The Research On Facial Expression Recognition Method

Facial Expression Recognition is one of the important branches of biometrics and it is alsoone of the most active fields of computer vision and pattern recognition. Among all thecharacteristics of human, the characteristics of facial expression are the most direct meanswhich are friendly and convenient.In this paper, we studied feature extraction and pattern classification. And a method offacial expression recognition based on Gabor wavelet transform and support vector machine isproposed.To overcome the high-dimension and complex computation problem of Gabor features, afacial expression recognition algorithm based on Gabor wavelet transforms with2D-PCA&2D-ICA&2D-LDA algorithm is proposed. For the method can effectively usevaluable feature vectors, it could improve the recognition rate. When using 2D-LDA algorithmin feature dimention reduction, in order to reduce the computation, we adopt the image blockmethod in the algorithm framework. And we analyse the effect of different block methods tothe recognition rate, choosing the best image block method in the algorithm. Throughcombining the feature dimention reduction algorithms with the linear SVM in the test, wecompare the results and make the conclusion that Gabor features with 2D-LDA algorithm asthe feature reduction method can achieve the best recognition rate.About Support Vector Machine (SVM) classifier, we use a convex combination of linearkernel and polynomial kernel with good global effect and radial basic function kernel withgood local effect in the classification of facial expressions. Therefore the SVM classifier withRBF kernels has better classified effect than linear and polynomial kernel classifiers.In the end, Gabor features with 2D-LDA algorithm and SVM classifier with RBF kernelare testified to be the best classification method in JAFEE human facial expression imagedatabase. The experimental results show the validity of the proposed algorithm framework.