biometrics iris - welcome to cedargovind/cse666/fall2007/biometrics_iris.pdf · iris and when iris...
TRANSCRIPT
http://www.cubs.buffalo.edu
Iris Biometrics
University at Buffalo CSE666 Lecture Slides
Resources:• John Daugman’s webpage: http://www.cl.cam.ac.uk/~jgd1000/
http://www.cubs.buffalo.edu
Eye anatomy
http://www.biometrics.dod.mil/bio101/7.aspx
J. Daugman “Iris Recognition”, American Scientist, v.89
http://www.cubs.buffalo.edu
Eye anatomy
• The structure is formed during first two years of person’s life and does not change (except some rare eye illnesses)
http://www.cubs.buffalo.edu
Examples of iris images
http://www.cubs.buffalo.edu
Examples of iris images
http://www.cubs.buffalo.edu
Retina Biometrics?
• Low intensity infrared light reveals the structure of blood vessels on the back of the retina• More intrusive than iris scan – the camera should be positioned very close to the eye, enrollee training and cooperation required• Limited to indoor use (low light – larger pupils – better image)• Earlier deployment (CIA, FBI, NASA)
http://www.cubs.buffalo.edu
Algorithm steps
T. Tan et al.”Iris recognition: Fundamentals and State-Of-the Art”
http://www.cubs.buffalo.edu
Iris localization
( )( ) ( )
∫∂∂
0000
,,,, 2
,*max
yxryxr
dsr
yxI
rrG
πσ
Daugman:search for values which give ( )00 ,, yxr
( ) 2
2
2
1 σσ σπ
r
erG−
=, where
Explanation:
( )∫
00 ,, 2
,
yxr
dsr
yxI
πaverage grey-scale intensity on the circumference of a circle of radius r and center ),( 00 yx
http://www.cubs.buffalo.edu
•If center is properly located there will be two big changes in intensities – when pupil changes to iris and when iris changes to sclera•Convolution with searches for a peak at r with gaussian of size
Iris localization
( )∫∂
∂
00 ,, 2
,
yxr
dsr
yxI
r πChange in average grey-scale intensity on the circumference of a circle when r is increasing
),( 00 yx
)(rIr avg∂
∂
)(rIavg
( )rGσ σ
( )rGσ
http://www.cubs.buffalo.edu
Example of localization
http://www.cubs.buffalo.edu
Zones for calculating features
Inner boundary (with pupil)
Outer boundary (with sclera)
“Unwrapping” the iris
http://www.cubs.buffalo.edu
Gabor Features
( ) 220
2200
00
/)(/)()(,,;, , βφθαρφθωωβαθ φρ −−−−−−= eeeG ri
r
Daugman: use Gabor filters
Oscillation in direction of φ ρ Spread in
direction of φSpread in direction of
to extract features
0:1?)0)),(),(((Re),,;,( ,,;,00Re 00>= ∫ ∫
ρ φωβαθ φρρφρφρωβαθ ddIGrh r
0:1?)0)),(),(((Im),,;,( ,,;,00Im 00>= ∫ ∫
ρ φωβαθ φρρφρφρωβαθ ddIGrh r
Total 256*8 bit features are extracted for each iris
http://www.cubs.buffalo.edu
Examples of Gabor Filters
http://www.cubs.buffalo.edu
Matching
maskBmaskA
maskBmaskAcodeBcodeABAHDraw
∩
∩∩⊗=),(
Hamming Distance between two binary iris codes A and B:
maskBmaskA, - specifies which bits are valid
It turns out additional normalization is required to compensate for the number of actually compared bits:
911),(
2
1
2
1),(
maskBmaskABAHDBAHD rawnorm
∩
−−=
http://www.cubs.buffalo.edu
Performance
http://www.cubs.buffalo.edu
Performance
http://www.cubs.buffalo.edu
NIST evaluation (NISTIR 7408)
http://www.cubs.buffalo.edu
Examples of NIST evaluation images
http://www.cubs.buffalo.edu
Examples of NIST evaluation images
http://www.cubs.buffalo.edu
• Bits in iris code seem to be independently randomly generated • Model the distribution between two irises as a Hamming Distance between two randomly generated binary strings of length n• n=249 seems to give a theoretical distribution of HD close to real-life impostor distribution• 256*8 total –249 random = “not needed”?
- spatially close bits might be dependent
Individuality of Irises
http://www.cubs.buffalo.edu
Deployment
• Largest deployment – more than 6,500 passengers entering UAE each day are screened against a database of 420,000 persons previously expelled from country• Time of each search – 1 second
http://www.cubs.buffalo.edu
Deployment
http://www.securimetrics.com/solutions/
Operation Iraqi Freedom
US Correctional Facilities
http://www.cubs.buffalo.edu
Deployment
Employee access control at airports
Condominium entrance
http://www.cubs.buffalo.edu
Deployment (near future)
Advertising
Law enforcement
http://www.cubs.buffalo.edu
Afghan girl
http://www.cubs.buffalo.edu
Afghan girl