Fall 2001 Update to CSU PCA Versus PCA+LDA Comparison


This short paper updates results presented in two previous publications, “ A Nonparametric Statistical Comparison of Principal Component and Linear Discriminant Subspaces for Face Recognition” presented at CVPR 2001 and “Parametric and Nonparametric Methods for the Statistical Evaluation of Human ID Algorithms” presented at the Workshop on the Empirical Evaluation of Computer Vision Algorithms held in conjunction with CVPR 2001. The update reflects changes in the measured performance of our PCA+LDA algorithm following refinements to the numerical precision used to determine the PCA+LDA subspace. The new results show improved performance of the PCA+LDA algorithm relative to PCA algorithm. Where as before, PCA+LDA was clearly inferior to PCA alone, PCA still appears to have a slight edge for this test, but the difference is no longer statistically significant as measured by the methodology laid out in the previous papers. This paper is not intended to be read alone, but instead after the papers cited above. Likewise, those who read the papers above should read this paper to get an improved picture of how the PCA+LDA algorithm performs.

6 Figures and Tables

Cite this paper

@inproceedings{Beveridge2001Fall2U, title={Fall 2001 Update to CSU PCA Versus PCA+LDA Comparison}, author={J. Ross Beveridge and Kai She}, year={2001} }