IriScan is a new technology, using the unique patterns of the human iris. It provides positive identification of an individual without contact or invasion. Recent formal testing for the Defense Nuclear Agency has confirmed the laboratory test results. Commercial products were first available in October 1995.
IriScan is a new technology, using the unique patterns of the human iris. It provides positive identification of an individual without contact or invasion. Recent formal testing for the Defense Nuclear Agency has confirmed the laboratory test results. Commercial products were first available in October 1995.
IriScan is a new technology, using the unique patterns of the human iris. It provides positive identification of an individual without contact or invasion. Recent formal testing for the Defense Nuclear Agency has confirmed the laboratory test results. Commercial products were first available in October 1995.
IriScan is a new technology, using the unique patterns of the human iris. It provides positive identification of an individual without contact or invasion. Recent formal testing for the Defense Nuclear Agency has confirmed the laboratory test results. Commercial products were first available in October 1995.
Iris Recognition Technology
Gerald 0. Witlams
Abstract-—~ triScan Inc. has for the past (wo
years, been developing an
dentification/vetification system capable of
positively identifying and vertying the identity of
Individuals without physical contact or a person
in the loop. Personal identifeation
Historically been based on what a
possesses (a card); knows (a Personal
Ientiication Number); or is (an_ inherent
physiological or behavioral - characteristic).
Facial features (both infra-red signatures and
geometry), fingerprints, hand geometry, vein
Patterns, retinal patterns, voice patterns, and
signature dynamics have all been explored as
biometric identifiers with varying levels. of
sucease, All but voice and facial recognition
quire contact or have been characterized by
some as invasive or intrusive, Many have
Suffered from high cost and unsatisfactory error
rates. A new technology, using the unique
patterns of the human ile, shows promise of
overcoming previous shortcomings and
providing positive identification of an individual
Without contact or invasion, at extremely high
confidence level,
‘The videosbasod systom locates the eye and iris;
evaluates the degree of occlusion by eyelid and
spectral reflection; determines. the quality of
Jmage focus; and determines the center and
‘boundary ofthe pupil and the limbus (outer edge
‘ofthe iris) for processing. The iris is zoned, and
the features therein measured and encoded into
256 byte (2048 bit) IrisCode for enrollment or
Identiication. The presented biometric is
compared to an. extensive database for
Identification, or to a referenced irisCode for
verfication. "Computations and decisions are
accomplished at extremly high rates of speod,
Tesulting in processing times of lose than two
seconds.
“The process is based on the unique nature and
tho extreme richness of the human ie, Tho:
muiiple contraction furrows, collagenous fibers,
crypts, coronas, ‘serpentine
vasculature, freckles, rift, and pite produce. @
omduplicable organ with more than 400
degrees of treedom, or measurable variables
‘The IriScan process typically uses about 200 of
these to create a code which can be compared to
fan entre databace in milliseconds, producing a
Positive identification with “imposter odds" as
high as 1 In 10". The process. dotines a
"Hamming Distance" or measure of dliference
between the presented biometric and the
feference irisCodes. The Hamming sistance:
determines under which portion of the frequ
Gitributions “(developed from over 6 millon
comparisons of Iie data) the would-be entrant
falls when the irs ls presented. This location
also determines where in the “Imposter”
0700395376 10884.00 6196 EEE
frequency distribution the individual fale and
thus. provides quantifiable Imposter odds on
which to base an entry decision. Calculated
‘error curves for Type | (False Reject) and Typo
(alse Accopt) errors overlap ata mathomatically
determined point which provides a "Crossover
Error Rate" where the probablity of Type | and
‘Type Il errors are equal. In comparison to the
best Crossover Error Rates obtainable by current
biometri technology of about 3%, the IriScan
Fate of 0.00076% represents a quantum leap
orward for postive personal identification.
Recent formal testing for the Defense Nuclear
Agency (ONA) has confirmed the laboratory te
results. Beta testing in an operational field
fenvironment has demonstrated the practically
fof such an identification dovice in a real world
Scenario. Together, thie recent testing has
Substantiated tho operational viability ofpersonal
|donttiers baeed on ie identification technology.
Commercial producte were first available in
October 1998.
1.0 INTRODUCTION
1.4 Background
Personal identification has historically been
based on what a person possesses (a latter of
Introduction perhaps), knows (a secret passwor),
fr is (personal recognition). in today's secunty
Industry, possession has become an encoded cara
knowledge equates to knowing one's Personal
Identiication "Number (PIN), and personal
recognition has given way t0 physiological and
behavioral characteris known ‘ae. biometic,
Various human foatures have been used as the
basis for biometric mansuremant. Thee include
fingerptints, paimprints, hand goomety, vein
patterns, facial characteristics, and capilary pat
Inthe retina. Additionally, bohavioralcharactoristics
such as signature dynamics, voice pattern, and
keystroke dynamics have been used for
identification of verflction. Many of these require
Contact or are perceived as Invasive or intrusive
(thors require person to make a final judgement
fo are cost, oF suffer from unsatisfactory error
rates,
IiScan nc. has forthe past two years, been
developing an automated identiication/eriicaton
{system capablect posively identiying and verying
without physiol Contact or a parson in the loop.
‘Tho. iis dentiieation technology, patented by
IniScan, shows promise of mecting ths challenge
without suttering many ofthe inadequacies exhibited
by the technologies mentioned above. In addition
to developing a commercialy viable idetiicaton
system for industria application, IiScan completeddovelonment of a brasstoard mod for use by the
Department of Detonse (000). This paper attempts
to report the progress of development of is
identication technology and inckides data from
both the DNA and commercial application program
‘mentioned above as well as IiScan's independent
research, This paper is intended for securty
Pracitioners. who! are knowledgeable, but not
technically or scientifoaly oriented. Appendox A
High Confidence Personal identiicaton By Rad
Video “Anaisis of Ie Texure, by Dr John
DDaugman, i Itended for the more tochnioally of
scientiialy oriented readers.
1.2 Caveats
1.21 Ins idonticaton technology and the [Scan
process are not associated with, of In any way
‘similar ‘0 retina (capilary) pattern recognition. The
iis isthe colorul donut-shaped organ surrounding
the pupil The retina is the hemispherical organ
behind the comea, lens, is, and pupil The IiSoan
process eaplures a video Image (takes a video
ictus) of the extemal ie whi the retinal soan
process scans the fovea or innermost surface ofthe
intemal orga, the retina,
41.22 The intial production run for the: fist
‘commercial product (The IiScan System 2000EAQ),
intended for Access Control application is complete
land units are avaiable, Because of thei recent
Inroduction, however, systems ave not been
formally tested by an independent testing agoncy.
Systems have been formally tested by InScan in @
laboratory environment in accordance with a test
an spproved by the Defense Nuclear Agency
(NA) as part ofthe brassboard project mentioned
‘atl. Adkdtonaly, systems havo been Beta tested
in operational environments ata unversty medical
‘school, @_national laboratory, and major
‘corrections facil
1.28 Much ofthe data reported in this paper has
been obtained from laboratory models, &
baseboard modal, and pre-production prototypes
‘operated by trained personnel whose orientation is
te optimize system performance. As a result, the
data may of may not reflect the resus thal wil be
‘obtained in an uncontroled environment subject to
the fobles and vagaries of the general population
1.24 The developmental and pre;production
Prototype idontfeation systems have to-date
‘operated in. the identfication (tecogniion) mode
conly. The veriation mode (where card or PIN
inputs ae usedto preselect the appropriate enrolled
fle) has been testod for the DNA, and future
prototypes and production modele may cortain this
feature as ar optional tom,
1.9 Dafictions
1.91 Identiication: The process of presenting a
biometric feature without selecting a single i inthe
dtabase fo compared to. Requires a comparison
‘of the presented biometric with each fle in the
database to. determine idently. Colloquially
desorbed as “one-on-many’
1.8.2 Verficstion: The process of presenting a fle
number along with a biometric feature so that the
system compares the presontod biometric wth only
fone fo in the database (‘one-on-one
1.89 Hamming Distance (HO): A quantitative
measure of diferance between two. IrisCodes
‘expressed as a two-decimal number. Itcan also be
expressed as a percentage. For example, an HD of
(08 indicates that two IrisCodes are &% diferent
2.0 BASIS OF THE TECHNOLOGY
2.4 Features of the ls
‘The human iris i ich in features which can
be used to quantitatively and postively distinguish
fone trom another. The iis contains multiple
collagenous fives, contraction furows, coronas,
crypts, color, serpentine vasculature, striations
Ireokles, rts, and pits, Further, measuring these
features for height, with, angle of deflection, and
location provides other quartiiable parameters
Useful to the Kietiicaion process. In al, there are
‘ver 400 Degress of Freedom (OOF) on which to
base a statistical analysis and comparison. In
practice, when all onjunct features ave excluded,
approximately 200 DOF are useable inthe Scan
comparison process. This is three to fourtimes the
frumber available to conventional tingerptint
identification systems
2.2 Uniqueness ofthe is
The Irie I unique because of the chaotic
‘morphogenesis ofthat organ. To quote Or. John
Daugman (Append A), "An advantage the iis
shares with fingerprints Is the chaotic
morphogenesis of is minutiae, Theis texture has
Chaotic dimension because Re details depend on
intial conditions in embryonic genetic expression:
Yet. the limtation of paral genetic penetrance
(beyond expression of form, function, color and
‘general textural quality), ensures thet even identical
‘wins have uncorelated ie minutiae. Thus the
Uniqueness of every is, including the pair
Postessed by one. indlidual, paralele the
Uniqueness of every fingerprint regardless. of
whether there is a common genome” Given this
the statistical probabil that two lrses would be
Identical by random chance Is calculated st
‘approximately 1 in 10
29 Stablty ofthe iis
Notwithstanding ts delicate nature, the sis
protected behind eyetd, comes, aqueous humor,
and frequently eyestasses or contact lenses wich
have negigible effect on te identification process)
‘Anonmel isis usualy kibrieated, not contaminated
‘with foreign material, and instinctively, one of the
most carefully protected organs In one's body.
‘Adaltionaly, the features of thei, thir placement,
size, shape, and orientation remain stable and fod
‘rom about one year of ege throughout fe
24 Natural protection fom ation
The iis has physiologiea! characteristics
which can be exploited to insure that reproductions
fof a human iis cannot be Used to fool the system
One of these Is known as pupilary unrest, the
autonomic reeponees of the eye to ight variationsthat are go minute and rapid that they fll below our
level of consoousness, Study the pupils of
someone under normal’ ambiont "lighting
Conditions, and you wil note that they are almost
‘constantly in mation, opening and closing within &
‘small range without the introduction of any
detectable varying ght source. A. simple
comparison of pupll size between multiple video
frames can detect pupllary unrest as acorfimation
athe *humanness’ ofthe is being presented. An
fdational ive eye" tests the response ofthe pupil
to light pulses datectod by the aye either directly
inta the aye being presented (primary rflex) or the
fee aye (oonsensusl reflex). Finally, there ie &
complex interaction of reflections and rafactions
which cours inthe human eye. producing
‘measurable “Purkinf images" which distinguish a
Ihe eye froma non-human representation of an ay.
3.0 PROCESSING
9.4 Imsge acquisition
‘The optical platform of the is identification
system acquires muiipie images of the presented
ins through @ simple lens, @ monochrome COD
camera. and a frame grabbing board. A low:
level@20 wat, operating at 7 watts) quartz halogen
umipator, operating trom .1 to 1-2 microns at
3200" provides ilursntion approximately 14° rom
the lens (9" from the front of the optical platform),
fand a liquid crystal cisplay with beam splitter
provides feedback through the lens to ald the user
Inalignmant. There isa diamond shaped reflection
in the pupil which shrinks to a dot as the user
approaches the optimum focal point which acts a=
‘8 secondary aid t0 user alignment and Image.
facquistion.
82 tis detintion
‘The image which best meets the focus and
otal laity requirements of tho system is then
‘analzedto locate the limbus (he outer boundary of
the iis, whore t meets the white sclera ofthe eye =
see Figure $1), the nominal boundary where the
Creu pup meets the iri, and the center of the
pup. The precise location ofthe eroular is hes
ow been defined and further processing can occur
39 Zones of analysis
The system dotinestes eight zones of
analysis on the iis. (Figure 1) using @
Statie/dymamic system. Through software, the static
portion of the. system truncates what would
btharwiea be circular zones much fia a bull2ye
target, The zones are truncated at top and bottom
te avoid analyzing special reflection fom the sb.
clock luminator, and te ignorethe occlusion from
the upper eyo. ‘The dynamic nature ofthe system.
‘s such that t automatically adjusts tho wiath of the
zones in realtime to maximize the amount of ine
fanalyzed glen varying rtios of pupil to is sizes
Truncation of the zonee has itle effect on the
analysis process. Even the reduced-size zones
have mere than enough feature variants toute all
256 bis. In actual practice, excelent enrlimenis
land subsequent Kentfcations are obiainod with
only 40% oF less of the ins avallable for analysis,
Features are located based on e polar coordinate
system
9.4 Image ansiysie
‘The features ofthe i's are then analyzed and
dligtized Into a 256 byte (2048 bi) lrisCode Figure
5:1, upper lft comer) which is then stored in the
database for future comparison. When an irs is
presented at poral, the same process occurs, and
the rosutant IneCode is compared to every fle In
‘the datebase In the idertfication or “recognition™
‘ode, or fo a selactad fein the veication mode,
9.5. Hamming distance caleulation
When a would-be entrant presents an iis at
42 pottal, the lrisCode oblained in the manner
described above is compared tothe IisCode(s) in
the database to calcula a Hamming Ditans (HD)
‘25a quantitative measure of how differant the
presented InsCode is trom each isCode. Each of
2088 pairs of bis are compared (Figure 62). Bi#t
‘rom tho presented IrsCode is compared to bit #1
fom the reference IisCoda, bit #2. from the
Presented IrsCode is compared to bit #2 trom the
Felerence IreCode, and so on. i two bits are ake
(two "1's, ortwo 0's) tho system assigns a value of
zero to that pair comparison, if two bits are
‘ifort, the system assigns a value of one to thet
pelr comparison, After al pars are compared, the
assigned values are summed, and divided by the
{otal numberof pair comparisons ruling in a two
dligt quantitative expression of how diferant the two
lisCodes are (sometimes the full 2048 pais are not
compared because the systom detects and
fxchides spectral refections that were not presant
during ervolmend.
‘36 entiation ot rection
Many thousand ICode comparisons of
both known impostore and known authentes have
statsticaly defined the Frequency Distributions
(Probabilty Dente) in Figure 8° iis clear that
the mean imposter Hamming Distance is near 05,
hich is what one would expect in a truly random
relationship tlipsing a coin several lion times and
noting the occurrence of heads versus tal), The
‘mean vahie for authertics onthe other hand is lese
than’ 0.10 (or 10%). Another feature of the
Frequency Dietibutions ie the very emall Standard
Deviation, resulting in very steop slopes and tty
Clustered values for HD (Hamming Distance) in
both, Itappears, visually at leas, to Be two discrete
Fee
| HOLWWLOaduad = OWaZ = — lam;
FONVLSIC SNINIAVHost oot os
o0z
ooe
ose
sourisig SuwWeH eve ombra
go so yO eo
A L
*
qwunog,
H
siejsoduyj pue solueyjny Joy seoueisiq Bulwuweyi-p ean6ya eourjsiq Bujwwey
a
=o 1
4000'0
\ r 2000°0
‘ \ |
\deooy asjey \ t ora
. |
eed \ 7000'0
000161: 1--eouajeainb3 \
s000°0
\
\\ +9000°0
sioug || adh, — \ |
sioug | adh, — pofey asjey | 40000
Coes 80000
(seyey 10113) Lee
SAONVNHOSAYHSd 10113 Jo Pd