| Facial expression of human is a main media for information transmission between communications of people. It includes much emotion information, and plays an important role in our lives. It is one of the most important manner of non-verbal communication and it is a supplementary of language communication. Facial expression recognition is the basis of emotion understanding, and premise of computer understand emotion of human. And it is also the basis of analyzing human emotions by using computers.A new approach using Gabor features and MutualBoost is proposed to recognize the facial expression. Since the high-dimensional Gabor feature vectors are quite redundant and the redundancy between Gabor features selected by AdaBoost is not considered, MutualBoost is introduced as a method of features selection and redundancy exclusion. Furthermore, nearest distance classifier is used for classification. This approach takes the advantages of the favorable ability of Gabor features in representing expression variability, the effective function of MutualBoost in feature selection and redundancy exclusion. Experiments with JAFFE (Japanese Female Facial Expression) database show that the redundancy of features selected by MutualBoost is lower than that of features selected by AdaBoost, and the recognition rate using MutualBoost to selecte features is higher than that using AdaBoost with the same condition. The experiments also show that the highest recognition rate is 97.62%, when select 180 features by MutualBoost, and it is higher than the highest recognition 97.14% of AdaBoost. And it also much higher than that of the approach using non features selection algorithm, it's highest recognition is 95.71%. So the advantage of features selected using MutualBoost over those learned by AdaBoost, specifically, Gabor features selected using MutualBoost are both non-redundant and achieve higher recognition accuracy. And the approach is practical in the area of Facial Expression Recognition. |
PDF Full Text Request