Facial expression recognition based orthogonal local fisher discriminant analysis

In recent years, feature extraction method make an achievement in pattern recognition. It extracts not only useful feature for classification, but also reduces the dimension of pattern sample. Linear discriminant analysis is an important method for image recognition, it achieve significant development both in theory and applications. Local fisher discriminant analysis redefines the between-class and with-class matrix, it can work well when with-class multimodality or outliers exist. Simultaneously, it can remove the limitation of tradition LDA which its embedding space dimension should be less than the number of classes. In this paper, we propose orthogonal local fisher discriminant analysis for facial expression recognition. Experiment on JAFFE database and Cohn-Kanade database show our method can get better performance than PCA, LDA, LFDA.