An Optimal Set of Discriminant Vectors

Key Points

• Develop a feature extraction method for two-class pattern recognition that selects discriminant vectors directly optimized for class separability instead of data fitting.
• Derived a mathematical formulation selecting discriminant vectors corresponding to the largest discrimination values.
• Developed a recursive algorithm for computing sequential discriminant vectors.
• Evaluated the framework against classical eigenvector-based fitting techniques using data-dependent and theoretical benchmarks.
• The proposed approach prioritizes features that maximize separation between classes rather than fitting overall data variance.
• The recursive method successfully extracts optimal discriminant vectors across both theoretical and empirical test cases.