Biometric Iris Recognition System
INTRODUCTION Iris recognition is fast developing to be
a foolproof and fast identification technique that can be administered cost effectively. It is a classic biometrics application that is in an advanced stage of research all over the world.
Biometric Iris Recognition System
Iris localization and alignment procedure
Image acquisition
Original iris image present in the database
Formation of a circular contour around the iris
The circular contour is concentric with the circular pupil
Removing the portion of the iris occluded by the eyelids
The ratio of limbus diameter and pupil diameter forms the first criterion in the comparison of
any two irises.
First step involves determining the limbus diameter
Determining the pupil diameter
The ratio of the limbus diameter and the pupil diameter is determined which is an important criterion in the identification/comparison of the irises.
Pattern matching
To decide if this pattern matches with the one existing in the database.
Multi-scale representation of the iris is achieved using the pyramid algorithm of Laplacian of Gaussian filters (4 levels) employing down sampling in the Gaussian and up sampling in the Laplacian of the Gaussian, filtered image.
Construction of the Laplacian Pyramid begins first with convolution of the Iris Image with LoG Mask ‘W’, so as to yield low-pass Gaussian filtered images gk The expression is as follows: gk=(W*gk-1)↓2,
g1–g4—low-pass Gaussian images each obtained after filtering the previous image and down sampling by 2
The LoG filter can be specified as
where σ—standard deviation of the Gaussian, ρ—radial distance of a point from the filter's center.
A discrete approximation which is derived from the above filter expression.
W = [1 4 6 4 1]/ 16
Laplacian pyramid lk
Is formed as the difference between gk and gk+1
lk=gk-4W*(gk+1)↑2,
gk+1 expanded before subtraction so that it matches the sampling rate of gk
The expansion is accompanied by up sampling and interpolation:Up sampling is achieved by the insertion of zeros between each row and column of the down sampled Gaussian image. The Mask ‘W’ is used as an interpolation filter and the factor 4 is necessary because 3/4 pixels are newly inserted zeros
We then evaluate the degree of match between the acquired image and images from the database. The approach taken is to quantify for the degree of match using normalized correlation between the acquired image and the images from the database.
Normalized correlation between the acquired
image and the images from the database Normalized correlation can be defined in discrete form as follows
Let lk1[i,j] and lk2[i,j] be the two iris images of size r×c (rows×columns)
Normalized correlation between lk1 and lk2 can be defined as
Pattern Matching
Result obtained from this iris recognition system for iris images of two different persons
Corr0=0.6625—Normalized correlation value for the 1st Laplacian of Gaussian of the 2 iris images
Corr0=0.8101—Normalized correlation value for the 2nd Laplacian of Gaussian of the 2 iris images
Corr0=0.8513—Normalized correlation value for the 3rd Laplacian of Gaussian of the 2 iris images
Corr0=0.9284—Normalized correlation value for the 4th Laplacian of Gaussian of the 2 iris images
Both the irises are not identical
• Result obtained from this iris recognition system for iris images of same person
• Corr0=1.0000—Normalized correlation value for the 1st Laplacian of Gaussian of the 2 iris images
• Corr0=1.0000—Normalized correlation value for the 2nd Laplacian of Gaussian of the 2 iris images
• Corr0=1.0000—Normalized correlation value for the 3rd Laplacian of Gaussian of the 2 iris images
• Corr0=1.0000—Normalized correlation value for the 4th Laplacian of Gaussian of the 2 iris images Both the irises are identical
Both the irises are identical
Pattern Matching 2
Result obtained from this iris recognition system for iris image of a person with the iris image of the same person but modified by
incorporating 3 dots in the iris region
Corr0=0.9953—Normalized correlation value for the 1st Laplacian of Gaussian of the 2 iris images
Corr0=0.9990—Normalized correlation value for the 2nd Laplacian of Gaussian of the 2 iris images
Corr0=0.9988—Normalized correlation value for the 3rd Laplacian of Gaussian of the 2 iris images
Corr0=0.9998—Normalized correlation value for the 4th Laplacian of Gaussian of the 2 iris images
Both the irises are not identical
Conclusion
• The systems considers the ratio of limbus to pupil diameter as the initial criterion for recognition, which saves the processing time involved in pattern matching and calculating the correlation when the two ratios are not equal.
•The pattern matching technique employs multiscale pyramid representation using a LoG filter, which provides optimal enhancement of the features of the iris patterns, and then normalized correlation is employed for evaluating the degree of similarity and has been shown to give accurate results.