Research Of Human Facial Expression Recognition

Facial Expression plays a very important role in information transmission when people communicate with each other. It succeeds in transmitting plenty of information people can not express by languages. Human Facial expression recognition is a process of feature extraction and classification, which makes the computer detect the expression state from given expression images and ascertain the subject's specific emotion in order to achieve smarter and more natural inter-action between human beings and computers. Generally, facial recognition system consists of mainly four parts, which are expression image acquisition, expression image preprocessing, expression feature extraction and classification. This paper presents a detailed description of the research on the key issues of expression feature extraction and classification, and the research work can be summarized as:(1) The significance and application future of research on the Facial Expression Recognition is generalized in this paper. The current domestic and overseas study situation is reviewed and some existing facial recognition methods are summarized.(2) Facial expression image is preprocessed. The Japanese JAFFE expression database is used as experimental data. Firstly, expression image sub-sampling image preprocessing is done to reduce the dimension, and the image of each pixel gray value belongs to a normalized [0,1] for feature extraction and classification. Finally, the expression image histogram is equalized. After equalization the details of the image get clearer, and the distribution of gray levels of histogram gets evener.(3) In the period of feature extraction PCA is adopted, and nearest neighbor classifier is used for classification, using Euclidean distance as the similarity measurement, and we've got a good result. Kernel nearest neighbor is also used for facial recognition and the experimental result show that it is better than nearest neighbor classifier.(4) Using Support Vector Machine(SVM) classification method and one-against-rest portfolio theory, we construct seven support vector machines, and use the combination of SVM for expression recognition. We improve the original SVM and according to the study sample estimate the nuclear parameters of each SVM adaptively without sharing the same nuclear parameters. Recognition rate improved from 94.761905% to 95.238095%.(5) Expression recognition method based on the nearest neighbor and expression recognition method based SVM are compared. Expression recognition method Based on the nearest neighbor is adopted when requesting real-time and rapid classification but not high correct classification. Expression recognition method based SVM is adopted when requesting high correct classification but not rapid classification or not caring speed.