DocumentCode :
579727
Title :
Face recognition: Impostor-based measures of uniqueness and quality
Author :
Klare, Brendan F. ; Jain, Anil K.
Author_Institution :
Noblis, Falls Church, VA, USA
fYear :
2012
fDate :
23-27 Sept. 2012
Firstpage :
237
Lastpage :
244
Abstract :
We present a framework, called uniqueness-based nonmatch estimates (UNE), which demonstrates the ability to improve face recognition performance of any face matcher. The first aspect of the framework is a novel metric for measuring the uniqueness of a given individual, called the impostor-based uniqueness measure (IUM). The UNE the maps face match scores from any any face matcher into non-match probability estimates that are conditionally dependent on the probe image´s IUM. Using this framework we demonstrate: (i) an improved generalization of matching thresholds (and, subsequently, improved matching accuracy), (ii) a score normalization technique that improves the interoperability for users of different face matchers, and (iii) the predictive ability of IUM towards face recognition accuracy. Studies are conducted on an operational dataset with 16,000 subjects using three different face matchers (two commercial, one proprietary) to demonstrate the effectiveness of the proposed framework.
Keywords :
estimation theory; face recognition; image matching; probability; IUM; UNE; face matching; face recognition accuracy; impostor-based measures; impostor-based uniqueness measure; interoperability; matching accuracy; matching thresholds; nonmatch probability estimates; operational dataset; predictive ability; probe image; score normalization technique; uniqueness-based nonmatch estimates; Accuracy; Face; Face recognition; Measurement; Probes; Sociology; Training;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Biometrics: Theory, Applications and Systems (BTAS), 2012 IEEE Fifth International Conference on
Conference_Location :
Arlington, VA
Print_ISBN :
978-1-4673-1384-1
Electronic_ISBN :
978-1-4673-1383-4
Type :
conf
DOI :
10.1109/BTAS.2012.6374583
Filename :
6374583
Link To Document :
بازگشت