A statistical approach towards performance analysis of multimodal biometrics systems

Date of Award


Publication Type

Master Thesis


Computer Science



Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.


Fueled by recent government mandates to deliver public functions by the use of biometrics, multimodal biometrics authentication has made rapid progress over the past a few years. Performance of multimodal biometrics systems plays a crucial role in government applications, including public security and forensic analysis. However, current performance analysis is conducted without considering the influence of noises, which may result in unreliable analytical results when noise levels change in practice. This thesis investigates the application of statistical methods in performance analysis of multimodal biometric systems. It develops an efficient and systematic approach to evaluate system performance in different situations of noise influences. Using this approach, 126 experiments are conducted with the BSSR1 dataset. The proposed approach helps to examine the performance of typical fusion methods that use different normalization and data partitioning techniques. Experiment results demonstrate that the Simple Sum fusion method working with the Min-Max normalization and Re-Substitution data partitioning yields the best overall performance in different noise conditions. In addition, further examination of the results reveals the need of systematic analysis of system performance as the performance of some fusion methods exhibits big variations when the level of noises changes and some fusion methods may produce very good performance in some application though normally unacceptable in others.