An Improve Linear Discriminant Analysis Method Based on Regularization
Lihua Guo, Lianwen Jin · 2010
Since the Linear Discriminant Analysis (LDA) method has the ability to choose the discriminant low-dimension subspace from the high-dimension feature space, this method has been successfully applied in some research fields. This paper proposes an improved LDA (ILDA) method to overcome the multi-model problem of LDA. In our ILDA method, the between-class scatter matrix and within-class scatter matrix are regularized, and some rules are introduced to optimize the Eigen analysis of LDA using matrix trace judgment. Some experimental results show that ILDA method can preserve the ability to choose the discriminate low-dimension subspace, and overcome some multi-model problems.