Fast multistage Gaussian maximum likelihood classifier

The authors propose a multistage classification algorithm based on the Gaussian maximum-likelihood (ML) procedure. A discriminant function is calculated for classes at each stage, and the classes whose discriminant function values are less than a threshold are truncated. The algorithm reduces processing time substantially without losing any significant accuracy. The computing time can be reduced by a factor of 3-7 using the proposed multistage classifiers when the Gaussian ML classifier is to be used. Therefore, after features which depend on an accuracy requirement have been selected, the processing time can be reduced substantially without losing any significant accuracy by employing the multistage classifiers.