Research On Facial Expression Recognition Methods Based On Multi-features Fusion

As an important part of emotional research,facial expression recognition is a necessary condition for intelligent interaction between human and machines,it has important research significance and potential commercial value.Due to the diversity of human emotion expression and the complexity of facial expression itself,the single-category feature can not work well for facial expression recognition,for this reason,more and more researchers are beginning to study the directionof multiple features fusion.Thesis focuses on the research of facial expression recognition algorithm based on facial image illumination compensation denoising,texture feature fusion facial expression recognition and multi-feature fusion facial expression recognition:(1)Research on the multi-method combining algorithm for illumination compensation of face images.Due to the fact that changes of illumination will seriously affect the face image recognition,an improved illumination compensation algorithm is proposed to eliminate noise,such as changes in illumination.By using YaleB07 library simulation test,the effects of each algorithm are compared from the subjective(human eyes)and objective(characteristic point efficiency and running time,brightness,and information entropy),the result of test shows the effectiveness of the algorithm.(2)Research on the texture feature fusion algorithm for face expression recognition.Aiming at the disadvantages of the directionality local binary patterns(DLBP)to obtain image texture features in a single scale,an xor-asymmetric-directional local binary pattern(XOR-AR-DLBP)multi-scale and multi-directional fusion feature extraction algorithm is proposed.The preprocessed facial expression image is compensated by illumination.The key areas of face and eyebrow,eye and mouth are divided and normalized,and the contribution degree of key areas(CM)is calculated.The feature information of XOR-AR-DLBP histogram is extracted from face and key area,CM is used to cascade the feature information of key area,and the information is connected with the feature information of whole face image to get texture fusion feature.Texture fusion features are trained and identified by the support vector machine(SVM)classifier.The experimental results show that,compared with the traditional algorithm,the texture feature fusion algorithm based on the simulation of JAFFE library and CK library can effectively improve the recognition rate and real time of facial expression.(3)Research on the multi-feature fusion algorithm for face expression recognition.The multi feature fusion algorithm based on the weight selection of expression recognition rate is proposed to overcome the limitation of single feature expression recognition.Texture fusion features,geometric features and global features are trained and identified by SVM classifier,and three different results are obtained.The result is fused by recognition rate weight selection algorithm,and the recognition results of multi class fusion features are obtained.The experimental results show that,compared with the three kind of single feature expression recognition algorithm,the multi feature fusion algorithm improves the average recognition rate of facial expression.Compared with the existing multi feature fusion algorithm,the average recognition rate of multi feature fusion is improved by0.52%~2.27%.