Kernel-Based Dual-Space Method and Its Fast Solution Algorithm
Zhou Xiao · 2010
To overcome the drawbacks of the dual-space linear discriminant analysis method,a kernel-based dual-space discriminant analysis method,namely KDS-DA is proposed.A fast algorithm based on the inverse operator of bordered matrix is also proposed for solving the discriminant vectors of KDS-DA.The algorithm utilizes the fact that the inverse computation of higher order bordered matrix can be converted to the inverse computation of a lower order matrix.To solve the(r+1)-th discriminant vector in the principal subspace of the within-class scatter matrix,the computational results obtained in computing the r-th discriminant vector are fully used,which significantly reduces the computational cost.The experimental results on ORL and AR face databases demonstrate the effectiveness of the proposed methods.