Supervised Classification of Whole Populations of Objects

J. Cutrona and N. Bonnet (France)

Keywords

Supervised Classification; Population; Principal Component Analysis; ParzenRosenblatt method; KullbackLeibler divergence

Abstract

In this paper we present a supervised classification method. Contrary to classical approaches dealing with the categorization of a unique object thanks to a training set, we propose here to cope with the classification of whole populations of objects. The scheme begins with a dimensionality reduction of the training set performed by Principal Component Analysis. The known and unknown classes are then projected in the reduced space. Next probability density function (pdf) estimation is performed for each class, known or unknown. Finally, a dissimilarity measure is computed between the pdf of the training sets and the unknown class. Here we chose the Kullback Leibler divergence as an example of dissimilarity measure.

Important Links:



Go Back