Research On Facial Expression Recognition Method

The recognition of facial expression have a great value in social psychology and related disciplines,and now has become a research highlight.From the point of emotional understanding,the smiling face has a distinct particularity,which reflects the psychological state of the object directly,meanwhile the technology of smiling face recognition is applied in many kinds of software.In order to improve the accuracy of smiling face recognition,this thesis studies the smiley face recognition technology.The research process of smiling face recognition is divided into three stages,and they are facial detection,feature extraction and recognition.First,the Viola Jones algorithm is used to detect the region of face,which calculates the black and white pixel value of the picture through the Haar feature.Furthermore,it selects the pixel value by using the Adaboost classifier,thus extracting the facial region and the mouth region with the expression information.Second,the mouth angle coordinate method is proposed to extract the feature of the mouth region.The corner detection of mouth is implemented by the Shi&Tomasi corner detection algorithm at first.Then the least square approach is used to fit the corner points.The average value of two endpoint coordinates of the curve is calculated to represent the eigenvalue of the picture.In order to verify the validity and feasibility of the mouth angle coordinate method,this thesis compares the three feature extraction methods of Multiscale Block Local Binary Patterns(MB-LBP),Gabor and Histogram of Oriented Gradient(HOG)with the mouth angle coordinate method.Finally,Support Vector Machine(SVM)is applied to classify and identify the characteristics of different methods.In order to further improve the accuracy,the Simple Multiple kernel learning(Simple MKL)method based on SVM is used to fuse the four features,and the classification results of the fusion features are analyzed.The experimental results show that the accuracy of the method to mouth corner coordinate is 92.6%,at the same time,based on the fusion feature of the mouth corner coordinate method,the overall accuracy of the smiley face recognition is further increased to 96.33%.However,the fusion feature takes more time than mouth angle coordinate method.As a consequence,in the system with higher real-time requirements,it can be identified by the feature extracted from the corner of the mouth.If it is in the system with higher recognition accuracy,the fusion feature based on the corner of the mouth can be used to identify the smiling face.