Modified Class-Incremental Generalized Discriminant Analysis
Yunhui He · 2009
In this paper, we propose an efficient method for resolving the optimal discriminant vectors of generalized discriminant analysis (GDA) and point out the drawback of high computational complexity in the traditional class-incremental GDA [W. Zheng, "Class-Incremental Generalized Discriminant Analysis", Neural Computation 18, 979-1006 (2006)]. Because there is no need to compute the mean of classes and the mean of total samples in the proposed method as needed in the traditional class-incremental GDA, the computational complexity is reduced greatly. The theoretical justifications of the proposed batch GDA and the class-incremental GDA are presented in this paper.