Research On Facial Expression Recognition Based On SVM

In recent years, the study of facial expression recognition in new human-computer interaction system is taken more and more seriously. It has a wide application prospect in education, robot, medical and other industries, and plays an important role in disciplines of pattern recognition, computer vision, and artificial intelligence etc. especially in enhancing the intelligence and humanity of computer development model human-computer interaction; facial expression recognition has practical significance, economic benefit and social benefit.Taking facial expression recognition algorithm as research object, this paper deeply researches the facial expression feature extraction and dimensionality reduction, and facial expression classification. Taking advantage of the excellent properties in the image feature extraction of the Gabor wavelet transform and the advantage of support vector machine’s excellent properties in small sample problem and good classification ability, this article puts forward the new method of fusion of two-dimensional Gabor wavelet combined with PCA+FLD expression feature extraction with based on (RBF) function One-against-one voting strategy classification SVM facial expression classification algorithm.The main work and innovation of this paper are as follows:(1) The method of gray-level integration projection in facial area positioning. Firstly, image preprocessing and then using digital images of the two values on integral projection to determine expression regions, and geometry pre-processing finally normalized cut standard size expression image.(2) Using multiple frequencies and direction in the two-dimensional Gabor wavelet transform and fusion PCA+FLD method of feature reduction to transfer image in high dimension feature space to a lower dimensional feature space.(3) This paper using RBF (Radio Basis Function) radial basis function as the kernel function of SVM classifier. Then the SVM classifier is used to train facial expression sample and to recognize facial expression.This paper carries on three groups of experiments, two experiments, original image and the PCA+FLD method based on the SVM expression recognition, and the integration of two-dimensional Gabor wavelet transform, related with people, and the last experiment, PCA+FLD reduced SVM expression recognition nonrelated with people. Experiments show that the method of SVM expression recognition based on the proposed fusion two Gabor-dimensional wavelet transform and PCA+FLD is a better method in facial expression recognition.