A novel approach to the analysis of feature spaces in statistical pattern recognition is described. This approach starts with linear dimensionality reduction, followed by the computation of selected sections through and ...