Biometric Identification
Francesc Serratosa
Iris recognition
MESIIA – Master’s Degree in Computer Security Engineering and Artificial Intelligence
Part 5
Iris recognition
Biometric Identification
Francesc Serratosa
[Link]
[Link] 2
Iris Identification
Application of Iris Identification
Iris Identification
Iris Recognition Revenues
IBG: International Biometric Group
Contents
1. Iris anatomy
2. History of biometrics based on iris
3. The Daugman method
4. Eye image acquisition
5. Iris localisation
6. Polar representation of Iris image
7. Iris code generation
8. Iris matching
5
Bibliography
Basic bibliography
[1] J. Daugman, How Iris Recognition Works, Transactions on circuits and systems for
video technology, 14(1), 2004.
[2] Daugman’s patent: Numer: 5.291.560
Other bibliography
[3] Kaushik Roy & Prabir Bhattacharya, Iris Recognition. A Machine Learning Approach,
Editorial Verlag Dr. Muller, 2008.
6
1. Iris anatomy
Observable structure
1. Iris anatomy
Features of Iris Structure
1. The structure of iris is unique to an individual
1. The main structure is genetically determined
2. The small particularities are dependent on the embryonic conditions
2. The structure is almost stable with age
1. Vasculature: They are largely in place at birth
2. Musculature: They mature around two years of age
3. Pigmentation: It changes until adolescence
4. Pupil size: It increases slightly until adolescence
5. Following adolescence: The iris is very stable except for some illnesses
3. Liveness
1. Moment dynamics: Hippus
1. The diameter of the pupil is in constant state of oscillation (0.5 Hz)
2. Fast reaction to the light (0.1 seconds)
2. History of biometrics based on iris
History of Iris Recognition
• In 1936, Frank Burch, an ophthalmologist, pointed out that iris pattern is
unique and may be used for personal identification.
• In 1987, Aran Safir and Leonard Flom, who are also ophthalmologist, proposed
the idea of automatic iris recognition and applied US patent.
• In 1991, Johnson from Los Alamos National Laboratory reported to actually
realize a personal identification system based on iris recognition.
• In 1993, John Daugman proposed a successful iris recognition method, which
is widely used in commercial products now.
• In 1996, Richard Wildes developed another iris recognition system, including
both iris image acquisition and recognition algorithm.
• In 2000, CASIA developed the first iris recognition system in China.
[Link]
3. The Daugman method
Main scheme
3. The Daugman method
Main scheme
Acquisition Image
Localization
IrisCode Gabor Filters Polar Representation
Matching
Demarcated Zones
···
4. Eye image acquisition
Iris Acquisition
Principal aspects on acquisition:
-High resolution and sharpness
-Iris diameter (1 cm) : From 100 to 200 pixels
-Distance of acquisition: 15 to 75 cm
-Good contrast in the pattern
-While restricting illumination to be within limits of the eye safety
-Infrared light (Monochrome):
-Less invasiveness and greater contrast
-Easy to eliminate reflections
-Image has to be centred and well oriented
-Without requiring the subject to employ an eyepiece or chin rest
-Feedback with real image
4. Eye image acquisition
Iris Normalisation
Four geometric degrees of freedom to be compensated:
-Shift: Accounts for offsets of the eye in the plane parallel to the camera
-Scaling: Accounts for offsets of the camera optical axis
-Rotation: Accounts for deviation in angular position
-Pupil dilatation: Accounts for the non-linear deformation of the pupil due to light
Compensations for:
-Shift and Scaling: Centre the pupil in the axis of the image
-Rotation: It is compensated for by brute force search.
Explicitly shifting an iris signature in q.
-Pupil dilatation: Several templates with different light conditions
4. Eye image acquisition
Non-linear Iris Deformation
4. Eye image acquisition
Iris Liveness Detection
To analyse the response of the pupil radius to illumination through
controllable visible lighting
4. Eye image acquisition
Iris Image Quality Assessment
• Can only classify defocused images:
• Dependent on illumination and contrast
• Defocused images: Low frequency response
• Eyelashes, hair and dirt on eyeglasses: high-frequency response
Frequency:
In the middle of
the image:
horizontal and
vertical
frequency =
(0,0)
4. Eye image acquisition
Iris Image Acquisition Devises
4. Eye image acquisition
Iris Image Acquisition Devises
Panasonic (Japan)
4. Eye image acquisition
Iris Image Acquisition Devises
LG Electronics (Korea)
4. Eye image acquisition
Iris Image Acquisition Devises
OKI (Japan)
4. Eye image acquisition
Iris Image Acquisition Devises
Eyeticket (USA)
4. Eye image acquisition
Iris Image Acquisition Devises
Sarnoff (USA)
4. Eye image acquisition
Iris Image Acquisition Devises
Iridian (USA)
5. Iris localisation
Convolution of an image given a mask
Suppose we have a 3x3 convolution mask: Mask
And an image, which is much larger: Input image
The convolution is a process to generate another image: Output image
The mask is slid over the image, manipulating a square of 3x3 pixels at a time.
In the example, b22 in the Output image is deduced by the equation below
5. Iris localisation
Iris Localisation
Based on Convolution the following masks given several different radius.
There is a different pair of masks per radius.
The radius are deduced by the convolution that generates the minimum value.
First, the pupillary border is detected and then the limbic border is detected.
White pixels in the mask represent mij=-1
Black pixels in the mask represent mij =1
Grey pixels in the mask represent mij =0
Mask to detect the pupillary border Mask to detect the limbic border
6. Polar representation of Iris image
Angular Conversion
Input: rs, rp, x0, y0, N, M
α = atan2(x-x0, y-y0)Î[π, -π]
rs: Sclera Radius
rp: Pupil Radius r = sqrt((x-x0)2 + (y-y0)2)Î[0, ∞)
(x0 ,y0 ) x’ = (M-1)·(α+ π)/2 π Î[0, M-1]
y’ = (N-1)·(r-rp)/(rs-rp) Î[0, N-1]
y (x,y)
(0 ,0)
x
y’ (x’,y’) M
N
x’
6. Polar representation of Iris image
Angular Conversion
Problem:
Not all pixels in the new image are filled
with information
6. Polar representation of Iris image
Angular Conversion
Yp
α = [2πx’/(M -1)]- π
r = rp+ [(rs-rp)· y’]/(N-1)
rs: Sclera Radius
rp: Pupil Radius
x = r·sin(α)+x0
(x0 ,y0 ) y = r·cos(α)+y0
y (x,y)
(0 ,0)
x
y’ (x’,y’) M
N
x’
7. Iris code generation
Feature Extraction
Gabor Filter Iris Image Gabor Filter
Real part Imaginary part
Convolution Convolution
K1 K2
If Real(Conv(I,G)>0 à K1=1 If Ima(Conv(I,G)>0 à K2=1
Else K1=0 Else K2=0
7. Iris code generation
Feature Extraction
7. Iris code generation
Gabor Filter
Expression of the real part of the Gabor Filter in the spatial domain :
x, y: Spatial position of the filter
λ: Wavelength of the cosine factor
θ: Orientation of the filter
ψ: Phase offset
σ: Sigma of the Gaussian envelope
γ: Spatial aspect ratio
8. Iris matching
Iris Matching
Normalisation &
Cropping
Feature Extraction Angular Conversion
Matching with rotation
8. Iris matching
Matching with Rotation
c M-c
X (Iris Code) N
Y (Iris Code) N
M -1 N -1 c -1 N -1
D( X , Y , c ) = åå ( ( ) ( )) åå ( ( ) ( ))
2 2
X x , y - Y x - c , y + X x , y - Y M - c + x , y if M >c³0
x =c y =0 x =0 y =0
M −1 N−1 −c−1 N−1
2 2
D ( X,Y, c) = ∑ ∑(Y ( x, y) − X ( x + c, y)) + ∑ ∑(Y ( x, y) − X ( M + c + x, y)) if −M <c<0
x=−c y=0 x=0 y=0
3. The Daugman method
Main scheme
Acquisition Image
Localization
IrisCode Gabor Filters Polar Representation
Matching
Demarcated Zones
···