CSE 40537/60537 Biometrics

Daniel MoreiraSpring 2020

Iris Recognition II

Today you will…

2

Get to knowIris acquisition and enhancement.

Enrollment

Iris Recognition

3

IrisAcquisition

UserID

sensor

presentation

IrisEnhancement

acquired iris, ID

Feature Extraction

enhanced iris, ID

templatedatabase

feature, ID

Verification

Iris Recognition

4

IrisAcquisition

IDsensor

presentation

IrisEnhancement

Feature Extraction

FeatureMatching

feature, ID

templatedatabase

1. ID

2. templatefeature

Decision

dissimilarity

outputdevice

decision

genuine

impostor

User

acquired iris, ID enhanced iris, ID

Identification

Iris Recognition

5

IrisAcquisition

sensor

presentation

IrisEnhancement

Feature Extraction

FeatureMatching

feature

templatedatabase

1. featurequery

Decision

query, gallery (with IDs),and dissimilarities

2. featuregallery

(with IDs) outputdevice

ID or Null

ID: John Doe

Null: unknown

User

acquired iris, ID enhanced iris, ID

Iris Recognition

6

IrisAcquisition

sensor

presentationUser

Acquisition

7

https://www.youtube.com/watch?v=BYN4oF_bi4c

On-line

?Off-line

It depends on the resolution and angle of capture.

Schiphol Airport

Acquisition

8

Iris CaptureVisible light.Can you see the iris texture(crypts, furrows, and collarette)?

Dr. Adam Czajka

Acquisition

9

Iris CaptureVisible light.Can you see the iris texture(crypts, furrows, and collarette)?

Acquisition

10

Iris CaptureVisible light.Melanin poses achallenge to visible-lightiris recognition.

300

2.0

1.0

1.5

0.5

400 500 600 700 800 900 1000

light wavelength

1100

visible light

mel

anin

ligh

t abs

orba

nce

Inspired by J. DaugmanAbsorption spectrum of melanin.www.cl.cam.ac.uk/~jgd1000/melanin.html

Acquisition

11

Iris CaptureSolution: near-infrared(NIR) light.

300

2.0

1.0

1.5

0.5

400 500 600 700 800 900 1000m

elan

in li

ght a

bsor

banc

elight wavelength

1100

visible light

Inspired by J. DaugmanAbsorption spectrum of melanin.www.cl.cam.ac.uk/~jgd1000/melanin.html

Typical wavelengthsused by sensors:750-890 nm.

Acquisition

12

visible light NIR

Acquisition

13

StandardsEye SafetyIEC 60825-1:1993(+ addendum A1:1997 and A2:2001),ANSI RP-27.1-96Maximum Permissible Exposure ( ):MPEMPE < 0.1 × MPEmax

eye damage due tolight exposure

Acquisition

14

StandardsImage QualityISO/IEC 19794-6 and ISO/IEC 29794-6wavelength: 700-900 nmresolution: 20 lines per iris diameternon-occluded iris area: 70%gray scale: 6 bitstypical resolution: 640 x 480 pixels

≥≥

≥

Acquisition

15

SensorsWith cooperation.

LG Iris Access 3000

Jim Wilson / The New York Times Dr. Adam Czajka Dr. Adam Czajka

CrossMatch IG-AD100

Acquisition

16

SensorsWith almost no cooperation.Multiple-Resolution Cameras Wide-angle camera for face detection.Narrow-angle cameras for iris capture.

https://www.youtube.com/watch?v=bolNgCrCZW0

Sarnoff Corp., Iris-on-the-move Gate

Acquisition

17

SensorsWith almost no cooperation.Deformable MirrorsSimilar to astronomical telescopes.Fast adaptation at presentation time.Capture at 1.5-2.5mof distance.

Dr. Adam Czajka

AOptix Insight SD, 2008

Acquisition

18

SensorsCurrent trend: miniaturization.Example 1Android-basedFidelys smartwatch.

linuxgizmos.com/worlds-first-iris-recognition-smartwatch-runs-android

Acquisition

19

SensorsCurrent trend: miniaturization.Example 2IriShield USBSee demonstration.

https://urvashicomputers.com/irishield-mk-2120-series/

Acquisition

20

ChallengesDeformations and OcclusionsEyelids and eyelashes.Specular reflections.Pupil dilation.Head movement, off-axis gaze.

http://www.cse.nd.edu/BTAS_07/John_Daugman_BTAS.pdf

Acquisition

21

ChallengesUser CooperationIt is easy for people toprotect their irises from capture.

Acquisition

22

ChallengesDiseasesE.g., cataracts, conjunctivitis.Is iris visible?Is the disease contagious?

E.g., cataracts.

commons.wikimedia.org

Acquisition

23

ChallengesAttacksObfuscation with texturizedcontact lenses.Presentation attack(see demonstration).

Jain, Ross, and NadakumarIntroduction to Biometrics Springer Books, 2011

Iris Recognition

24

IrisAcquisition

sensor

presentation

IrisEnhancement

acquired iris, ID

User

Enhancement

25

StepsSegmentationKeep only useful information(iris texture).NormalizationMake different captures of the sameiris look as similar as possible.

Enhancement

26

Segmentation (1/2)Iris and Pupil LocalizationLocalize limbus and pupillaryboundaries.

http://www.cse.nd.edu/BTAS_07/John_Daugman_BTAS.pdf

Enhancement

27

Segmentation (1/2)Iris and Pupil LocalizationLocalize limbus and pupillaryboundaries.Eyelid, Eyelash, and Specular Reflection DetectionDeal with iris texture occlusions.

http://www.cse.nd.edu/BTAS_07/John_Daugman_BTAS.pdf

Enhancement

28

Segmentation (1/2)Iris and Pupil LocalizationLocalize limbus and pupillaryboundaries.Eyelid, Eyelash, and Specular Reflection DetectionDeal with iris texture occlusions.

http://www.cse.nd.edu/BTAS_07/John_Daugman_BTAS.pdf

Enhancement

29

Segmentation (1/2)Iris and Pupil LocalizationMethod 1: Integral-differentialoperatorObjective:Find of limbus and

of pupillary boundaries.(r, x0, y0)

(r, x0, y0)

http://www.cse.nd.edu/BTAS_07/John_Daugman_BTAS.pdf

Segmentation (1/2)Iris and Pupil LocalizationMethod 1: Integral-differentialoperatorStrategy:Try various values for .(r, x0, y0)

Enhancement

30

1. pixel values2. integral of circle

with radius r3. derivative of

integral

4. smoothing function (gaussian)

5. get the configuration with maximum values

(r, x0, y0)

Segmentation (1/2)Iris and Pupil LocalizationMethod 1: Integral-differentialoperatorStrategy:Try various values for .(r, x0, y0)

Enhancement

31

different values of r

pupil limbus

Segmentation (1/2)Iris and Pupil LocalizationMethod 1: Integral-differentialoperator

Enhancement

32

result example

J. DaugmanHow Iris Recognition WorksIEEE TCSVT, 2004

Segmentation (1/2)Iris and Pupil LocalizationMethod 2: Image processingending with Hough circletransform.

Enhancement

33

https://github.com/olesia-midiana/iris-recognition-py

Enhancement

34

1.grayscale 2. inverted 3. median blur 4. Canny edge detector

https://github.com/olesia-midiana/iris-recognition-py

5. Hough circletransform

Enhancement

35

Segmentation (1/2)Eyelids, eyelashes,specular highlightsFit of parabolic curves for eyelids.Active contours (curve evolution)to avoid eyelashes.Fit of elliptical curves for specular highlights.Machine learning from annotated examples.

GutfeterActive contours for iris segmentation.BSc Thesis, WUT, 2010

Enhancement

36

Segmentation (1/2)Iris LocalizationConvolutional Neural Networks(machine learning trained withannotation examples).

Kerrigan et al.Iris Recognition with Image Segmentation EmployingRetrained Off-the-Shelf Deep NeuralNetworkshttps://arxiv.org/abs/1901.01028, 2019

Enhancement

37

Segmentation (1/2)Manual SegmentationNext slide: iris recognition toolthat we have developed at Notre Dame.

Enhancement

38

Segmentation (1/2)Iris and Pupil LocalizationLocalize limbus and pupillaryboundaries.Eyelid, Eyelash, and Specular Reflection DetectionDeal with iris texture occlusions.

Enhancement

39

Normalization (2/2)

source target

Enhancement

40

Normalization (2/2)

source target

Enhancement

41

Normalization (2/2)

source target

Enhancement

42

Normalization (2/2)

source target

Enhancement

43

Enhancement

44

LimitationsSegmentationPupil and iris are not concentric(pupils are slightly shifted to the nasal corner).They are not perfectly round.

Enhancement

45

Dr. Adam Czajka

coloboma condition

Enhancement

46

Dr. Adam Czajka

Enhancement

47

LimitationsNormalization Forcing circular modelsmay lead to poor mapping.

John DaugmanEvolving Method in Iris RecognitionBTAS, 2012

Iris Recognition

48

IrisAcquisition

sensor

presentation

IrisEnhancement

User

acquired iris, ID

Iris Recognition

49

IrisAcquisition

sensor

presentation

IrisEnhancement

Feature Extraction

FeatureMatching

feature

templatedatabase

1. featurequery

Decision

query, gallery (with IDs),and similarities

2. featuregallery

(with IDs) outputdevice

ID or Null

ID: John Doe

Null: unknown

User

acquired iris, ID enhanced iris, ID

Iris Description andMatching

S’up Next?

50

AcknowledgmentsThis material is heavily based on

Dr. Adam Czajka’s and Dr. Walter Scheirer’s courses.Thank you, professors, for kindly allowing me to use your material.

https://engineering.nd.edu/profiles/aczajka

https://www.wjscheirer.com/