Research And Implementation Of Deep-learning-based Static Facial Expression Recognition System

Facial expression recognition is to identify the state of facial expression and is an important research area in the current artificial intelligence research.Facial expression plays an important role in people's daily communication activities.In recent years,the degree of automation in life is getting higher and human-computer interaction is increasingly valued.If computers can better recognize the expression state of people,then the experience of human-computer interaction will be greatly improved.Not only that,face recognition can also be widely used in medical,education,driving,games,etc.,with important scientific and practical value.Researches on facial expression recognition is generally divided into two aspects,the one is to recognize facial expressions in static pictures,and the other is to recognize facial expressions in dynamic videos.Among them,identification of facial expression in pictures is the main content of this thesis.Deep learning is currently the hottest method in the field of artificial intelligence.With the dramatic increase in the performance of computer hardwares and the explosion of data volume in recent years,deep learning has achieved remarkable results in areas such as speech recognition,image recognition,and natural language processing.Various deep-learning models have also emerged and have achieved breakthrough success in various fields.How to use deep learning theories to improve the accuracy of facial expression recognition is a frontier research direction of artificial intelligence.In view of the above situation,this thesis studies how to apply deep learning methods on static facial expression recognition to improve its accuracy.Based on studying deep learning theorise and various models,we propose a deep-CNN-based ensemble classifier model using bagging framework.This model can commendably maintain the accuracy and stability of facial expression recognition.Compared with existing models,this model greatly reduces the complexity of the ensemble classifier,and reduces the cost of time and computer hardware resources in training and testing.Furthermore,in order to make its identification accuracy more stable in practical applications,this thesis trains the model with movie frames.Compared with the commonly used face expression database,training data set made from movie frames has more face angles,and the state and environment of the faces are closer to the real situation.Based on this model,this thesis designs and implements a static facial expression recognition system using B/S architecture.The system has two recognition modes: taking photo to recognition facial expression and uploading multiple local images to recognition facial expressions.In this thesis,the operation process and expression recognition accuracy of the system are tested respectively.For the test of recognition accuracy,two test data sets of the movie frames and real photos are used.Experiments have proved that the static facial expression recognition system in this thesis can maintain a good recognition accuracy in real environment.