Performances of the likelihood-ratio classifier based on different data modelings

C. Chen; Raymond N.J. Veldhuis

doi:10.1109/ICARCV.2008.4795718

Performances of the likelihood-ratio classifier based on different data modelings

C. Chen, Raymond N.J. Veldhuis

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

1 Citation (Scopus)

199 Downloads (Pure)

Abstract

The classical likelihood ratio classifier easily collapses in many biometric applications especially with independent training-test subjects. The reason lies in the inaccurate estimation of the underlying user-specific feature density. Firstly, the feature density estimation suffers from insufficient number of user-specific samples during the enrollment phase. Even if more enrollment samples are available, it is most likely that they are not reliable enough. Furthermore, it may happen that enrolled samples do not obey the Gaussian density model. Therefore, it is crucial to properly estimate the underlying user-specific feature density in the above situations. In this paper, we give an overview of several data modeling methods. Furthermore, we propose a discretized density based data model. Experimental results on FRGC face data set has shown reasonably good performance with our proposed model.

Original language	Undefined
Title of host publication	10th International Conference on Control, Automation, Robotics and Vision, 2008. ICARCV 2008.
Place of Publication	Los Alamitos
Publisher	IEEE
Pages	1347-1351
Number of pages	5
ISBN (Print)	978-1-4244-2287-6
DOIs	https://doi.org/10.1109/ICARCV.2008.4795718
Publication status	Published - 17 Dec 2008
Event	10th International Conference on Control, Automation, Robotics & Vision, ICARCV 2008 - Melia Hanoi, Hanoi, Viet Nam Duration: 17 Dec 2008 → 20 Dec 2008 Conference number: 10

Publication series

Name
Publisher	IEEE Computer Society Press

Conference

Conference	10th International Conference on Control, Automation, Robotics & Vision, ICARCV 2008
Abbreviated title	ICARCV
Country/Territory	Viet Nam
City	Hanoi
Period	17/12/08 → 20/12/08

Keywords

SCS-Safety
EWI-15180
CR-I.5
METIS-256467
likelihood-ratio classifier
IR-65417
Density estimation
Quantization

Access to Document

10.1109/ICARCV.2008.4795718

Cite this

@inproceedings{4268a8c73b2d4f73a6e1e59b5efba7a9,

title = "Performances of the likelihood-ratio classifier based on different data modelings",

abstract = "The classical likelihood ratio classifier easily collapses in many biometric applications especially with independent training-test subjects. The reason lies in the inaccurate estimation of the underlying user-specific feature density. Firstly, the feature density estimation suffers from insufficient number of user-specific samples during the enrollment phase. Even if more enrollment samples are available, it is most likely that they are not reliable enough. Furthermore, it may happen that enrolled samples do not obey the Gaussian density model. Therefore, it is crucial to properly estimate the underlying user-specific feature density in the above situations. In this paper, we give an overview of several data modeling methods. Furthermore, we propose a discretized density based data model. Experimental results on FRGC face data set has shown reasonably good performance with our proposed model.",

keywords = "SCS-Safety, EWI-15180, CR-I.5, METIS-256467, likelihood-ratio classifier, IR-65417, Density estimation, Quantization",

author = "C. Chen and Veldhuis, {Raymond N.J.}",

note = "10.1109/ICARCV.2008.4795718 ; 10th International Conference on Control, Automation, Robotics & Vision, ICARCV 2008 ; Conference date: 17-12-2008 Through 20-12-2008",

year = "2008",

month = dec,

day = "17",

doi = "10.1109/ICARCV.2008.4795718",

language = "Undefined",

isbn = "978-1-4244-2287-6",

publisher = "IEEE",

pages = "1347--1351",

booktitle = "10th International Conference on Control, Automation, Robotics and Vision, 2008. ICARCV 2008.",

address = "United States",

}

Chen, C & Veldhuis, RNJ 2008, Performances of the likelihood-ratio classifier based on different data modelings. in 10th International Conference on Control, Automation, Robotics and Vision, 2008. ICARCV 2008.., 10.1109/ICARCV.2008.4795718, IEEE, Los Alamitos, pp. 1347-1351, 10th International Conference on Control, Automation, Robotics & Vision, ICARCV 2008, Hanoi, Viet Nam, 17/12/08. https://doi.org/10.1109/ICARCV.2008.4795718

Performances of the likelihood-ratio classifier based on different data modelings. / Chen, C.; Veldhuis, Raymond N.J.
10th International Conference on Control, Automation, Robotics and Vision, 2008. ICARCV 2008.. Los Alamitos: IEEE, 2008. p. 1347-1351 10.1109/ICARCV.2008.4795718.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Performances of the likelihood-ratio classifier based on different data modelings

AU - Chen, C.

AU - Veldhuis, Raymond N.J.

N1 - Conference code: 10

PY - 2008/12/17

Y1 - 2008/12/17

N2 - The classical likelihood ratio classifier easily collapses in many biometric applications especially with independent training-test subjects. The reason lies in the inaccurate estimation of the underlying user-specific feature density. Firstly, the feature density estimation suffers from insufficient number of user-specific samples during the enrollment phase. Even if more enrollment samples are available, it is most likely that they are not reliable enough. Furthermore, it may happen that enrolled samples do not obey the Gaussian density model. Therefore, it is crucial to properly estimate the underlying user-specific feature density in the above situations. In this paper, we give an overview of several data modeling methods. Furthermore, we propose a discretized density based data model. Experimental results on FRGC face data set has shown reasonably good performance with our proposed model.

AB - The classical likelihood ratio classifier easily collapses in many biometric applications especially with independent training-test subjects. The reason lies in the inaccurate estimation of the underlying user-specific feature density. Firstly, the feature density estimation suffers from insufficient number of user-specific samples during the enrollment phase. Even if more enrollment samples are available, it is most likely that they are not reliable enough. Furthermore, it may happen that enrolled samples do not obey the Gaussian density model. Therefore, it is crucial to properly estimate the underlying user-specific feature density in the above situations. In this paper, we give an overview of several data modeling methods. Furthermore, we propose a discretized density based data model. Experimental results on FRGC face data set has shown reasonably good performance with our proposed model.

KW - SCS-Safety

KW - EWI-15180

KW - CR-I.5

KW - METIS-256467

KW - likelihood-ratio classifier

KW - IR-65417

KW - Density estimation

KW - Quantization

U2 - 10.1109/ICARCV.2008.4795718

DO - 10.1109/ICARCV.2008.4795718

M3 - Conference contribution

SN - 978-1-4244-2287-6

SP - 1347

EP - 1351

BT - 10th International Conference on Control, Automation, Robotics and Vision, 2008. ICARCV 2008.

PB - IEEE

CY - Los Alamitos

T2 - 10th International Conference on Control, Automation, Robotics & Vision, ICARCV 2008

Y2 - 17 December 2008 through 20 December 2008

ER -