Study On Facial Expression Recongtion

With the development of the society, all the nations around the world have set off a way of constructing high-speed rail. China has known a significant achievement in constructing high-speed rail in the past few years, which made a great contribution to the progress of human civilization and the development of the society. Therefore, the proper evaluation of the degree and the way to improve the abilities of dealing with emergency when the train is under the extreme operating conditions are very important. So, this paper explores the use of facial expressions information of passengers to evaluate the comfortable degree and analyze passengers'response behavior under the extreme operating conditions. It lays the foundation for the simulation experiment of passengers'ride environments under the extreme operating conditions which is based on National laboratory of Track Transportation. It provides first-hand feedback information for evaluating comfortable degree. It's benefit for dealing with emergency.Firstly, this paper explains the background and the meaning of research, it introduces facial expression of the current state of research, discusses the key steps of face detection, preprocessing, feature extraction, and classification. Secondly, the shortcomings of Gabor filters and down-sampling of traditional methods have been analyzed. Aiming at shortcomings, a new method is desgnined. To begin with, using Adaboost for detecting face and eyes in order to get sub-image. Next, a bank of 24 Log-Gabor filters is applied on images to extract features after the sub-image was preprocessed.Then the statistical sampling and Principal Component Analysis (PCA) are used successively in order to reduce dimensions. Last, the experiment has been done based on the JAFFE and ORL in order to valid confirmation of the methodOn this basis, SVM (Support Vector Machines) have been involved as classifiers. The features vectors have been input to SVM for training, and then classify the test samples after getting SVM classifier. Two groups of simulation experiment have been designed based on the JAFFE database, the experimental results show that the novel method is superior to traditional methods, which achieves higher efficiency and stronger robustness.