INTRODUCTION
The conventional identification system is based on token and knowledge, such
as key, password, magnetic tape and chips, etc. However, all of these can be
embezzled, forgotten or forged and hence password and tokenbased recognition
systems are nowadays supplanted by a biometric recognition system. Even in some
systems where password and token are still used, on top of it a biometric layer
is added for more secure authentication. Hence, there is a tremendous growth
in biometric based identification system when compared to past decade. Biometric
includes reference to the measurement, analysis, classification, science of
personal recognition and verification or identification by using distinguishable
biological (physiological) or behavioral trait, features or characteristic of
that person. Biometric identification is the process of associating an identity
of the input biometric data by comparing it against the enrolled identities
in a database (Jain et al., 2004). Some of the
commonly used physical or behavioral traits are: Face, fingerprint, palm print,
hand geometry, iris, keystroke, signature, voice and gait. Apart from these,
there are also some traits which are used to identify a person uniquely based
on odour, tongue, ear, knuckle print.
Among the various biometric traits, ‘Iris’ is proved to be a predominant
and popular biometric feature for authentication and identification with a high
degree of accuracy and distinctiveness. It is also proved for its reliability
and universality. The finger print has also received substantial concentration
and is being effectively used in many applications over one hundred years. Face
and speech have also been extensively studied over 25 years, whereas, ‘Iris
recognition’ is a newly emerging trait in personal identification and is
reported to be the most reliable biometric trait in identification and authentication
(Daugman and Downing, 2001). For its biological development
and the features present in the iris is proved to be unique and stable throughout
the life time of the person (Wildes, 1997). The performance
of the identification system depends more on the search space and the accuracy
rate. Existing systems are till now smallscaled, databases and henceforth identifying
an imposter is not a difficult task but as the size grows the unique identification
becomes cumbersome.
MATERIALS AND METHODS
Anatomy of human eye and iris: The human eye is an asymmetrical globe
which measures nearly an inch in diameter. The front view of the human eye (the
part what we see) as shown in Fig. 1 includes:
• 
Pupil (center of the eye) 
• 
Iris (the pigmented darker part) 
• 
Cornea (a dome structure over the iris) 
• 
Sclera (white part) 
• 
Conjunctiva (layer of tissue covering the eye) 
The lens is placed behind the iris and the pupil which is responsible for the
focus of light at the back of the eye. Major part of the human eye is packed
with vitreous which is a clear gel. The light passes to the back of the eye
through the pupil and then to the lens. The retina is the group of sensing cells
which converts the received light into electrical impulses. The impulses are
carried to the brain by the optic nerves which are behind the eye. A small extrasensitive
area called the macula is present within the retina that gives central vision.
It is positioned at the center of the retina and contains a small depression
or pit, the fovea at the center of the macula that gives the clear sight. Parts
of the eye and their functions are presented in Table 1.
Iris is a Greek word means goddess of the rainbow due to the many colors of
the iris and hence the name. The iris is a thin circular ring region, a part
of the human eye positioned between the black pupil and white sclera presenting
a unique and rich texture information, such as patterns, rifts, colors, rings,
spots, stripes, filaments, coronal, furrows, minutiae and recess and other detailed
characteristics seen under the infrared light.
Table 1: 
Function of human eye 

The iris is composed of loosely woven soft tissues. The iris are responsible
for the amount of light reaching the retina by controlling the diameter and
size of the pupil which is hole a located in the center of the iris. The Iris
is the only internal organ which seen from outside. Further, it is a measurable
internal physical feature which is unique and cannot be easily altered due to
ecological effects. The probability of finding any two eyes identical, are approximately
1 in 10^52. It has nearly about 250 points of identification for comparison.
The minute iris patterns and textures which is the phenotypic feature are determined
during the foetal development and not inherited from the gene and for this reason
the right and left eye of a person are not alike and same even for the identical
twins. It has been proved that the color of the eye and the appearance of the
eye are genetically inherited (genotype) but the texture patterns are not inherited
(epigenetic).
The color of the iris, can be green, blue, or brown. In some cases it can be
hazel (a combination of light brown, green and gold), grey, violet or even pink.
Although there are a variety of colors in the iris the dark pigment melanin
is responsible for the ranges of colors in the iris. The amount of melanin is
the deciding factor in the iris color. The color of the iris is brown or black
which is due to the quantity of melanin when it is more. The red or yellow is
because of pheomelanin caused by melanocytes which is found less in blue and
green color iris. If it is a moderate quantity of melanin then it is hazel.
In brown color iris the surface of the iris is smooth and is heavily pigmented.
With Blue and hazel color iris the surface is irregular with crypts. The color
of the eye may be influenced by more than 50 genes and hence the genetics of
eye color are more complex than what they appear (Schwab,
2012).
Iris originates in the sixth week of gestation period. At end of the third
month the ciliary fold are formed and by the completion of the fourth month
the vascular layers of the iris is formed. During the fifth month the pupillary
membrane, the iris stroma and sphincter muscle are formed. The development is
completed by the end of eight month but the stroma layer is very thin, the cellular
framework is incomplete and the collarette is very nearer to the pupil in pupillary
region. During this period the patterns and structures are complete but the
pigmentation process is carried out until one year after the birth (Berggren,
1985). The size of the eyeball of a newborn baby is about 18 mm starting
from front to back. The eye grows until the age of two and increases in length
to 2425 mm for an adult approximately. The diameter of the iris is 1112 mm,
the radius is 5.5 mm approximately, circumference is about 38 mm and thickness
is 0.6 mm at the pupillary margin which is collarette and at the ciliary margin
the thickness is 0.5 mm. The pupil which is a hole in the center (so called)
occupies onethird of the iris say about nearly 5 mm in diameter and 2.5 mm
radius. The edge of the pupil is the collarette which is 1.5 mm away from the
pupil as shown in Fig. 2.
The iris is a multilayered structure consisting of anterior layer and posterior
layer (Simon et al., 1979). The anterior layer
is visible and the posterior layer is invisible.

Fig. 2: 
Pupil, collarette and iris boundary 
The anterior layer is larger and lies between the cornea and the iris. The
posterior layer is smaller and lies between the iris and the lens. The anterior
layer consists of the pupillary zone and the ciliary zone. The pupillary zone
and the ciliary zone is separated by the collarette which is a circular ridge.
The pupillary zone is relatively flat in the iris and the width varies with
the degree of atrophy of the anterior border layer. The ciliary zone contains
radial interlacing ridges giving an appearance of gossamer. The iris crypts
are formed due to the irregular atrophy in the ciliary zone. The anterior border
layer of the iris is a loose collagen tissue in which cells are densely packed.
The posterior layer is a heavily pigmented epithelial cell.
Proposed geometric features: The geometric features of the iris can
also be used for human identification. The structure of the iris is more related
to the geometric shape and hence the extraction of these features is also possible.
The idea of including iris as biometric feature for human recognition is long
back hundred years ago (Bowyer et al., 2008),
but the computerization of iris recognition system is very young than other
biometric system. Dr. Frank Burch an ophthalmologist in the year 1936 recommended
iris for the human identification. After a long time in 1985 Dr. Leonard Flom
and Aran Safir identified that the two irises of the same person are unique
and not same. Then after in 1993 Dr. John Daugman proceeded with the novel approach
of Dr. Flom and Safir and effectively automated the iris recognition system
in 1994. The Daugman’s approach (Daugman, 2004)
is based on the frequency domain and he used Integrodifferential operator to
detect pupil and 2D Gabor filter phase modulation to extract iriscode. The
iriscode was matched using hamming distance. Wildes (1997)
approach used Hough transform and gradient edge detection method for pupil detection
and Laplacian pyramid for iris image analysis. Boles and
Boashash (1998) uses zerocrossing method with dissimilarity functions of
matching. The reader can have a thorough review of iris recognition in Burge
and Bowyer (2013) study. A precise pupil center and radius is extracted
in the study of Xu and Shi (2006). To extract the pupil
feature Hough transform is used and a weight ring mask is estimated. In the
study of Mohammed et al. (2010) the pupil feature
is extracted from gray scale image for identification. The Angular Integral
Projection Function (AIPF) method is used to accurately detect the center of
the pupil and then the boundary points are computed. In the study of He
and Shi (2005), feature extraction was based on Daugman approach. The crossed
chord segmentation method was used for iris segmentation and zigzag collarette
was detected for identification. The geometric based feature in the previous
study is used only for medical analysis (Ma et al.,
2013). In this study, we had attempted to use these geometric features for
identification purpose. The geometric features which are extracted can be used
as a primary level of identification and some can be used as secondary, for
example, the size of the pupil may not be unique but when it is associated with
other features say iris or collarette it becomes unique. The geometric characteristics
are classified as:
• 
Pupil based features 
• 
Collarette based features 
• 
Iris based features 
Pupil based feature: Pupil is defined to be circular or round but they
are not actually a circle. So, the geometrical properties of the circle can
be used to identify the pupil and these features can be used for recognition.
The following are the pupil based features:
• 
Pupil Roundness (PR) 
• 
Pupil Largeness (PL) 
• 
Pupil Smoothness (PS) 
Pupil Roundness (PR): The roundness measurement requires to trace the
iris in 360°C and hence the consistency of the diameter can be measured
at a number of orientation points on the iris. In general roundness is defined
as a measure of sharpness at the corners and edges of a circle which is related
with the sphericity and compactness of the circle. In simple it is the radii
around the iris at regular intervals. The roundness feature is calculated based
on the diameter at various orientations in Ma et al.
(2013), where the ratio between the maximum value and the minimum value
is calculated at horizontal (h), vertical (v), left (l) and right (r) at 0,
45, 90 and 135°, respectively:

Given a circle the roundness value is 1.0 for a standard circle, for a noncircular
it is less than 1.0 and greater than 1.0 for an irregular shape. Even though
the constant diameter is one of the necessary conditions for roundness, it is
not sufficient. In this proposed approach we used the assessment procedure Least
Square Circle (LSC) which popularly used in the field of metrology to find the
roundness of an object. The LSC fits the data at equal spaced angles in the
circle which estimates PR, the noncircularity where R is the radius of the
circle and P is the distance from the center of the circle to the point on the
path of the circle. The center of the circle is at a point (a, b) in the circle.
The radius (r_{i}) is measured at each angle in 360°. The LSC (y_{i})
is calculated as:
The minimum difference between the LSC and the measured trace point is defined
as:
Hence, the deviation from the LSC is calculated from Eq. 3
and calculate PR as:
where, I = {1, 2, 3, …., n} and n is the number of measure points on the
circle.
Pupil Largeness (PL): The largeness of the pupil is described based
on the radius of the pupil. As the pupil is not an actual circle the general
diameter calculation of distance from the center to any point on the pupil is
not sufficient. The diameter is calculated at eight direction on the pupil at
0° (horizontally), 90° (vertically), 30, 45, 75, 120, 135 and 165°.
Hence, the Pupil Largeness (PL) is defined as:
where, dp is the diameter of the pupil.
Pupil Smoothness (PS): The smoothness of the iris can be represented
based on the curvature of the circle. The curvature (K) of the circle is defined
as the reciprocal of the radius:
The curvature of a straight line is 0. As the pupil is considered to be a circle,
in this study we have adopted the curvature of circle to measure the pupil smoothness
feature (PS):
where, ρk is curvature and
is the mean curvature.
The is calculated at the points (X_{1}, X_{2}, X_{3},
....X_{k}, ....X_{K}, X_{1}) and measured between the
angles (X_{(k+3)} mod KX_{k}) and (X_{(k3)} mod KX_{k})
the 3rd order curvature is used. The curvature ρk is:
Geometrically the curvature is define as how fast the unit tangent vector to
the curve rotates. If the a curve keeps close to the same direction the unit
tangent vector change very little and the curvature is small whereas the curve
undergoes a light turn the curvature is large.
Collarette based feature: The collarette area is insensitive to pupil
dilation and not affected by eyelid or eyelash. The collarette is concentric
with the pupil and the radius of the area is restricted certain range. The collarette
can be defined as a snaky scalloped line that splits the iris ancillary zone
and ciliary zone. The collarette is restricted to the inner half of the iris
and contains ether radial spokes or dots. As the collarette based feature cannot
be detected directly from the iris it requires some first to detect the collarette
and then compute the geometric measures. In the proposed method uses the adjusted
angle zigzag sampler to detect the collarette (Strindberg
and Buckland, 2004). In the given image the collarette along the line is
detected. Here the height of the collarette is denoted as h_{s} (x)
where s is the sampler at the location x along the axis and defined as:
In this case the coverage probability is denoted as π (u) which varies
spatially. The vertical component at the point Y, is also used for coverage
probability and is π_{x} (y). For horizontal component it is defined
as:
Hence, by using the horizontal component and vertical component it is easy
to estimate the coverage probability. The minimum length of the boundary can
be obtained as:
where, x_{max} is the minimum value of x within the boundary and x_{min}
is the maximum value of x within the boundary. In the adjusted angle zigzag
line detector the sample height h_{s}(x) is proportional to the height
H(x). Then π_{x}(x) = π_{c} and the mean coverage
probability is calculated as:
Hence, the angle θ(x) is continuously adjusted proportionally to the height
H(x). Along the x axis the angle sampler is defined as:
Where:
where, R_{s} denotes the sample region R.
The algorithm to find ROI_2 the zigzag collarette is as follows:
• 
Select a point randomly at the collarette
region and obtain the θ(x), where: 
• 
Determine the angle of the sample along
xaxis using θ(x), find 180°θ(x) 
• 
Along the x axis extend the sampler from the selected
point in both direction by continuously adjusting the angle with respect
to θ(x) 
• 
When the upper or lower boundary of R is intersected
and the length at x coordinate is obtained 
• 
Continue the above steps until the complete collarette
zone is covered 
Collarette Roundness (CR): As the collarette feature is extracted from
the iris, the next step is working out of geometric feature of the collarette.
The roundness feature of collarette is computed as the ratio between the maximum
and minimum values of the six diameter of the collarette.
where, dc is the diameter of the collarette at various direction.
Collarette Iris Ratio (CIR): The collarette iris ratio is computed based
on the diameter of the iris. The diameter of iris is nearly 1011 mm. The collarette
occupies one third of the iris and is located at about 1.5 mm away from the
pupil. Hence the distance of the collarette and iris diameter is calculated
at eight points, where the iris is divided into equal space ordinates with respect
to the polar coordinates. For each intersection point on the edge of the iris,
the distance is computed and the collarette iris ratio is computed as:
where, dc is the diameter of the collarette and di is the diameter of the iris.
Collarette Pupil Ratio (CPR): The collarette pupil ratio is calculated
with respect to the collarette and pupil. The same formula used for the CIR
can be used to compute CPR but based on the pupil.
where, dc is the diameter of the collarette and dp is the diameter of the pupil.
Iris based feature: In the proposed work the geometric feature of iris
is used for recognition rather than the texture features of iris. The following
are the iris geometric features used for recognition:
• 
Iris Roundness (IR) 
• 
Iris Diameter (ID) 
Iris Roundness (IR): The roundness feature of the iris is calculated
same as the pupil using the Least Square Circle. The deviation from the LSC
is calculated from Eq. 3 and the IR is calculated as:
where, i = {1, 2, 3, …., n} and n is the number of measure points on the
circle.
Iris Diameter (ID): This feature is calculated based on the average
of diameter at eight point on the iris. It is known that the iris is not an
actual circle the general diameter calculation of distance from the center to
any point on the outer circle of the iris is not sufficient. Hence, the average
diameter of the iris is calculated at eight direction on the pupil at 0°
(horizontally), 90° (vertically), 30, 45, 75, 120, 135 and 165°. Hence,
the Iris Diameter (ID) is defined as:
where, dp is the diameter of the pupil.
Feature fusion and matching: After the iris geometric feature extraction,
the iris is represented as a feature vector. The extracted three features are
fused into single feature vector as:
Feature vector = {PR, PL, PS, CR, CIR, CPR, IR, ID}
where, PR is Pupil roundness, PL is pupil Largeness and PS is Pupil Smoothness
CR is Collarette roundness, CIR is Collarette Iris Ratio and CPR is Collarette
Pupil Ratio IR is Iris Roundness and ID is the Average Iris Diameter.
This system is operated at two modes, say identification and verification.
During the identification the iris features are extracted and stored in the
database. Later at the verification stage the iris features are extracted and
matched with the database. As the iris is saved as feature vector, the Euclidean
distance measure is used for matching.
RESULTS AND DISCUSSION
The proposed approach is tested on CASIA and Real Time Database. The first
database is publicly available dataset CASIAIris V3 which consists of 22,035
images from more than 700 subjects. The other is our own real time dataset collected
from 800 subjects which consist of 1,600 images. The iris for the real time
database was collected environment, where the lighting was maintained and clear
instructions were given to the subject in all respect. The performance of the
iris recognition system is evaluated using False Acceptance Rate (FAR) and False
Reject Rate (FRR).
False accept rate: The frequency of access with wrongful claims of identity
or nonidentity that are falsely accepted by the system. The FAR is defined
as:
where, FA is the number of false access.
False reject rate: The frequency of access with truthful claims of identity
or nonidentity that are falsely rejected by the system. The FRR is defined
as:
where, FA is the number of false access, FR is the number of false reject and
n is the total number of access. The system performance can be represented by
Receiver Operating Curve (ROC) curve. The ROC curve visualizes the relationship
between FAR which is on the xaxis and FRR, the yaxis for the various threshold
value. The system was experimented as monomodal and multimodal. From Table
2 it is apparent that the system gives better recognition rate when the
features are fused rather than as a mono model. In Table 3
the proposed geometric fused feature is compared with the existing texture based
iris recognition and the results proved to have a better accuracy rate.
The proposed method is compared with that of Wildes (1997)
and Chen and Yuan (2003). Table 4
shows the accuracy rate against the proposed work and shows a better recognition
rate than the existing methods.
As the system is implemented as a monomodal the results are compared as single
feature with that of the fused feature as shown from Fig. 38
shows the performance of the system with a comparison of collarette and pupillary
features. The features are extracted for the CASIA database and the real time
database.
Table 2: 
Comparison of single feature and proposed fused features 

Table 3: 
Comparison of texture feature and proposed features 

Table 4: 
Comparison of existing methods with proposed method 


Fig. 3: 
ROC curve for iris geometric features and proposed fused features 

Fig. 4: 
ROC curve for pupil geometric features and proposed fused
features 

Fig. 5: 
ROC curve for collarette feature and proposed fused features 

Fig. 6: 
ROC curve for all geometric features and proposed fused features 

Fig. 7: 
ROC curve for texture feature and proposed fused features
for CASIA DB 

Fig. 8: 
ROC curve texture features and proposed fused features for
Real time database 
CONCLUSION
The study presented in this study investigates the feasibility of using geometric
features of iris in biometric identification system. The iris features were
used for medical analysis and therefore this study is an initiation made using
geometric features for human identification. Three different features were extracted
and matched. The first feature is based on the pupil’s roundness, largeness
and smoothness features; the second set of features are based on collarette’s
roundness, collarette iris ratio and collarette pupil ratio and the third feature
is based on iris roundness and diameter. These features of iris was extracted
and used for identification as these features are not affected by occlusion
such as eyelash or eye lid. The proposed approach was first experimental and
the results were analyzed as a single feature. Later the three features were
fused as single feature and were matched. The experimental result with this
fusion technique proved a better identification than as a single feature and
time taken to match is also less.