Chapter 12Classification Techniques for Biometrics 1

12.1. Introduction

Automatic learning is a technique that allows machines to learn from examples or experiences. When no mathematical models exist for resolving certain problems, such as pattern recognition, a natural approach inspired from human learning was introduced to overcome the limitations of traditional programming techniques. The concept includes a family of algorithms, whose functionality is to predict the class of an object by the variables that characterize it, called features. The prediction is performed by a decision function built after a learning phase that involves a set of objects considered as training examples defined by the same features, and so called the training set. Learning is said to be supervised or simply called classification when the class memberships of the examples are known a priori; otherwise, it is called non-supervised learning or “clustering”. The authentication process in biometrics is an example of supervised learning, since it uses a training set composed of features extracted from persons known as authentic or imposters. After the training phase, a decision function is built to categorize new persons in two classes labeled respectively as “authentic” and “impostor”. Non-supervised learning, consists of grouping together given objects into clusters, by considering the similarity criteria, from which we cite the Euclidean distance as an example. The history of automatic learning dates ...

Get Signal and Image Processing for Biometrics now with the O’Reilly learning platform.

O’Reilly members experience live online training, plus books, videos, and digital content from nearly 200 publishers.