Research Of Facial Expression Recognition Based On Multi-feature Fusion

With the development of Computer Science Technology and the rise of Artificial Intelligence research,Facial Expression Recognition(FER)has become an important part of artificial psychological theory and artificial emotion research,it is also the main research object in Man-Machine Interaction technology.Based on a large number of data and literature,we studied each segment and the main parts of the Facial Expression Recognition System,aiming to solve practical problems and achieve a robust and efficient real-time expression recognition system.Firstly,this paper expounds the research background of facial expression recognition and the research status at home and abroad.After the study and implementation of the facial recognition system,a real-time facial expression recognition system was developed based on the aforesaid expression recognition theories.The main contents of our research are as follows:1)We studied the Haar-Like feature based on integral graphs combined with the cascaded AdaBoost classifier method to achieve real-time face detection.Face detection is the basis of expression recognition system,experimental results showed that this method can reach a good performance for real-time face detection in dynamic scenes.Detection of face can be accurately performed in the presence of face deflection and similar interference.2)Constrained Local Model(CLM)was used to complete the positioning of key facial feature points,which reduces the computational complexity.The location of facial feature points is a precondition for extracting geometric features and local regions.Compared with the current mainstream feature point localization algorithm-Active Appearance Model(AAM),CLM models have fewer features to detect feature points,which ensures the real-time performance of the system.3)A method combining geometric feature and Local Binary Pattern(LBP)texture feature of key region was studied for extracting facial features,which improved the efficiency of expression recognition system.After extracting the facial expression feature,the local linear embedding(LLE)method was used to reduce the dimensionality of the feature and further integrated it.Finally,the support vector machine(SVM)was used to implement feature classification.Experiments were conducted on the JAFFE dataset with seven expressions and the Yale dataset containing four expressions manually selected.Compared our method with other different methods respectively,the accuracy rate has advantages.4)A real-time expression recognition system was developed to verify the performance of the proposed algorithm.Based on the theoretical research and verification results of the above algorithms,using Visual Studio 2010(VS2010)as a development platform,a real-time facial expression recognition system was developed under the support of OpenCV 3.0 computer vision library.This system can realize the dynamic recognition of seven basic facial expressions.