Date of Award

2009

Publication Type

Master Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

Applied sciences

Supervisor

Xiabou Yuan

Rights

info:eu-repo/semantics/openAccess

Abstract

This thesis proposes to study and extend the ability of the statistical methodologies that have been established to measure the performance of multimodal biometric systems. In particular, it takes into account the various noise factors that are inevitable in a real world scenario, which influence the performance of biometric systems. The work completed in the past uses the Design of Experiment framework to create a systematic approach to test the performance of biometric systems. Input parameters are varied including the data fusion methods and the normalization schemes (both controlled), and using discrete intervals based deviations in the matching scores (uncontrolled) of genuine and impostor users to represent noise. This work however, is limited provided the manual interface to the developed application. All parameters are fixed and operate over a comparatively small dataset. Further, the design of the existing application limits the extensibility of the same to incorporate additional data sources, increase or decrease the deviation values that contribute to the noise, and generate analytical graphs and reports.

It is the purpose of this thesis to establish a framework that is scalable to accommodate additional biometric databases for a larger subject pool. The developed application will also allow users to identify a larger set of deviation values for noise, automatically generate test cases for all possible biometric modalities defined within the system, etc. It is also the intent to provide, as results, the ability for the user to choose from a set of possible graphs and reports that are in tune with the common industry (commercial) standards as opposed to purely technical reports.

Download

Share

COinS