Facial Expression Recognition Algorithm Based On Dynamic Changes Of Facial Landmarks

Expression is important for humans to convey their emotions,it can reveal people's emotions,intentions,personalities and other information.Therefore,facial expression recognition can be used in the fields of medical care,education,driving,and interrogation.Research on facial expression recognition is of great practical significance,and researchers have carried out a lot of research on facial expression recognition.Facial expression recognition can be divided into recognition of images or video.This paper focuses on the facial expression recognition in video.The facial expression in the video is a dynamic process.The facial landmarks of the face contain rich information during the process.And this paper proposes the concept of dynamically changed image of facial landmarks(abbreviated as DCIFL),which can effectively describe the changes of expression.Also,based on the DCIFL and deep learning,more comprehensive feature is constructed.In addition,a new loss function is proposed to solve the problem of misclassification of similar expression.Therefore,the main work of this paper is as follows:(1)The concept of the DCIFL and the method to construct the DCIFL are proposed.With the dynamically changed image of facial landmarks,the change of each region of the face at various moments can be intuitively reflected.(2)The algorithm for facial expression recognition combining the DCIFL and neural network is proposed,which can extract comprehensive features to describe changes of expression;the DCIFL can describe contour information of the face,while the convolutional neural network can extract the texture information,and the recurrent neural network can extract the timing information of sequence.With these advantages,accuracy is greatly improved.(3)A new loss function is designed.In order to solve the problem of misclassification of similar expression,this paper integrates the variance information into the loss function,which makes the similar variation easier to be distinguished,thus reducing the misclassification ratio of similar expressions and improving the overall recognition accuracy.Experiments are taken on the facial expression dataset of CK+,MMI and Oulu-CASIA.The results show that the DCIFL can effectively describe the changes of facial expressions.The loss function with additive variance information can alleviate the misjudgment of similar expressions,and the overall algorithm can effectively improve the final recognition accuracy.