Biometric Authentication: A Machine Learning Approach

4.2. Fisher's Linear Discriminant Analysis

The study of linear classifiers with supervised training has a long history, going back to that of Fisher's linear discriminant analysis (LDA) and perceptron. The development of perceptron has led to a vast literature on neural classifiers, especially the class of decision-based neural networks; thus, its discussion is deferred to Chapter 7. This section describes the basic Fisher classifier and highlights its fundamental difference from the support vector machine.

4.2.1. A Two-Class Classifier

Assume that one is given N sets of labeled input/output pairs {x_i, yi; i = 1,..., N} ∊ X × {+1, -1}—shown in Figure 4.1(a)—where X is the set of input data in ℜ^D and yi are the labels. The task is to compute the ...