中文摘要：

传统的基于纹理特征的表情识别采用单一纹理特征构建单核支持向量机（SVM）进行表情特征分类,势必会造成表情特征信息的丢失,影响识别率;然而过多的特征又会带来冗余,产生过拟合现象,降低识别率。针对传统方法的不足,提出了基于多核学习特征融合的人脸表情识别方法,即提取图像的Gabor纹理特征、灰度直方图特征、LBP纹理特征三种特征并进行主成分分析（PCA）降维,在多核支持向量机训练中利用基于核函数组合的特征融合模型,寻找一组最优的特征组合系数,构建基于特征融合模型的核函数,进行表情的分类。该方法能更大限度利用表情图像中的有用特征,还能避免无关特征和冗余特征带来的过拟合现象。通过在学生听课表情表情库中的实验结果表明,方法的识别率为88%,好于传统方法 80%的识别率。

英文摘要：

The traditional facial expression recognition based on texture characteristics uses single texture feature to build single-core Support Vector Machine（ SVM） to classify expressions. It will inevitably lead to loss of facial expression characteristic information and negative effect on recognition rate,but too many features will bring redundant,over-fitting and low recognition rate. For the shortcomings of the traditional methods,a facial expression recognition method based on multifeature fusion by multi-kernel SVM was presented,which extracted three features such as Gabor texture feature,histogram feature and LBP texture feature,then reduced the dimension with Principal Component Analysis（ PCA）,used feature fusion model based on combination of Radial Basis Function（ RBF） in multi-core SVM training,found an optimal set of features combination coefficients to build RBF based on features fusion model,classified expressions. Features fusion can not only make better use of useful features than single feature,but also avoid over-fitting resulted from irrelevant and redundant features. The results on students＇ class expression database show that the proposed method with recognition rate of 88% is better than traditional method with recognition rate of 80%.