Facial Expression Recognition Based on Linear Discriminant Locality Preserving Analysis Algorithm

摘要

The traditional LDA algorithm is used to seek the optimal distinguishing vector spaces, but it only considers the global structure of samples. The LPP algorithm can preserve the local structure of data very well, however, its projection direction is likely to cause the class overlap when applied to the facial expression recognition which is one of the multiple-classification problems. To resolve the class overlap problem, this paper presents a new Linear Discriminant Locality Preserving Analysis (LDLPA) algorithm which takes full advantage of LDA and LPP for facial expression recognition. It utilizes the strong discriminative superiority of LDA to find the optimal distinguish vector space and joins the thinking of LPP to preserve the neighborhood relationship among the data points. By reconstructing the between-class scatter and minimizing within-class scatter of LDLPA objective function, it makes the sample points of different classes be away from each other and the sample points of same classes be close to each other to get the optimal discriminative features. Finally, the validity of LDLPA is verified by the experiments on JAFFE and Cohn-Kanade facial expression databases.