Facial Expression Recognition Based On Compression Perception Algorithm

While the sensor is widely used, many companies have developed the human-computer interaction devices, such as Surface Tablet PC from Microsoft and Ipad devices from Apple company. But they just provide Touch Sense and Voice Control, rarely come to Computer Vision Perception. Which is not to deny that Computer Vision Perception is important to human-computer interaction, pattern recognition and automated image understanding. Especially the facial expression recognition technology allows a computer to read people in mind, and provide more and better services in the medical field, the security field, public services and distance education. So according of this opinion, the paper mainly completed the following work:(1) Real-time face detection, firstly this paper summarizes the knowledge-based face detection and statistic-based face detection, and then focuses on the basic theory of AdaBoost algorithm to detect the human face. At the same time, this paper summarizes the database used in domestic and foreign, the final expression training set consists of the Cohn-Kanade face database, CAS-PEAL face database, JAFFE database and some images found from Internet.(2) The facial feature extraction based on compressed sensing. This paper outlines the basic theory of compressed sensing, introduces and analyses compressed sensing signal in the measurement matrix, sparse representation and signal reconstruction; Finally, we use a very sparse matrix which satisfied Johnson-Lindenstrauss theorem to extract facial expression features. This approach take advantage of speed of compressed sensing theory, has laid a good foundation for the real-time facial expression recognition.(3) Expression classifier training. Firstly this paper introduces several classification methods such as the Support Vector Method, Decision Tree and Naive Bayes Theory, and proposed using Support Vector Method (SVM) to classify the face expression. Then we focus on the use of the LibSVM tools to train the expression classifier, and discuss the classification results of different kernel function.(4) Facial expression recognition. After training the classifier, this paper concludes two expression classification methods:expression classification based on the Decision Tree and expression classification based on voting method. We compare these two methods in speed and stability, and finally choose the expression classification based on voting method which has better stability.