Research On Facial Expression Recognition Based On Deep Neural Networks

With the development of artificial intelligence,the application of emotion recognition scene more and more widely,the typical advertising effectiveness evaluation,product evaluation,video analysis,medical rehabilitation,safe driving and emotional robot.At present,emotional recognition in t he field of human-computer interaction particularly fast,especially in the safe driving,emotional robot applications,the machine better understanding of people,more intelligent and humane human service is the fundamental of the recent artificial intelligence revolution.The machine gradually learn enough emotional cognitive ability,you can in the human-computer interaction in the user experience a series of upgrades,and ultimately,the machine can eventually like ordinary people into human life.Emotional recognition can be broadly expressed through facial expressions,phonetic intonation,or EEG capture.Currently the most mature technology is widely used is the expression recognition technology,which is based on computer vision algorithms to identify facial expression movements and inferred emotions and other basic emotions.Automatic facial expression recognition(FER)remains a challenging and interesting problem in computer vision because of differences in the level of expression of different people.Although efforts have been made to develop various methods for FER,the existing methods lack universality when applied to pictures that are not labeled or captured in the natural environment.Most existing methods are based on manual features(e.g.,HOG,LBP,and Gabor characterization operators)and then combined with a classifier(SVM)where the hyperparameters of the classifier are optimized to be given in a small set of individual databases or similar databases Best recognition accuracy.Different characterization operators have different ability to characterize the expression images under different backgrounds.The most suitable feature description operator must be found for a specific background image,which greatly increases the complexity of the work.And depth learning can automatically learn facial features,and belongs to the end-to-end model,that feature learning and classification in a model to complete.In this paper,a deep neural network architecture is proposed based on Google's incepti on structure to solve the problem of finding different character description operators in different background images and to simplify the model so that it can be successfully applied to the mobile terminal.Specifically,our network consists of two convolu tion layers,each layer is followed by the largest pool,followed by three inception layer.The network is a single component architecture that takes the registered facial image as input and categorizes it into any of six basic expressions or neutral expre ssions.In this paper,seven publicly available facial expression databases were comprehensively tested,Multi PIE,MMI,CK+,DISFA,FERA,SFEW and FER2013.The results show that the generalization ability of the method based on depth learning is better than that of the traditional one.In addition,the generalization ability of the method based on the traditional learning method is better than that of the current mainstream learning method.Models such as VGG,Google Net,Res Net and other models in the expression recognition task made a comparison further illustrates the structure based on the inception to ensure the accuracy of expression recognition under the premise of the model can be as simple as possible to reduce the size.