Sparsity reconstruction error-based discriminant analysis dimensionality reduction algorithm

Abstract

The inter-class and intra-class information of existing discriminant analysis method is more sensitive to external disturbance such as the imperfection and occlusion. Aiming at this problem, from the view of the local sparsity representation, a kind of sparsity Reconstruction Error-based Discriminant Analysis dimensionality reduction algorithm is proposed. The algorithm firstly applies the sparsity representation to complete the intra-class local sparsity reconstruction, and then applies the average value of the non-intra-class sample to complete the inter-class local sparsity reconstruction of all the samples, finally keeps the inter-class and intra-class sparsity reconstruction ratio during the dimensionality reduction process. The algorithm improves the sparsity representation computation efficiency and discriminant analysis performance. The experimental results offace data set ofARandUMIST face database verify the effectiveness of the proposed algorithm.