Multivariate Statistical Classifiers for Extracting Discriminant
Information in Limited Sample Size Problems

In classification problems, when the number of examples per class is
less than or comparable to the dimension of the feature space, the
performance of statistical pattern recognition techniques tends to
deteriorate. This problem, called the ‘limited sample size problem’, is
indeed quite common nowadays, especially in image recognition
applications. In this talk, I will present a couple of ideas of using
multivariate statistical classifiers to identify and analyse the most
discriminating hyper-planes separating two populations. The goal is to
analyse all the image features simultaneously rather than segmented
versions of the data separately, feature-by-feature, or distinct models
for texture and shape information. To demonstrate the performance of
these statistical pattern recognition approaches I show some
experimental results on medical data composed of 3D magnetic resonance
images and on frontal 2D face images.