Chapter 6
Iris and Retina Recognition
Iris Recognition
2
Eye Components
3
Iris Recognition vs Retina
Recognition
4
Iris: Anatomy
• Iris is located behind the cornea and in front of the
lens.
• Complex mesh of muscle beams, blood vessels,
nerves, and pigmented skin.
5
Anatomy
6
Function of the Iris
• To regulate the amount of light entering the
eye by dilating or contracting the pupil.
7
Introduction
• Iris recognition works on the basis of iris pattern in
human eye.
• The iris is the pigmented elastic tissue that has
adjustable circular opening in center.
• It controls the diameter of pupil.
• In adult humans, the texture of iris is stable
throughout their lives.
• The iris patterns of left and right eyes are different.
• The iris patterns and colors change from person to
person.
8
Iris Structure
9
Iris Structure
• The iris constricts and dilates in a nonlinear
fashion
10
Iris Recognition
• It involves taking the picture of iris with a capable camera,
storing it, and comparing the same with the candidate eyes
using mathematical algorithms
11
Why Irises?
12
Why Irises?
• Universality
– Does everybody have the trait? – Probably
• Uniqueness
– How likely two or more individuals will present the same
trait?
• Permanence
– How easily does the trait change?
• Needed Research - There seems to be a degradation of True Match
Rate (TMR) as a function of time.
• Some traumas and diseases might degrade/change the iris. For
example, cataracts
• Measurability
– How easy is it to acquire and digitize the trait? 13
Why Irises?
• Acceptability
– Will individuals collaborate during data collection?
• Circumvention
– How hard can the trait be forged or imitated? – contact
lens
– Irises can be used in identification soon after death.
• Performance
– How good is the trait quantitatively according to objective
metrics?
• Accountability
– How easy is it for everyone to understand the trait
comparison? 14
Brief History of Iris Recognition
1. 1883: Alphonse Bertillon
– eye color (not used today in iris recognition); extension of
the Bertillonage system
2. 1936: Frank Burch
– presentation of the iris recognition idea at the opening of
the American Academy of Ophthalmology Annual Meeting
3. 1949: James Doggarts
– one of a few reproductions of Burch’s idea in
ophthalmology textbooks
4. 1987: Aran Safir, Leonard Flom
– US patent, somewhat based on Burch’s idea
15
Brief History of Iris Recognition
5. 1992: John Daugman
– first algorithm of iris image coding based on two-dimensional Gabor
filtering
– first prototype of the iris recognition system (Bench Model 2.5)
6. 1994: John Daugman
– iris code generation methodology becomes patented
– the so called ‘Daugman’s method’ presents absolutely brilliant
accuracy, and it becomes de facto standard in iris recognition,
inspiring iris recognition scientists in their research
7. 2005: Safir’s and Flom’s patent expires, new solutions start
to appear on the market
16
Iris Image Representation
ISO/IEC 19794-6
1. Raw image (ISO: UNCROPPED or KIND 1)
2. Image with standard resolution of 640×480 pixels (ISO: VGA or KIND 2)
3. Cropped image (ISO: CROPPED or KIND 3)
4. Cropped image with eyelids and sclera masked (ISO:
CROPPED_AND_MASKED or KIND 7)
17
Application of Daugman’s method
• Left - Photo of young Sharbat Gula captured in 1984 in the refugee camp
in Pakistan during the time of Soviet occupation of Afghanistan – National
Geographic Magazine cover 1985
• Right - John Daugman applies his method for verification of the girl’s
identity in 2002
18
Iris Image Capture and Representation
• Use of visible light
– Possible, sometimes necessary ...
– ... but may be problematic for dark eyes
19
Iris Image Capture and Representation
• Solution: near-infrared (NIR) light
20
Iris Image Capture and Representation
• Use of visible light
– Melanin: important protective substance
– Melanin poses a challenge to visible-light iris recognition
21
Iris Image Capture and Representation
• Use of Near Infrared (NIR) light
• Visible vs. near infrared (NIR) - Example
22
Standard Requirements
1. Eye safety
– standards: IEC 60825-1:1993 (+ addendum A1:1997 and
A2:2001), ANSI RP-27.1-96
– Maximum Permissible Exposure (MPE) not greater than 0.1 ∗
MPEmax, where MPEmax is the exposure resulting in severe
eye damage in 50% of population
2. Image quality
– standards: ISO/IEC 19794-6 and ISO/IEC 29794-6
– wavelength (700-900 nm), resolution (at least 120 lines per iris
diameter), usable iris area (at least 70%), gray levels (8 bit
dynamic range, minimum 6 bits of useful information), pupil
location (centered within the image), gaze (eye fully on-axis
with the camera lens), no patterned contact lens, etc.
– typical iris image resolution: 640×480 pixels 23
Typical Problems
1. Iris: a small, three dimensional and fidgety object
– conflict of interests: CCD sensitivity, exposure times and aperture vs.
optical depth-of-field
2. User cooperation and habituation
3. Deformations and obstructions
– easy: eyelids, specular reflections
– hard: eyelashes, hairs, patterned contact lenses
– iris constriction and dilation, hippus, pupil dynamics: surface warping,
nonlinear deformations
– head movement (important in high zoom)
– off-axis gaze
4. Interoperability of capture devices
– different wavelengths of illuminating light
– different number and location of illuminants
24
Interoperability of Capture Devices
• Different wavelengths of illuminating light
25
Iris Imaging with user Cooperation
• Example devices
26
Iris Imaging with (almost) no user
Cooperation
1. Use of multiple-resolution cameras (Sarnoff Corp., USA)
– wide-angle camera for face detection
– narrow-angle camera(s) for iris capture
2. Use of deformable mirrors (AOptix, USA)
– idea used before in astronomical telescopes compensating the deformations
introduced by the atmosphere
– Zernike polynomials used to describe the deformations of the mirrors (in optics
used for describing aberration of the lens)
– fast (a fraction of a second) and at-the-distance capture (typically 1.5m – 2.5m)
27
Image Acquisition Challenges
28
Image Acquisition Challenges
29
Image Acquisition Challenges
30
Iris Image Pre-processing
1. Enhancing image structure: removing sensor noise,
compensating interlaced imaging, specular reflections
removal
2. Enhancing image intensity: contrast enhancement,
histogram equalization, linear trends removal
31
Iris Image Enhancement
Steps
• Segmentation
– Keep only useful information (iris texture).
• Normalization
– Make different captures of the same iris look as
similar as possible.
32
Iris Image Enhancement
• Segmentation (1/2)
– Iris and Pupil
Localization
• Localize limbus and
pupillary boundaries.
– Eyelid, Eyelash, and
Specular Reflection
Detection
• Deal with iris texture
occlusions.
33
Iris Image Enhancement
Segmentation (1/2)
• Iris and Pupil Localization
– Method 1: Integral-differential
operator
• Objective: Find (r, x0,y0) of limbus and
(r, x0,y0) of pupillary boundaries.
• Strategy: Try various values for (r,
x0,y0)
– Method 2: Image processing ending
with Hough circle transform.
34
Iris Image Enhancement
Segmentation (1/2)
• Eyelids, eyelashes, specular
highlights
– Fit of parabolic curves for eyelids.
– Active contours (curve evolution)
to avoid eyelashes.
– Fit of elliptical curves for specular
highlights.
– Machine learning from annotated
examples.
35
Iris Image Enhancement
Normalization (2/2)
36
Iris Image Enhancement
Normalization (2/2)
37
Iris Image Enhancement
Limitations
Segmentation
• Pupil and iris are not concentric (pupils are slightly shifted to the
nasal corner).
• They are not perfectly round.
Normalization
• Forcing circular models may lead to poor mapping.
38
Iris Feature Extraction
39
Iris Feature Extraction
Zero-Crossing Approach (1/3)
• Proposed by W. W. Boles.
• Iris image is treated as a 1D
signal (iris signature).
1. Iris signature is filtered by
Laplacians of Gaussians (LoG)
(second derivative of
Gaussian).
40
Iris Feature Extraction
Zero-Crossing Approach (1/3)
2. Zero-crossings lead to bits up; everything else is
zero.
41
Iris Feature Extraction
2D-Gabor Filtering Approach (2/3)
• Proposed by John Daugman.
• De facto iris description
solution.
• More complete and robust than
zero-crossing.
• 2D Gabor filters are convolved
with the normalized iris image
42
Iris Feature Extraction
2D-Gabor Filtering Approach (2/3)
• Gabor wavelets are a good model of neural receptive
fields found in the visual cortex.
• Empirical selection of a proper Gabor wavelet
(adequate to encode iris texture).
43
Iris Feature Extraction
2D-Gabor Filtering Approach (2/3)
44
Iris Feature Extraction
2D-Gabor Filtering Approach (2/3)
45
Iris Feature Extraction
BSIF Approach (3/3)
• Binarized Statistical Image Features (BSIF)
• General-purpose local image descriptors designed for
texture encoding.
• Examples of textures that one might want to describe.
46
Iris Feature Extraction
BSIF Approach (3/3)
• Subspaces of representative image
patches (further used as filters) are
learned from a set of example
patches through Independent
Component Analysis(ICA).
– ICA: N filters of size lxl are estimated
from examples by maximizing their
mutual statistical independence.
47
Iris Feature Extraction
BSIF Approach (3/3)
• Images are convolved with each
BSIF filter leading to various
projections in the target subspace.
• BSIF code: a threshold is used to
make the image projections
binary; anything above zero is
ONE, everything else is ZERO.
48
Iris Feature Extraction
BSIF Approach (3/3)
• In the case of irises…
• Solution's
performance is on
par with the Gabor-
based one.
49
Iris Feature Matching
How to compare binary codes
50
Iris Feature Matching
How to compare binary codes
Problems (1/2)
• How to consider iris masks?
51
Iris Feature Matching
How to compare binary codes
52
Iris Feature Matching
How to compare binary codes
Problems (2/2)
• How to deal with iris rotations?
• They happen when heads are tilted…
How to match Iris rotations
53
Iris Feature Matching
How to compare binary codes
Problems (2/2)
• How to deal with iris rotations?
• Solution: provide different shifts for one of the iris
codes
• Compute various normalized Hamming distances
(one for each shift).
• Take the smallest distance as the score. 54
Retina Recognition
55
Introduction
• Retina is the lining layer at the back of the eyeball that covers
65% of the eyeball’s inner surface.
• It contains photosensitive cells.
• Each person’s retina is unique due to the complex network of
blood vessels that supply blood.
• It is a reliable biometric as the retina pattern remains
unchanged throughout the person’s life, barring the patterns of
persons having diabetes, glaucoma, or some degenerative
disorders.
56
Retinal Scanning
• In retinal scanning process, a person is asked
to remove lenses or eyeglasses.
• A low-intensity infrared light beam is casted
into a person’s eye for 10 to 15 seconds.
• This infrared light is absorbed by the blood
vessels forming a pattern of blood vessels
during the scan.
• This pattern is then digitized and stored in the
database.
57
Retinal Scans
• Retinal scans measure the pattern of blood vessel in
retina or the back part of the eye.
• Blood patterns in retina are not genetically affected,
thus resulting in being completely unique from one
individual to another, even within twins.
• The device that measures these patterns requires a
user to stand still within inches of the device which
has a shining light source.
58
Retina Features
• Physical features
– vessels: blood transfer
– optic disk: the nerves of eye connected to brain
• Pathological features
– red lesion
– bright lesion
59
Advantages of Retina Biometrics
• accuracy
• stability of biometrics sample
• resistant to fraud
• small template
60
Challenges/Disadvantages of
Retinal Scans
• The scanning device is rather expensive
• In addition, the retinal blood patterns can change with body
condition and diseases, such as high blood pressure, pregnancy,
cataracts, astigmatism and AIDS
• Difficulty to use
• Consumer perceptions
• Static design
• Not convenient (not user friendly) if you wear glasses or is in close
contact with the reading device
61
