度图像表情库,提出了基于多种脸部表情特征多级分类的表情识别算法.首先在选取的人脸特征点上作局部的Gabor小波变换.为了提高特征提取速度,利用改进的弹性图匹配算法来提取图像中的人脸有效区域,在提取的人脸区域中提取几何特征,并通过Fisher脸法提取统计特征,利用几何特征与建立的相应一级集成SVM来进行初次分类.最后利用Fisher特征与建立的相应二级集成SVM进行最终分类.通过在JAFFE与Cohn-Kanade表情库中实验,证明该方法与单个特征相比较,具有更高的表情识别率以及更强的鲁棒性.%Based on the static gray image expression database, this paper gave a recognition algorithm by using multiple facial expression features to construct multi-classifier.Aiming to improving speed of extracting features, features of expression that were extracted by local Gabor wavelet transformation on the selected facial landmark were used to constructing facial elastic templates.Extracted geometric features and Fisherfaces features on the facial effective area extracted by elastic templates.Primary integrated SVM should be constructed by combining with Geometric features; secondary integrated SVM should be constructed by combining with Fisherfaces features.Compared with the single features, the experimental results show that recognition rate and robustness are improved by experiments based on JAFFE and Cohn-Kanade.