Facial Expression Recognition Based On Deep Learning

The Internet of things(IOT)is in vogue today,face identification applications has been very mature,but the application of facial expression recognition is still blank,if facial expression recognition can be applied to IOT,to the computer gived feelings,this is the real era of IOT.Facial expression recognition is also an important part of the intelligent technology for human machine interaction,and is a active research topic in the fields of artificial intelligence,pattern recognition,etc.it has catch various disciplines researchers' attention in recent years,and the researchers has proposed many new methods.In this paper,the development of facial expression recognition at home and aborad in recent years to be a brief introduction,and a detailed analysis and summary for the mian facial expression recognition technology used in this system is made,such as feature extraction and facial expression classification.Finally,the potential problems in the process of Facial expression recognition are summarized and presente the future development direction of Facial expression recognition.This paper analyzes and summarizes some important methods of the feature extraction and classification in facial expression recognition.In this paper,puts forward some new identification methods by combining with the method of deep learning and the traditional identification methods.Finally,the experiment verified.The main works of this paper are as follows:Deep belief networks is one of the most representative model of deep learning methods.DBNs have the ability of unsupervised learning features,but do not have the ability to identify the classification.So,this paper presents a new facial expression recognition method by combine deep belief networks and Multilayer Perceptron.This method abandon some cumbersome steps,such as:face detection,image proprocessing and feature extraction,and so on.Firstly,deep belief networks are used to learn primitive facial expression features,and get a higher level of abstract features,After several pre-training network and fine-tune the depth of faith,and ultimately choose the best model,and the weights and bias of the model is used to initialize the hidden layer weights and bias of the traditional model of multi-layer perceptron(MLP).Finally,we use MLP model to achieve the classification of facial expression.Experimental results on the JAFFE database,the proposed method can obtain the best accuracy of 91.25% and 88.57% under dataSet database,for facial expression recognition,recognition accuracy obviously the highest compare to other used classification algorithms.Thus,this method is proposed for facial expression recognition can greatly improve performance.