Research On Facial Expression Recognition Based On Feature Fusion And Sparse Representation

Facial expression recognition technology is a research hotspot in the field of artificial intelligence and pattern recognition, it is also a challenging subject, which involves human-computer interaction, computer vision, image processing and other fields, and has broad development space. But because of facial features susceptible to light, noise and other interference, resulting in the recognition effect is not ideal.Based on the complexity of facial features, the purpose of this paper is to improve the expression recognition rate. Specific work is as follows:1. The technical methods of facial expression recognition are systematically summarized, among which the feature extraction method and the classifier design are studied. This paper mainly studied three methods of facial feature extraction:PCA,HOG and ULBP, and discussed the problem of energy value in PCA method. And the facial expression recognition experiments were carried out on PCA, HOG and ULBP.This paper introduces several commonly used expression classification methods, and the sparse representation classification method is studied emphatically. This paper discusses several sparse solution methods and the construction of super-complete dictionary, and introduces the principle and process of sparse representation classification algorithm in detail.2. Because different expression features are complementary in facial expression recognition, the feature fusion method is used to fuse different features. This paper focuses on the feature fusion method based on CCA and its application in expression fusion. Through the transformation function will feature from high-dimensional mapping to low-dimensional space, to reduce the dimension of the purpose. And the validity of the CCA method is verified by comparing with the expression of the single feature.3. Aiming at the multi-class classification problem of facial expression recognition, this paper mainly studies the classification method based on sparse representation. The feature extraction method based on CCA feature fusion isexperimentally verified by sparse representation classifier. The PCA and HOG characteristics, ULBP and HOG characteristics are fused based on CCA and the expression classification experiments are carried out. And compared with the expression recognition rate of PCA, HOG and ULBP feature extraction methods, and further validated the effectiveness of CCA method; Compare the expression recognition rate of different sparseness to find the most suitable sparseness.Experiments were carried out on different enery values in the PCA method to find the best energy values. By comparing with the expression recognition rate under SVM classification method, the superiority of sparse representation classification method is verified.