Comparative Study And Analysis Of Quality Based Multibiometric Technique Using Fuzzy Inference System

Biometric is a science and technology of measuring and analyzing biological data i.e. physical or behavioral traits which is able to uniquely recognize a person from others. Prior studies of biometric verification systems with fusion of several biometric sources have been proved to be outstand...

Full description

Saved in:
Bibliographic Details
Main Author: Hamid, Lydia Abdul
Format: Thesis
Language:English
Published: 2013
Subjects:
Online Access:http://eprints.usm.my/43781/1/Lydia%20Binti%20Abdul%20Hamid24.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Biometric is a science and technology of measuring and analyzing biological data i.e. physical or behavioral traits which is able to uniquely recognize a person from others. Prior studies of biometric verification systems with fusion of several biometric sources have been proved to be outstanding over single biometric system. However, fusion approach without considering the quality information of the data used will affect the system performance where in some cases the performances of the fusion system may become worse compared to the performances of either one of the single systems. In order to overcome this limitation, this study proposes a quality based fusion scheme by designing a fuzzy inference system (FIS) which is able to determine the optimum weight to combine the parameter for fusion systems in changing conditions. For this purpose, fusion systems which combine two modalities i.e. speech and lip traits are experimented. For speech signal, Mel Frequency Cepstral Coefficient (MFCC) is used as features while region of interest (ROI) of lip image is employed as lip features. Support vector machine (SVM) is then executed as classifier to the verification system. For validation, common fusion schemes i.e. minimum rule, maximum rule, simple sum rule, weighted sum rule are compared to the proposed quality based fusion scheme. From the experimental results at 35dB SNR of speech and 0.8 quality density of lip, the EER percentages for speech, lip, minimum rule, maximum rule, simple sum rule, weighted sum rule systems are observed as 5.9210%, 37.2157%, 33.2676%, 31.1364%, 4.0112% and 14.9023%, respectively compared to the performances of sugeno-type FIS and mamdani-type FIS i.e. 1.9974% and 1.9745%.