Dr. Arun Ross 1
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
Panel: Fundamentals of Identity Science
Arun RossAssociate Professor
West Virginia [email protected]
http://www.csee.wvu.edu/~ross
Dr. Arun Ross 2
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Biometric Traits
Dr. Arun Ross 3
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Gender
Age
Ethnicity
Medical ailment
Familial relation
Name/PIN
Levels of information
Dr. Arun Ross 4
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Universality (all users possess this trait)
Uniqueness (varies across users)
Permanence (does not change over time)
Collectability (ease of acquisition and measurement)
Performance (low error rates and processing time)
Acceptability (degree of approval by target population)
Circumvention (can it be easily spoofed or altered?)
Attributes of a Biometric Trait
Dr. Arun Ross 5
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Multiple levels
Increasing Resolution
Level I (<250 dpi)
Level II(250 – 512 dpi)
Level III (>1000 dpi)
• Increasing the resolution of the scanner reveals biometric details that can enhance the “uniqueness” of the trait
• However, this also leads to an increase in noise content
Dr. Arun Ross 6
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Biological models
Evaluation based on:
Feature sets
match scores
Approaches to establish “uniqueness”
Dr. Arun Ross 7
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Illustration only
Biological Models
Also see Kucken, Newell, “Fingerprint formation,” Journal of Theo Biol, 2005
Dr. Arun Ross 8
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Given two fingerprints with m and n minutiae, resp, what is the probability they will share q minutiae?
Feature Models
1. m=n=52, q=12PRC = 4.4 x 10-3
(Observed value = 3.5 x 10-3)
2. m=n=52, q=26PRC = 3.4 x 10-14
Dass et al, “Compound Stochastic Models for Fingerprint Individuality", ICPR, Aug 2006.
Dr. Arun Ross 9
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
316,250 Subjects
632,500 Iris Classes
200 Billion Comparisons
Score Models
Daugman J (2006) "Probing the uniqueness and randomness of IrisCodes: Results from 200 billion iris pair comparisons." Proceedings of the IEEE, 94(11), pp 1927-1935
Dr. Arun Ross 10
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Modeling Error Rates
Modality Test Label Test Parameter FNMR (FRR)
FMR (FAR)
Fingerprint FpVTE 2003 US Government operational data (>25,000 subjects)
0.6% 0.01%
Face FRVT 2006 Controlled Illumination, low-resolution images (~36,000 subjects)
2.4 –2.7%
0.1%
Iris ICE 2006 Controlled Illumination, broad quality range (240 subjects)
1.1-1.4%
0.1%
Iris ITIRT 2005 Indoor environment (1224 subjects)
0.99% 0.94%
Dr. Arun Ross 11
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Biometric Acquisition
Finger Image # 1
Sensor # 2Finger Image # 2
Sensor # 1
The acquisition process “perturbs” information
Dr. Arun Ross 12
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Choice of features
MINUTIAE-BASED TEXTURE-BASED
Dr. Arun Ross 13
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Image Enhancement
(a) Noisy image (b) Enhanced image
Dr. Arun Ross 14
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Template aging
Time duration: 6 months
Time duration: several years
Uludag, Ross, Jain, "Biometric Template Selection and Update: A Case Study in Fingerprints", Pattern Recognition, 2004.
Dr. Arun Ross 15
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Intra-class variations (FNMR)
Rn
Dr. Arun Ross 16
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
Existence of a biometric “zoo”: sheep, goats, lambs, wolves
Different categories of users impact error rates in a different manner
Capacity of a template
U2U3U1
B1 B2 B3 B4 B5 B6 … … … Bn
Theoretically: 2n usersPractical limitations: << 2n users
U4
FEATURE 1
FEA
TUR
E 2
BIOMETRIC FUSION: To increase capacity
Dr. Arun Ross 17
NSF Workshop on Fundamental Research Challenges for Trustworthy Biometrics 2010
© Ross 2010
• A number of identity management systems now need a highly accurate, scalable, real-time, low cost, user-friendly biometric recognition system
Grand Challenge in Biometrics
Accuracy
Scale
Usability
Unusable
Hard to Use
Easy to Use
Transparent to User
101
103
105
107
90%
99%
99.99%
99.999%
Jain et al. “Biometrics: A Grand Challenge”, ICPR 2004