Performances of weighted sum-rule fusion scheme in multi-instance and multi-modal biometric systems

Dzati Athiar Ramli, Nurul Hayati Che Rani, Khairul Anuar Ishak

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

Biometric speaker authentication systems use speaker specific information in speech signal to discriminate individuals. However, intra-speaker variations such as the changing of speaking rates, emotional and health conditions can affect system performances. One of the solutions to this limitation is to implement multibiometric system approach by combining different sources of biometric information. In this study, two approaches of m ultibiometric systems i.e. multi-instance systems and multi-modal system are experimented. Multi-instance systems consider a combination of information from several samples of different verbal extracted from the same modality while multi-modal systems is the fusion of different information extracted from different modality. Due to the different performances of each instance and each modality in the respective multi-instance system and multi-modal system, the use of weighted sum-rule fusion is suggested in this study. Performances based on sum-rule fusion are also evaluated for comparison. This research focuses on the score level fusion and Min-Max normalization technique is employed for score normalization. For speech signal feature extraction, the information in term of Mel Frequency Cepstral Coefficient (MFCC) is extracted while region of interest (ROI) of face images has been used as a second modality for the multi-modal systems. The Support Vector Machine (SVM) classifier is executed for the verification process. Experimental results prove that performances of multi-modal systems with weighted sum-rule fusion are outstanding compared to the other system performances. EER performances for multi-modal system with weighted sum-rule fusion, multi-modal system with sum-rule fusion, multi-instance systems with weighted sum-rule fusion, multi-instance systems with sum-rule fusion and speech signal single system (verbal zero) are observed as 0.0563%, 0.2778%, 1.9904%, 2.0261% and 4.3206%, respectively.

Original languageEnglish
Pages (from-to)2160-2167
Number of pages8
JournalWorld Applied Sciences Journal
Volume12
Issue number11
Publication statusPublished - 2011

Fingerprint

Biometrics
Fusion reactions
Authentication
Support vector machines
Feature extraction
Classifiers
Health

Keywords

  • Multi-instance
  • Multi-modal
  • Speech signal biometrics
  • Sum-rule and weighted sum-rule

ASJC Scopus subject areas

  • General

Cite this

Performances of weighted sum-rule fusion scheme in multi-instance and multi-modal biometric systems. / Ramli, Dzati Athiar; Rani, Nurul Hayati Che; Ishak, Khairul Anuar.

In: World Applied Sciences Journal, Vol. 12, No. 11, 2011, p. 2160-2167.

Research output: Contribution to journalArticle

@article{3d30b57f9ae24308aaae582a0f56bb0b,
title = "Performances of weighted sum-rule fusion scheme in multi-instance and multi-modal biometric systems",
abstract = "Biometric speaker authentication systems use speaker specific information in speech signal to discriminate individuals. However, intra-speaker variations such as the changing of speaking rates, emotional and health conditions can affect system performances. One of the solutions to this limitation is to implement multibiometric system approach by combining different sources of biometric information. In this study, two approaches of m ultibiometric systems i.e. multi-instance systems and multi-modal system are experimented. Multi-instance systems consider a combination of information from several samples of different verbal extracted from the same modality while multi-modal systems is the fusion of different information extracted from different modality. Due to the different performances of each instance and each modality in the respective multi-instance system and multi-modal system, the use of weighted sum-rule fusion is suggested in this study. Performances based on sum-rule fusion are also evaluated for comparison. This research focuses on the score level fusion and Min-Max normalization technique is employed for score normalization. For speech signal feature extraction, the information in term of Mel Frequency Cepstral Coefficient (MFCC) is extracted while region of interest (ROI) of face images has been used as a second modality for the multi-modal systems. The Support Vector Machine (SVM) classifier is executed for the verification process. Experimental results prove that performances of multi-modal systems with weighted sum-rule fusion are outstanding compared to the other system performances. EER performances for multi-modal system with weighted sum-rule fusion, multi-modal system with sum-rule fusion, multi-instance systems with weighted sum-rule fusion, multi-instance systems with sum-rule fusion and speech signal single system (verbal zero) are observed as 0.0563{\%}, 0.2778{\%}, 1.9904{\%}, 2.0261{\%} and 4.3206{\%}, respectively.",
keywords = "Multi-instance, Multi-modal, Speech signal biometrics, Sum-rule and weighted sum-rule",
author = "Ramli, {Dzati Athiar} and Rani, {Nurul Hayati Che} and Ishak, {Khairul Anuar}",
year = "2011",
language = "English",
volume = "12",
pages = "2160--2167",
journal = "World Applied Sciences Journal",
issn = "1818-4952",
publisher = "International Digital Organization for Scientific Information",
number = "11",

}

TY - JOUR

T1 - Performances of weighted sum-rule fusion scheme in multi-instance and multi-modal biometric systems

AU - Ramli, Dzati Athiar

AU - Rani, Nurul Hayati Che

AU - Ishak, Khairul Anuar

PY - 2011

Y1 - 2011

N2 - Biometric speaker authentication systems use speaker specific information in speech signal to discriminate individuals. However, intra-speaker variations such as the changing of speaking rates, emotional and health conditions can affect system performances. One of the solutions to this limitation is to implement multibiometric system approach by combining different sources of biometric information. In this study, two approaches of m ultibiometric systems i.e. multi-instance systems and multi-modal system are experimented. Multi-instance systems consider a combination of information from several samples of different verbal extracted from the same modality while multi-modal systems is the fusion of different information extracted from different modality. Due to the different performances of each instance and each modality in the respective multi-instance system and multi-modal system, the use of weighted sum-rule fusion is suggested in this study. Performances based on sum-rule fusion are also evaluated for comparison. This research focuses on the score level fusion and Min-Max normalization technique is employed for score normalization. For speech signal feature extraction, the information in term of Mel Frequency Cepstral Coefficient (MFCC) is extracted while region of interest (ROI) of face images has been used as a second modality for the multi-modal systems. The Support Vector Machine (SVM) classifier is executed for the verification process. Experimental results prove that performances of multi-modal systems with weighted sum-rule fusion are outstanding compared to the other system performances. EER performances for multi-modal system with weighted sum-rule fusion, multi-modal system with sum-rule fusion, multi-instance systems with weighted sum-rule fusion, multi-instance systems with sum-rule fusion and speech signal single system (verbal zero) are observed as 0.0563%, 0.2778%, 1.9904%, 2.0261% and 4.3206%, respectively.

AB - Biometric speaker authentication systems use speaker specific information in speech signal to discriminate individuals. However, intra-speaker variations such as the changing of speaking rates, emotional and health conditions can affect system performances. One of the solutions to this limitation is to implement multibiometric system approach by combining different sources of biometric information. In this study, two approaches of m ultibiometric systems i.e. multi-instance systems and multi-modal system are experimented. Multi-instance systems consider a combination of information from several samples of different verbal extracted from the same modality while multi-modal systems is the fusion of different information extracted from different modality. Due to the different performances of each instance and each modality in the respective multi-instance system and multi-modal system, the use of weighted sum-rule fusion is suggested in this study. Performances based on sum-rule fusion are also evaluated for comparison. This research focuses on the score level fusion and Min-Max normalization technique is employed for score normalization. For speech signal feature extraction, the information in term of Mel Frequency Cepstral Coefficient (MFCC) is extracted while region of interest (ROI) of face images has been used as a second modality for the multi-modal systems. The Support Vector Machine (SVM) classifier is executed for the verification process. Experimental results prove that performances of multi-modal systems with weighted sum-rule fusion are outstanding compared to the other system performances. EER performances for multi-modal system with weighted sum-rule fusion, multi-modal system with sum-rule fusion, multi-instance systems with weighted sum-rule fusion, multi-instance systems with sum-rule fusion and speech signal single system (verbal zero) are observed as 0.0563%, 0.2778%, 1.9904%, 2.0261% and 4.3206%, respectively.

KW - Multi-instance

KW - Multi-modal

KW - Speech signal biometrics

KW - Sum-rule and weighted sum-rule

UR - http://www.scopus.com/inward/record.url?scp=81755165919&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=81755165919&partnerID=8YFLogxK

M3 - Article

VL - 12

SP - 2160

EP - 2167

JO - World Applied Sciences Journal

JF - World Applied Sciences Journal

SN - 1818-4952

IS - 11

ER -