Matching Score Level Fusion

Matching Score Level Fusion

David Zhang (Hong Kong Polytechnic University, Hong Kong), Fengxi Song (New Star Research Institute Of Applied Technology, China), Yong Xu (Harbin Institute of Technology, China) and Zhizhen Liang (Shanghai Jiao Tong University, China)
DOI: 10.4018/978-1-60566-200-8.ch014
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

With this chapter we aims at describing several basic aspects of matching score level fusion. Section 14.1 provides a description of basic characteristics of matching score fusion in the form of introduction. Section 14.2 shows a number of matching score fusion rules. Section 14.3 surveys several typical normalization procedures of raw matching scores. Section 14.4 gives an example of matching score level fusion method. Finally, Section 14.5 provides several brief comments on matching score fusion.
Chapter Preview
Top

Introduction

Matching score level fusion is the most commonly used biometric information fusion strategy because matching scores are easily available and because they retain sufficient information to distinguish genuine matching from impostor matching. Generally, a multi-biometric system based on the matching score level fusion works as follows: each subsystem of the multi-biometric system exploits one biometric trait to produce a matching score. Then these matching scores are normalized and integrated to obtain the final matching score or final decision for personal authentication. Studies on the basis and potential of score level fusion are very helpful for us to understand and implement fusion at the score level. For example, based on the likelihood ratio test Nandakumar, Chen, Dass and Jain (2008) proposed a framework for the optimal combination of matching scores. Their study showed that when finite Gaussian mixture model was used to model the distributions of genuine and impostor match scores, the matching score level fusion could produce good performance. Toh Kim, and Lee (2008b) studied the issue of optimizing the ROC performance of the multimodal biometric system using the matching score fusion strategy. Jain, Nandakumar, and Ross (2005) provided us with comprehensive descriptions of various score normalization rules. Snelick, Uludag, Mink, Indovina and Jain (2005) and Poh and Bengio (2006) presented the performances of matching score level fusion algorithms and systems obtained using elaborate evaluation.

A number of studies (Dass, Nandakumar, & Jain, 2005; Kittler, Hatef, Duin, & Matas, 1998; Poh & Bengio, 2006; Ross & Jain, 2003; Schmid, Ketkar, Singh, & Cukic, 2006; Snelick Uludag, Mink, Indovina, & Jain, 2005; Vielhauer & Scheidat, 2005) have demonstrated that the matching score level fusion strategy can lead to a higher accuracy than the single biometric system. A variety of cases of matching score fusion have also been proposed (Brunelli & Falavigna, 1995; Doddington, Liggett, Martin, Przybocki, & Reynolds, 1998; Islam, Mangayyagari, & Sankar, 2007; Kumar & Zhang, 2004; Scheidat, Vielhauer, & Dittmann, 2005; Tsalakanidou, Malassiotis, & Strintzis, 2007; Tulyakov & Govindaraju, 2005; Yan & Bowyer, 2005). For example, Ribaric and Fratric (2005) acquired images containing both fingerprints and palm prints and then used the extracted eigenpalm and eigenfinger features to perform matching score level fusion. By conducting experiments on a population approaching 1,000 individuals, Snelick, Uludag, Mink, Indovina and Jain (2005) demonstrated that the multimodal fingerprint and face biometric system, which combines the two biometric traits at the matching score level, was significantly more accurate than any individual biometric systems.

Complete Chapter List

Search this Book:
Reset