New Concept for Discriminator Design: From Classifier to Discriminator

Abstract

This paper introduces a new concept of designing a discriminant analysis method (discriminator), which starts from a local mean based nearest neighbor (LM-NN) classifier and uses its decision rule to direct the design of a discriminator. The derived discriminator, called local mean based nearest neighbor discriminator (LM-NND), matches the LM-NN classifier optimally in theory. The proposed LM-NND method is evaluated using the CENPARMI handwritten numeral database, the ETH80 object category database and the PolyU Palmprint database. The experimental results demonstrate the effectiveness of LM-NND and the LM-NN classifier based pattern recognition system.