基于信息融合的面部表情识别

Facial Expression Recognition Based on Information Fusion

文章提出用支持向量机融合四种基于不同特征表示的面部表情识别方法进行面部表情识别,即几何表示、PCA人脸表示、ICA人脸表示和FLD人脸表示。在用FLD和ICA提取表情特征前先进行PCA,把训练样本的人脸图像向量投影到一个较低维的空间,以达到降维和去除相关性的目的。然后对每一种表情特征表示都用最小距离分类器进行初步分类,最后用支持向量机融合这些分类结果来进行面部表情的最终识别,实验证明本文提出的方案是有效的。

In this paper,we propose a multi-feature and multi-classifier fusion system based on SVM for facial expression recognition.First,we obtain local features based on the geometric relation and three global features based on three face representations：PCA,ICA,FLD representations from pre-processed face images.The outputs of face recognizers based on four face representations are input to SVM information fusion to get facial expression recognition.We train and test the proposed algorithm with images from the Yale face database and Japanese female facial expression database, and the results show algorithm is effective.