The increasing use of distributed authentication architecture has highlighted interoperability issues of biometric systems. This presentation highlights the ongoing efforts at understanding fingerprint sensor interoperability. BSPA Labs has conducted experiments over the past few years aimed at addressing the challenges related to sensor interoperability. This presentation was given at a research seminar which covered the following: Importance of fingerprint sensor interoperability, sources of issues related to sensor interoperability, analysis framework for evaluating sensor interoperability, discussion of experimental results and its practical applicability
2. Agenda
Problem statement & Goal
Data collection protocol
Results
C l i
Conclusions and Future work
d k
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
3. Problem Definition
Thermal Swipe Optical Touch Capacitive Touch
Matching fingerprints collected on different types
of sensors increases probability of false matches
of sensors increases probability of false matches
and false non‐matches
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
4. Goal
Acquisition
Technology
Interaction Sensor
Type Characteristics
Sensor
Interoperability
Create a statistical analysis framework to examine the
effect of sensor dependent distortions and variations
on False Non Match Rate (FNMR) of mixed datasets
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
5. Sensor Description
8 commercially available fingerprint sensors
8 commercially available fingerprint sensors
1 thermal swipe sensor
1 capacitive swipe sensor
4O i l
4 Optical touch sensors
h
2 Capacitive touch sensors
All sensors were 500 dpi resolution
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
7. Description of Participants
Total 190
Gender Male Female
131 59
Occupation Manual Laborer Office Worker
17 173
Age Groups < 30 years 30‐50 years > 50 years
156 23 11
Handedness Right Left Ambidextrous
164 23 3
Ethnicity Caucasian Black Hispanic American
American Asian Other
Indian
133 8 11 1 34 3
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
8. Data Analysis Methodology
• VeriFinger extractor
• Descriptive Stats
Minutiae
Count • Test of similarity
• NFIQ
• Descriptive Stats
Descriptive Stats
Image
Quality • Test of Similarity
• VeriFinger matcher
• Native and mixed datasets
FNMR
FNMR
Matrix • Test of proportions
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
12. Image Quality Analysis
Image quality computed using NFIQ
g q y p g Q
Overall Hypothesis:
H20: µi qscore = µ2 qscore……= µ9 qscore
: µ = µ =µ
H2 A: µi qscore ≠ µ2 qscore…..≠ µ9 qscore
Pairwise Comparisons:
Pairwise Comparisons:
H2’0: µi qscore = µj qscore for all i ≠ j
H2’A: µi qscore ≠ µj qscore for all i ≠ j
NFIQ uses a 3‐layer feed forward nonlinear perceptron model
to predict the image quality values based on the input feature
vector of the fingerprint image
vector of the fingerprint image
Performed a Kruskal Wallis Test of similarity of ranks
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
13. Tukey’s HSD Pairwise Test
D2 D3 D4 D5 D6 D7 D8
D1 S S S S S S S
D2 S S S S S S
D3 S S S S S
D4 S S NS S
D5 S S S
D6 S S
D7 S
S‐ Significant difference
S Significant difference
NS – Not Significant
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
16. Test of Homogeneity of Proportions
FNMR Matrix is useful for numerical comparison
p
Test of Homogeneity of FNMR was performed
between native FNMR and mixed FNMR
b i d i d
Compare multiple proportions using Marascuillo
Compare multiple proportions using Marascuillo
Procedure
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
17. Results
D1 D2 D3 D4 D5 D6 D7 D8
D1 S S S S S S S
D2 NS NS NS S S S NS
D3 S S NS NS S S NS
D4 S S NS NS S S NS
D5 S S NS NS S NS NS
D6 S S S NS S NS NS
D7 S S NS NS NS NS NS
D8 S S NS NS NS S NS
S‐ Significant difference
S Significant difference
NS – Not Significant
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
18. Impact of Quality on Interoperable FNMR
Removal of low quality images will reduce FNMR
q y g
of native and interoperable datasets
But will it make the FNMR of mixed datasets
similar to FNMR of native datasets?
i il f i d ?
FNMR matrix regenerated using VeriFinger 5 0 for
FNMR matrix regenerated using VeriFinger 5.0 for
datasets comprised of images with NFIQ score of
1, 2, or 3
The test of homogeneity of proportions was rerun
on native and mixed dataset FNMR
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
19. Results
D1 D2 D3 D4 D5 D6 D7 D8
D1 S S S S S S S
D2 NS NS NS S S NS NS
D3 S S NS NS S S NS
D4 S S NS NS S S NS
D5 S S NS NS S NS NS
D6 S S S S S S S
D7 S S NS NS NS NS NS
D8 S S NS NS NS S NS
S Significant difference
S‐ Significant difference
NS – Not Significant
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
20. Consistency of Placement
Interaction with different sensors can lead to inconsistent placement
due to sensor design, interaction type, capture area
Metric of consistency – detection of core
Core detected in both images Core detected in one image
Samples from different sensors – Samples from different sensors –
Correctly Verified Incorrectly rejected
Core detected in none of the images
Samples from same sensor – Correctly verified
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
21. Consistency of Placement
Interaction with different sensors can lead to
inconsistent placement due to sensor design,
interaction type, capture area
Metric of consistency detection of core
Metric of consistency – detection of core
D1,D6
D6,D1
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
22. Interoperability ‐ Interaction & Technology
A l d i t
Analyzed interoperability of datasets separated by
bilit f d t t t db
interaction and technology type
Created three datasets
1. Fingerprints collected from swipe sensors
g p p
2. Fingerprints collected from optical touch sensors
3. Fingerprints collected from capacitive touch sensors
g p p
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
23. Interoperability ‐ Interaction & Technology
TEST
Swipe
p Optical Capacitive
p p
Touch Touch
E Swipe 4.81 11.75 6.58
N
R
O Optical 11.93 1.53 1.89
L Touch
Capacitive 4.76 1.47 0.45
Touch
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
24. Conclusions
No technology specific impact on similarity of
gy p p y
minutiae count and quality score
Image quality analysis reduced FNMR, but did not
increase similarity of FNMR
i i il i f
Consistency of placement improved performance
of mixed datasets
of mixed datasets
Data collected from same interaction type had
lower FNMR
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
25. Future Work
Apply the analysis framework to test False Match
pp y y
Rates (FMR)
Removing low quality images on interoperable
datasets did not make FNMR more similar
d did k i il
Analyze which factors are affected by removal of low quality images
Further analysis into skin characteristics and its
Further analysis into skin characteristics and its
impact on image quality
Develop a detailed ridge spacing profile which can
be used as a sensor agnostic method for
transforming images and compensate for
El ti D f
Elastic Deformation
ti
Capture mechanisms
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
26. Upcoming Research Seminar
Paper presented at BTAS 2009-
2009
http://www.bspalabs.org/files/185/BTAS%20Fprint%20Se
nsor%20Interop.pdf
Slides available at
http://www.bspalabs.org/presentations-on-
demand
Next seminar will be on Human Biometric
Sensor Interaction – Dr Eric Kukula
Dr.
Feb 10th at 3:30pm (EST)
Sign up at
http://www.bspalabs.org/archives/1236
Biometric Standards, Performance & Assurance Laboratory | www.bspalabs.org
27. Thank You!
Questions…
Contact Information:
Shimon K. Modi, Ph.D.
modis@purdue.edu
BSPA L b
Laboratory | www.bspalabs.org
b l b
Purdue University, Knoy Hall of Technology
401 North Grant Street
West Lafayette, IN 47907 2021
47907-2021
Phone: (765) 494-2311
Fax: (765) 496-2700