Orthonormal discriminant vectors of pattern recognition

Recently, concerning statistical feature extraction, Okada and Tomita presented an orthonormal discriminant vector method based on the Fisher criterion. The advantage of their method has been discussed experimentally. However, it has not been discussed theoretically in comparison with a well-known discriminant analysis. For a two-class problem, it is shown that the method is more powerful than discriminant analysis from the viewpoint of the discriminatory ability between classes. Numerical examples are presented to illustrate the results obtained in this paper.