Upload
firebazz
View
224
Download
0
Embed Size (px)
Citation preview
8/8/2019 Jain Presentation Update
1/17
Multimodal Biometric SystemsMultimodal Biometric Systems
Anil K. JainDept. of Computer Science and Engineering
Michigan State University
http://biometrics.cse.msu.edu
8/8/2019 Jain Presentation Update
2/17
Unacceptable error rates using a single biometric
Noisy biometric data
Flexibility to provide one of several possiblebiometrics
Reduce failure to enroll rate (increase populationcoverage)
Difficult to employ fake biometric
Why Multimodal Biometrics?Why Multimodal Biometrics?
8/8/2019 Jain Presentation Update
3/17
At NY airports, an average of ~ 300,000 passengers pass through daily. If all of these used biometric-authenticated smart cards for identification,
there would be 6,000 falsely rejected (and inconvenienced) passengersper day for fingerprints, 30,000 for face and 45,000 for voice. Similarnumbers can be computed for false accepts
State-of-the-art Error Rates State-of-the-art Error Rates
2-5%10-20%Text
Independent
NIST
[2000]
Voice
1%10%Varied lighting,outdoor/indoor
FRVT[2002]
Face
2%2%20 years(average age)
FVC[2004]
Fingerprint
FalseAccept Rate
FalseReject Rate
TestParameter
Test
8/8/2019 Jain Presentation Update
4/17
Poor Quality Iris ImagesPoor Quality Iris Images
Drooping eyelids Large pupil Off-centered iris
Failure to enroll (FTE) for iris is ~7% *
* BBC News, "Long lashes thwart ID scan trial", 7 May 2004, news.bbc.co.uk/2/hi/uk_news/politics/3693375.stm
8/8/2019 Jain Presentation Update
5/17
What is Multimodal Biometrics?What is Multimodal Biometrics?
8/8/2019 Jain Presentation Update
6/17
Intra-modal Fusion: Combination of
Fingerprint Matchers
Intra-modal Fusion: Combination of
Fingerprint Matchers
Jain, Prabhakar, Chen, " Combining Multiple Matchers for a High Security Fingerprint VerificationSystem", Pattern Recognition Letters, Vol 20, No. 11-13, pp. 1371-1379, 1999.
1. Minutiae-basedDynamicStringHough
2. Filter-basedFingercode
8/8/2019 Jain Presentation Update
7/17
Intramodal Fusion: Multiple Fingers,
Matchers and Templates
Intramodal Fusion: Multiple Fingers,
Matchers and Templates
8/8/2019 Jain Presentation Update
8/17
Cost vs. performance
Throughput
Verification vs. Identification mode
Choice and number of biometrics
Level of fusion
Fusion methodology
Assigning weights to biometrics
Multimodal databases
Design of Multimodal SystemsDesign of Multimodal Systems
8/8/2019 Jain Presentation Update
9/17
Which Biometric Modalities to Fuse?
Fingerprint, Hand geometry
Fingerprint, Face, Hand geometry
Voice, Lip Movement
Ear, Voice
Palmprint, Hand geometryIris, FaceFacial thermogram, Face
Fingerprint, Voice, Hand geometry
Fingerprint, Face, VoiceFingerprint, FaceVoice, Face, Lip Movement
Voice, Face
8/8/2019 Jain Presentation Update
10/17
Multibiometric Login SystemMultibiometric Login System
Face
Fingerprint
Hand geometry
Casacaded versus Parallel mode of operation
8/8/2019 Jain Presentation Update
11/17
Levels of FusionLevels of Fusion
Ross, Jain, "Information Fusion in Biometrics", Pattern Recognition Letters, September 2003.
8/8/2019 Jain Presentation Update
12/17
Three commercial fingerprint matchers and oneface matcher with EER values of 3.96%, 3.72%,2.16% and 3.76 % , respectively, were combined
972 individuals in the database
The best EER values in individual columns (rows)are indicated with bold typeface (star (*) )
1.62*1.501.824.651.73Tanh
* 0.631.16* 0.635.430.94QLQ
1.861.721.795.28*1.71Z-Score
* 0.631.160.865.430.99Min-Max
UWMWMaxMinSum
Fusion TechniqueNormalizationTechnique
MW Matcher Weighting; UW User Specific Weights
Fusion of Commercial Fingerprint
and Face Systems
Fusion of Commercial Fingerprint
and Face Systems
8/8/2019 Jain Presentation Update
13/17
Min/MaxNormalization of matching scores
Sum Rule
1000 Subjects
Fusing Commercial Face &
Fingerprint Systems
Fusing Commercial Face & Fingerprint Systems
8/8/2019 Jain Presentation Update
14/17
Performance VariabilityPerformance Variability
S c a
l a b i l i t y
G e n e r a
l i z a
b i l i t y
E f f e c
t s o
f V i r t u a
l
S u
b j e c
t s
8/8/2019 Jain Presentation Update
15/17
Cascaded Multibiometric SystemCascaded Multibiometric System
Capture biometric measurements as needed(Sequential pattern recognition)
Verification - reduces the average verification time Identification - successively prunes the database
(indexing)
To reduce the average verification time, themodalities must be cascaded in the decreasingorder of accuracy
In user-friendly systems, the user can be allowed tochoose the order of the modalities
8/8/2019 Jain Presentation Update
16/17
Fusion of Matching Scores
AdditionalBiometric?
No
Yes
ResultsFusion of Matching Scores
Cascaded Multimodal Biometric SystemCascaded Multimodal Biometric System
AdditionalBiometric?
Yes
Results
ResultsNo
8/8/2019 Jain Presentation Update
17/17
Cascaded System Performance
28% of the genuine users required only fingerprint, 13% required bothfingerprint and face, and 59% required all three modalities
Impostors had to submit all three modalities