Geometric interpretation of Fisher's linear discriminant analysis through communication theory

Abstract

This paper provides a geometrical aspect of Fisher’s linear discriminant analysis (FLDA), which has been widely used owing to its simple formulation and low computational costs. Our approach is based on a new framework of pattern recognition that can be modelded by a communication of class information. This model is quite different from a commonly used framework of pattern recognition as a mapping from the set of patterns to the set of classes. In the new framework, patterns can be regarded as class information with redundant encoding. We show that the geometry of two class FLDA can be described via communication theory of noisy channel.

Publication
IAPR International Conference on Machine Vision Applications (MVA)

Related