Approximate LDA Technique for Dimensionality Reduction in the Small Sample Size Case
The regularized linear discriminant analysis (LDA) technique overcomes the small sample size (SSS) problem by adding a regularization parameter to the eigenvalues of within-class scatter matrix. However, it has some drawbacks. In this paper we address its drawbacks and propose an improvement. The proposed technique is experimented on several datasets and promising results have been obtained.
Journal of Pattern Recognition Research
Artificial Intelligence and Image Processing not elsewhere classified