Research On Face Expression Recognition Based On Dynamic Sequence

Facial expression can transfer psychological status,mood changes and other information in communication.With the development of artificial intelligence,people have higher requirements for computers to understand human emotions and identify facial expressions.Expression recognition technology has become one of the current research hotspots,and it has a wide range of application prospects and important research value.In this thesis,dynamic sequence images are used as the object of study,and facial expression recognition is performed for six basic expressions.Firstly,the facial expression images in the dynamic sequence are obtained by face detection.The Haar-like feature is used to represent the human face,and a strong cascade classifier is constructed by AdaBoost method to detect the face.All face images are then geometric normalized by bilinear interpolation method.Secondly,in the feature extraction process,in order to obtain the dynamic information changes better and improve the recognition rate of facial expression,an expression feature extraction method based on Active Appearance Model and Gaussian pyramid Lucas-Kanada optical flow method is proposed.Neutral expression model is constructed by Active Appearance Model to get the expression feature points of the initial neutral frame.The feature points are tracked by the Gaussian pyramid Lucas-Kanada optical flow method in the dynamic sequence to obtain the expression dynamic change information.Facial expression features extraction is realized through the combination of two methods.The Lucas-Kanade optical flow method compensates the shortcomings of Active Appearance Model in the localization of severe expression changes.Gaussian pyramid also solves the problem that feature points shift out of motion window due to the high motion velocity of objects.Finally,the K-nearest neighbor classification method and the support vector machine are used in expression classification.The difference between the four methods is compared when using the K nearest neighbor classification method,and the influence of the parameters of SVM and the number of training samples is also discussed.With the experiments in Cohn-Kanade+ facial expression database,the result shows that the expression recognition method based on the Active Appearance Model and the Gaussian pyramid Lucas-Kanada optical flow method can improve the accuracy of facial expression recognition.