Research On Feature Description Method Of Face Expression Recognition

With the rapid development of the computer technology and artificial intelligence technology, the application of human-computer interaction is becoming more and more widely. Facial expression receives a great attention as an important kind of body language. We can use the computer to extract the facial expression characteristic information, according to these information, the facial expression can be divided into seven categories, i.e., anger, disgust, fear, excitement, neutral, sadness and surprise. In facial expression recognition, facial expression characteristic description and classification method are most important steps. Although the facial expression recognition technology has a lot of achievements, but further study is needed to achieve large-scale application.This paper analyzes the facial expression recognition research status at home and abroad, which focuses on facial expression feature extraction method and classification method, and conducted a series of experiments. The main research work of this thesis is as follows:①Study the influence of feature fusion to facial expression recognition rate. This paper uses the methods of local binary pattern and gradient direction pattern to extract the facial expression features respectively and then fusion these two features. In JAFFE database and Cohn-Kanada database, the recognition rate is69.52%and75.71%respectively. Feature fusion can obtain more abundant image characteristics. Experiment results prove that feature fusion can improve facial expression recognition rate.②Study the effect of binarized statistical image features in facial expression recognition rate. This method obtains a set of filters by using the statistical features of natural images, then using these filters to extract facial expressions features. The recognition rate is69.52%and77.43%based on JAFFE database and Cohn-Kanada database respectively. The experimental results show the effectiveness of binarized statistical image features in facial expression recognition.③Study the effect of block division strategy in facial expression recognition rate. This paper using three different ways of blocking:3x3,3x5,15x15. From the experimental results, the facial expression recognition rate with block15x15is higher than other two methods of block. At the same time, the computer will running longer time.Based on JAFFE and Cohn-Kanada facial expression library, the experiment results show:Fusion feature can improve the facial expression recognition rate and using the method of binarized statistical image features to extract features can effectively improve the facial expression recognition rate. Compared with support vector machine and other traditional classification method, experiment results prove the effectiveness of sparse representation classification method in facial expression recognition.