An efficient algorithm of solving the optimal discriminant vectors

摘要

In view of the limitations of traditional uncorrelated Linear Discriminant Analysis (uLDA) of failure with singular within-scatter matrix and computationally expensive in solving the optimal discriminant vectors for a large and high-dimension dataset, an equivalent uLDA to Linear Discriminant Analysis (IDA) and a algorithm of uLDA based on generalized singular value decomposition is proposed to simply the computation and get over the singularity problem. The classification experimental results of four image and text datasets demonstrate the superiority of our algorithmover other traditional algorithms.