Research On The Accuracy And Robustness Of 3D Facial Expression Recognition

Nowadays,the influences of gender,age and race are always neglected in the research of 3D facial expression recognition,which will reduce the accuracy and robustness of recognition.In general,there are some distinctions between male and female faces.For example,women have soft face,big eyes and thick lips,whereas men with rigid face,large jaw and coarse eyebrows.In addition,the differences in the character of men and women also lead to different facial expressions and women usually react to emotional stimuli more intense.With the growth of age,there will be many age features,such as wrinkles,skin pigmentation and so on.Gender and age features are both facial features,which can interfere with the accuracy of 3D facial expression recognition.Therefore,the influence of gender and facial wrinkle feature on facial expression recognition is studied in this paper.Major works and innovations of this paper are listed as follows:(1)Firstly,the 3D facial features extraction based on radial curves and the 2D conformal mapping features extraction based on local features are introduced.Then the corresponding optimal feature selection algorithm is applied to obtain the facial expression feature set with highest distinction.Secondly,the AdaBoost(adaptive boosting)and SVM(support vector machine)is used respectively to design the classifier,which could achieve automatic recognition of 3D facial expression.Finally,the comparative experiments of facial expression recognition through the different methods of features extraction and features classification are tested.(2)Extracting the gender features in 3D faces by the method of circle curves,and reducing the dimension of the features through PCA(Principal Component Analysis),then recognizing the gender of 3D faces automatically.Regarding gender as the prior knowledge of facial expression recognition,so the gender differences of 3D facial expression in training set will be reduced,and the accuracy and robustness of facial expression recognition will be enhanced.(3)A new wrinkle detection and elimination based on adaptive fractional differential is proposed for removing age features.First,the 3D facial expression image is mapped into 2D plane by conformal mapping.Next,constructing an autocorrelation differential mask region and improving the fractional differential mask operator through image gradient,entropy and contrast.So the facial wrinkles of the face image can be enhanced adaptively.Last,obtain curvilinear shapes of wrinkles with Gabor filter and eliminate them automatically.The robustness of extracting facial expression features is strengthened by using the 2D conform mapping face image,and the accuracy of the expression recognition is further improved.(4)Before identifying the facial expression,the gender of the 3D face image in BU-3DFE database has been recognized and facial wrinkles have been removed.Then,the effects of gender and age features on the accuracy and robustness of facial expression recognition are analyzed.Experimental results show that the accuracy and robustness of 3D facial expression recognition will be improved under the consideration of gender or age features.And the whole recognition rate could reach 86.43%by considering both of two cases.