A Statistical Identification and Verification Method for Biometrics

Abstract

A biometrics system is to find out the identity of a person by measuring physical and physiological features which can distinguish the corresponding person from others. When applying the conventional machine learning methods to design a biometrics system, one first runs into the difficulty of collecting sufficient data for each person to be registered to the system. In addition, there can be almost infinite number of variations of non-registered data. Therefore, it is very difficult to analyze and predict the distributional properties of data that are essential for the system to process real data in practical applications. These difficulties require a new framework of identification and verification, which is appropriate and efficient for the special situations of biometrics systems. As a preliminary solution, the present paper proposes a simple but theoretically well-defined method based on the statistical test theory.

This work is supported by the Korea Research Foundation (2001-003-E00234)