The multiclass, multivariate discriminant analysis [42] is used to derive the features in this paper. The discriminant analysis is characterized as follows: one has two types of multivariate observations - the first, called training samples, are those whose class identity are known, and the second type, referred to as test samples, consists of observations for which class identity are unknown and which have to be assigned to one of the class. The discriminant analysis consists of two stages. The first stage, concerned solely with the training samples, is to find a representation of these observations so as to, in some sense, clearly separate the groups. The second stage is concerned with assigning the test samples to one of the specific class.