It has been shown that facial expression plays an important role incommunication, so automatical facial expression recognition is quite use-ful for industry and daily life in nowadays. Most of the existing facial ex-pression recognition methods are based on either only texture features oronly geometrical features. In this paper, we propose to improve the perfor-manceoffacialexpressionrecognitionbycombiningbothtypesoffeaturesusing fuzzy integral. The geometric features used are the displacements ofpositionsofActiveShapeModel(ASM)featurepointsontheface. Wefrstembed them in a lower dimensional manifold space, then use a modifedversion of Support Vector Machine (SVM) as the classifer. The texturefeatures are boosted Gabor features. Since the dimension of Gabor fea-tures is quite high, we use Adaboost to select the most important featuresand then use SVM to classify them for diferent emotions. Finally, wecombine these two methods using fuzzy integral. The experiment result-s show that our method signifcantly improves the performance of facialexpression recognition. |