The project report details the development of an iris-based human recognition system, designed to accurately identify individuals using unique iris patterns as biometric parameters. The system involves several key processes including segmentation, normalization, feature extraction, and matching, utilizing the CASIA database for testing and validation. Overall, it aims to enhance security and reliability in identification systems amidst increasing concerns regarding privacy and authentication.

1. A
PROJECT REPORT
ON
“IRIS BASED HUMAN
RECOGNITION SYSTEM”
SUBMITTED BY:
MR. DESAI GAURAV S.
MR. JOSHI SUSHIL D.
MS. IYER ANANDHI G.
MR.WAZALWAR DHAWAL S.
(BE. EXTC)
GUIDED BY:
MR. R.R. MANTHALKAR
SGGS INSTITUTE OF ENGINEERING AND
TECHNOLOGY, VISHNUPURI, NANDED
SWAMI RAMANAND TEERTH
MARATHWADA UNIVERSITY, NANDED
1

2. CERTIFICATE
THIS IS TO CERTIFY THAT THE PROJECT REPORT
ENTITILED
“IRIS BASED HUMAN RECOGNITION SYSTEM”
SUBMITED BY
MR. DESAI GAURAV S.
MR. JOSHI SUSHIL D.
MS. IYER ANANDHI G.
MR.WAZALWAR DHAWAL S.
(BE. EXTC)
HAS BEEN COMPLETED AS PER THE REQUIREMENT OF
“SWAMI RAMANAND TEERTH MARATHWADA
UNIVERSITY, NANDED”.
IN PARTIAL FULFILMENT OF THE DEGREE
B.E. (Electronic and Telecommunication Engineering)
FOR THE ACADEMIC YEAR 2006-2007.
GUIDE HEAD OF DEPT.
DR. R.R MANTHALKAR Prof. A.N .KAMTHANE
2

3. INDEX
• ACKNOWLEDGEMENT………………………………........4
• ABSTRACT………………………………………………......5
• LIST OF FIGURES……………………………………..........6
• INTRODUCTION TO BIOMETRICS ………………………7
o ADVANTAGES OF BIOMETRICS………………………………8
o COMMON BIOMETRIC PATTERNS……………………………9
o THE HUMAN IRIS………………………………………………..10
• PROJECT INTRODUCTION…………………………...…..14
o OBJECTIVE………………………………………………………14
o BASIC ASSUMPTIONS………………………………………….14
o DATABASE CHARACTERISTICS……………………………...15
• IMAGE SEGMENTATION ………………………….…….16
o AVAILABLE TECHNIQUE…………………………….……….16
o OUR APPROACH……………………………………….……….18
• IMAGE NORMALIZATION ………………………………23
o BASIC MOTIVATION…………………………………..………23
o OUR APPROACH……………………………………… ……….24
• FEATURE EXTRACTION ………………………………...26
o BASIC TECHNIQUE……………………………………………26
o OUR APPROACH…………………………………………….....29
• MATCHING …………………………………….…………31
o DIFFERENT DISTANCE METRICS…………………………..31
o OUR APPROACH………………………………………………34
• ENROLLMENT AND IDENTIFICATION…….…….........35
o ENROLLMENT PHASE………………………………………..35
o IDENTIFICATION PHASE………………………………….....36
3

4. • CONCLUSION …………………………………………….37
 SUMMARY OF WORK………………………………37
 SUGGESTED IMPROVEMENTS…………………….38
• GRAPHIC USER INTERFACE……………………...…….39
• REFERENCES …………………………………………......43
ACKNOWLEDGEMENT
4

5. For the accomplishment of any Endeavour the most
important factor besides Hard-work & Dedication is the Proper
Guidance. Just as a light-house provides guidance to the sea
voyagers even in the extreme stormy conditions the same job has
been done by our guide Dr. R.R. Manthalkar.
We heartily thank him for his timely support & guidance.
His keen interest & enthusiasm in executing the project has always
been a source of inspiration for us.
We would like to thank our Head of department Prof.
A.N Kamthane and also Dr. M.B.kokare for their moral support &
encouragement. We would also take this opportunity to extend our
sincere thanks to all the other professors of our department for their
direct or indirect help extended by them which was a cap in the
feather of our efforts.
We would also like to thank all our friends who also
played an instrumental role in the completion of our project. Last
but not the least we would like to thank our parents who always
support us in our efforts.
MR. DESAI GAURAV S.
MR. JOSHI SUSHIL D.
MS. IYER ANANDHI G.
MR.WAZALWAR DHAWAL S.
(BE. EXTC)
5

6. ABSTRACT
The successful implementation of any system is largely
determined by its reliability, authenticity & the amount of secrecy it
provides. In today’s highly techno world where security & privacy
are the concerns of prime importance the crucial systems must
employ techniques to achieve this.
Our project is just a small step towards this. The Iris based
system can cope up with a lot of the individual biological variations
& still provide the identification system with much accuracy &
reliability.
In this project we have designed system which involves
recognition of a person using IRIS as biometric parameter. We have
first segmented pupil & iris structure from the original eye image.
Then we have normalized it to build a feature vector which
characterizes each iris distinctly. This feature vector is then used for
matching among various templates & identifies the individual.
The work provided in this project is mainly aimed at
providing a recognition system in order to verify the uniqueness of
the human iris and also its performance as a biometric. The paper
has implementation of all the algorithms on the CASIA database.
The various results of the different implementations and their
accuracies have been tested in this paper.
All in all this paper is a sincere effort in suggesting an
efficient system for the implementation of the Human Identification
system based on IRIS recognition.
LIST OF FIGURES
6

7. • Fig no. 1.3.1 Picture of Human Eye……………………….12
• Fig no. 3.1.1 Demonstration of Hough Transform………...17
• Fig no.3.2.1 Fully Segmented Iris Image………………....21
• Fig no 4.1.1 Daugman`s Rubber sheet model…………....23
• Fig no. 4.2.1 Fully and Enhanced Normalized Image……..24
• Fig no. 5.1.1 2 Level Decomposition of Wavelet………….26
• Fig no. 5.2.1 Application of db2 Wavelet to eye image…...29
• Fig no. 7.1.1 General Iris Identification System ………….34
CHAPTER 1
INTRODUCTION TO BIOMETRICS
7

8. In this vastly interconnected society establishing
the true identity of the person is becoming the most critical issue.
The questions like “Is she really who she claims to be?”, “Is this
person authorized to use this facility?” or “Is he in the watch list
posted by the government?” are routinely being asked in a variety of
scenarios ranging from issuing a driver’s license to gaining entry
into a country. With the advent of the newer networking
technologies the sharing of vast audio & video resources has
become much easier but has raised brows over the issues concerning
the security in the transactions. The need for reliable user
authentication techniques has increased in the wake of heightened
concerns about security and rapid advancements in networking
, communication and mobility.
Biometrics, defined as a science of recognizing
an individual based on her physiological, biological or behavioral
traits, is beginning to gain acceptance as a legitimate method for
determining an individual’s identity. Biometrics aims to accurately
identify each individual using various physiological or behavioral
characteristics such as fingerprints, Face, iris, retina, gait, palm-
prints and hand geometry. The developments in science and
technology have made it possible to use biometrics in applications
where it is required to establish or confirm the identity of
individuals.
Applications such as passenger control in
airports, access control in restricted areas, border control, database
access and financial services are some of the examples where the
biometric technology has been applied for more reliable
identification and verification. In recent years, biometric identity
cards and passports have been issued in some countries based on
iris. Fingerprint and face recognition technologies to improve
border control process and simplify passenger travel at the airports.
In the field of financial services, biometric technology has shown
a great potential in offering more comfort to customers while
increasing their security. Although there are still some concerns
about using biometrics in the mass consumer applications due to
8

9. information protection issues, it is believed that the technology will
find its way to be widely used in many different applications.
1.1 ADVANTAGES OF BIOMETRICS:
1. It links an event to a particular individual, not just to a
password or token.
2. It is very convenient from user friendliness point of view since
there is nothing to remember, unlike in the case of passwords
or some code words.
3. It can’t be guessed, stolen, shared, lost or forgotten.
4. It prevents impersonation by protecting against identity theft
and providing higher degree of non- repudiation.
5. It enhances privacy by protecting against unauthorized access
to personal information.
A good biometric is characterized by use of a feature
that is highly unique – so that the chance of any two people
having the same characteristic will be minimal, stable – so that
the feature does not change over time, and be easily captured – in
order to provide convenience to the user, and prevent
misrepresentation of the feature.
1.2 COMMON BIOMETRIC PATTERNS
1. Finger print Recognition
Features:
a. Measures characteristics associated with the friction ridge
pattern on the fingertip.
b. General ease and speed of use.
c. Supports both 1:1 verification and 1:N applications
Considerations:
a. Ridge patterns may get affected due to accidents or aging
effects.
9

10. b. Requires Physical contact with sensor.
2. Facial Recognition
Features:
a. Analyzes geometry of the face or the relative distance
between features.
b. No physical contact required.
c. Supports both 1:1 verification and 1:N identification
applications.
Considerations:
a. Can be affected by surrounding lightning conditions.
b. Appearance may change over time.
3. Hand Geometry
Features:
a. Measures dimensions of hand, including shape and length of
finger.
b. Very low failure to enroll rate.
c. Rugged.
Considerations:
a. Suitable only for 1:1 contexts.
4. Speech Recognition
Features:
a. Compares live speech with previously created speech model
of person’s voice.
b. Measures pitch, cadence and tone to create voice print
cadence.
Considerations:
a. Background Noise can interfere
b. Suitable only for 1:1 contexts.
10

11. 1.3 THE HUMAN IRIS
The human iris is rich in features which can be used to
quantitatively and positively distinguish one eye from another. The
iris contains many collagenous fibers, contraction furrows,
coronas, crypts, color, serpentine vasculature, striations, freckles,
rifts, and pits. Measuring the patterns of these features and their
spatial relationships to each other provides other quantifiable
parameters useful to the identification process. The iris is unique
because of the chaotic morphogenesis of that organ.
To quote Dr. John Daugman, “An advantage the iris shares with
fingerprints is the chaotic morphogenesis of its minutiae. The iris
texture has chaotic dimension because its details depend on initial
conditions in embryonic genetic expression; yet, the limitation of
partial genetic diffusion (beyond expression of form, function,
color and general textural quality), ensures that even identical
twins have uncorrelated iris minutiae. Thus the uniqueness of
every iris, including the pair possessed by one individual, parallels
the uniqueness of every fingerprint regardless of whether there is a
common genome.” Given this, the statistical probability that two
irises would be exactly the same is estimated at 1 in 10^72.
1.3. a. STABLITY OF IRIS
Notwithstanding its delicate nature, the iris is protected
behind eyelid, cornea, aqueous humor, and frequently eyeglasses
or contact lenses (which have negligible effect on the recognition
process). An iris is not normally contaminated with foreign
material, and human instinct being what it is, the iris, or eye, is one
of the most carefully protected organs in one’s body. In this
environment, and not subject to deleterious effects of aging, the
features of the iris remain stable and fixed from about one year of
age until death.
11

12. 1.3. b. STRUCTURE OF IRIS
Fig no. 1.3.1 Picture of Human Eye
The iris is a thin circular diaphragm, which lies between
the cornea and the lens of the human eye. A front-on view of the
iris is shown in Figure. The iris is perforated close to its centre by a
circular aperture known as the pupil. The function of the iris is to
control the amount of light entering through the pupil, and this is
done by the sphincter and the dilator muscles, which adjust the size
of the pupil. The average diameter of the iris is 12 mm, and the
pupil size can vary from 10% to 80% of the iris diameter.
The iris consists of a number of layers; the lowest is the
epithelium layer, which contains dense pigmentation cells. The
stromal layer lies above the epithelium layer, and contains blood
vessels, pigment cells and the two iris muscles. The density of
stromal pigmentation determines the color of the iris. The
externally visible surface of the multi-layered iris contains two
zones, which often differ in color. An outer ciliary’s zone and an
12

13. inner pupillary zone, and these two zones are divided by the
collarets – which appears as a zigzag pattern.
1.3. c. ADVANTAGES OF USING IRIS PATTERN
1. Highly protected internal organ of the eye.
2. Iris patterns possess a high degree of randomness.
3. Patterns are apparently stable throughout life.
4. Comparatively fast matching technique.
13

14. CHAPTER 2
PROJECT INTRODUCTION
2.1. OBJECTIVE
The objective will be to implement an open-source
iris recognition system in order to verify the claimed
performance of the technology. Our main system consists of
many subsystems. Important steps involved can be
demonstrated as follows:
1. Segmentation:
Locates the iris region in the eye image by eliminating the
unwanted parts.
2. Normalization
Creating a dimensionally consistent representation of iris
region.
3. Feature Encoding and Matching
Creating a template containing only the most discriminating
features of the iris and matching it.
So our main aim is to implement best possible
algorithm for each step and obtain required degree of accuracy.
2.2 BASIC ASSUMPTIONS
a) Methods for image segmentation and feature extraction will
assume all patterns have the same rotation angle.
b) The Iris and the pupil regions are assumed to be perfectly
concentric circles.
c) The pupil region has a constant intensity throughout.
d) Of the available seven images of a subject four are considered
as principle images while three are considered for test images.
e) We have resized normalized image of each of the iris into a
vector of constant dimensions irrespective of their original
size.
14

15. 2.3. DATABASE CHARACTERISTICS
This paper is based on the CASIA image database. Its
ethnical distribution is composed mainly of Asians. Each iris class
is composed of 7 samples taken in two sessions, three in the first
session and four in the second. Sessions were taken with an
interval of one month. Images are 320x280 pixels gray scale taken
by a digital optical sensor designed by NLPR (National Laboratory
of Pattern Recognition – Chinese Academy of Sciences). There are
108 classes or irises in a total of 756 iris images.Each of the iris
images is preprocessed to eliminate the effect of the illumination
variations and other noise effects.
15

16. CHAPTER 3
IMAGE SEGMENTATION
The main aim of the segmentation step is to distinguish
an iris texture from the rest of the eye image. Properly detecting
the inner and outer boundaries of iris texture is significantly
important in all iris recognition systems. The segmentation is a
crucial step in that any false representation here may corrupt the
resulting template in poor recognition rates.
The iris is an annular portion between the pupil (inner)
and the sclera (outer boundary). The iris regions can be
approximated by two circles, one for the iris/sclera boundary and
another, interior to the first, for the iris/pupil boundary. The eyelids
and eyelashes normally occlude the upper and lower parts of the
iris region. Also, specular reflections can occur within the iris
region corrupting the iris pattern. However, our project work is
mainly focused on CASIA iris database which do not contain
specular reflections due to the use of near infra-red light for
illumination. The success of the segmentation depends upon the
imaging quality of the eye image. Imaging of the iris must acquire
sufficient detail for recognition while being minimally invasive to
the operator.
3.1 Available Techniques
a. Daugman`s Integro Differential Operator
Daugman proposed a method by making use
of first derivatives of image intensity to signal the location of
edges that corresponds to the location of edges that corresponds
to the borders of the iris. The notion is that the magnitude of the
derivative across an imaged border will show a local maximum
due to the local change of image intensity. The limbus and pupil
are modeled with circular contours. The expected configuration
16

17. of model components is used to fine tune the image intensity
derivative information.
Daugman`s operator is expressed as
I(x, y) is an image containing an eye
The operator searches over the image domain (x, y) for the
maximum in blurred partial derivative with respect to increasing
radius r of the normalized contour integral of I(x, y) along a
circular arc ds of radius r and center coordinates (x0, y0) .
Is smoothing function
Operator serves and behaves as circular edge detector blurred at a
scale by σ, which searches iteratively for maximum contour
integral derivative with increasing radius at successively finer
scales of analysis through the three parameter space of center
coordinates and radius (x0, y0, r) defining a path of contour
integration.
b. Hough Transform
This is a standard computer algorithm that can be used to
determine the parameters of single geometric objects such as lines
and circles, present in an image. Circular Hough Transform is used
to deduce the radius and center coordinates of the pupil and iris
regions, while to detect the eyelids, parabolic Hough Transform is
used.Hough Transform can be demonstrated as follows:
Fig. no. 3.1.1 Demonstration of Hough Transform
17

18. These edge maps along with determination of some appropriate points
in Hough space will give us required parameters.
Some problems in use of Hough Transform are:
1. It is difficult to determine the threshold values to be chosen for
edge detection resulting in critical edge points being removed
sometimes, resulting in failure to detect circles/arcs.
2. This approach is computationally intensive due to its brute force
nature and thus may not be suitable for real time applications.
c. Other Methods for Segmentation
Iris localization proposed by Tisse. et al is combination of
Integro differential and the Hough Transform. The Hough Transform
is used for a quick guess of the pupil center and then the integro
differential is used to accurately locate pupil and limbus using a
smaller search space.
Lim et al localize pupil and limbus by providing an edge
map of the intensity values of the image. The center of the pupil is
then chosen using a bisection method that passes perpendicular lines
from every two points on the edge map. The center point is then
obtained by voting the point that has the largest number of line
crossovers. The pupil and limbus boundaries are then selected by
increasing the radius of a virtual circle with selected center point and
choosing the two radii that have the maximum number of edges by the
virtual circle as the pupil and limbus radii.
3.2 Our Approach
Our approach is mainly divided into four parts as listed below:
1. Separation of Pupil Region from Eye image.
2. Determination of Pupil Coordinates
3. Determination of Iris Edges based on the above two steps
4. Removing the unwanted part and getting the segmented part
18

19. a. Separation of Pupil Region
CASIA Database images have been preprocessed and
each has constant intensity pupil region. Firstly we determine
this constant value so that we can use it as a threshold value to
separate pupil region from eye image. For this we employ
following steps:
• First, We obtain mean of the image (say ’m’)
• Then we scan the image completely to determine the regions
having same pixel value consecutively for 15 times.
• Sometimes, Camera effect may develop bright constant
intensity regions on the eyelids. To avoid selecting this
region value as threshold value, we compare each region
value with the obtained mean (here ‘m’) and select that value
which is less than the mean value.
• After getting the threshold value, we stop scanning the
image.
• Sometimes, eyelashes also satisfy the threshold condition, but
have much smaller area than pupil area. Using this
knowledge and concept of 8 connected pixels, we can cycle
through all regions and apply the following conditions
For each region R
If Area(R) <1300,Set all pixels of R to 0
• To this finally obtained pupil image, some statistical
MATLAB functions are applied and it’s Center Coordinates
and radius is determined.
b. Determination of Pupil Coordinates
This function is mainly used to determine the edges of pupil.
Here, we take the help of MATLAB ‘find’ function to detect the
abrupt change in the image intensity. This function also helps us to
verify the previously determined Pupil parameters with the help of
following formula:
Center = (|Right/Top Edge - left/Bottom Edge|)/2
Coordinates coordinates Coordinates
19

20. c. Determination of iris edges
This is used to obtain the contour of the iris. Here, we
assume that iris and pupil are concentric circular regions. It takes
into consideration that areas of the iris at the right and left of the
pupil are the ones that most often present visible to data extraction.
The areas above and below the pupil also carry unique
information, but it is very common that they are totally or partially
occluded by eyelash or eyelid.
Firstly, we enhance the iris image so that edges can be seen
more prominently and their detection will become simpler. For this
we employ Histogram equalization method which gives us an
image with increased dynamic range, which will tend to have
higher contrast. Then we follow certain steps as listed below to get
the iris edges:
• Firstly, we get the pupil edges from previously explained
functions.
• Then, we start the process with 30 pixels left to the left edge
of pupil (say this point is ‘a’), so as to avoid abrupt intensity
changes in collarette region, which may mislead our
algorithm. We form a vector containing image intensity at
that point ‘a’ and 4 pixels above and below it. Mean of this
vector is obtained (say it is m1).
• Another vector containing image intensity at point ‘a-1’ and
3 points above and below it is obtained. Mean of this vector
is obtained (say it is m2).
• Now, We compute ‘m’, where m=|m1- m2|
• If m>. 04, there may be abrupt change in image intensity.
• To avoid false detection which may cause due to iris region
getting corrupted by some reasons, we obtain similar vector
for ‘a- 2’also and calculate its mean and store it in m2.
• Again step 5 is repeated and if condition is satisfied for 10
times consecutively we conclude that this is the left edge of
20

21. the iris and stop further calculations. We calculate this edge’s
distance from pupil center and thus we get iris radius towards
left i.e. r1.
• The above steps are repeated for right part also. Here, the
exception is that we start from 30 pixels right to the right
most pupil edge. This gives radius towards right i.e. r2.
• We take the larger value amongst r1 and r2 so as to avoid any
data loss.
d. Removal of unwanted part from the image
Now, the above data so far obtained can be used to
eliminate portions other than iris region. This is done by
approximating Iris as a circle with center at pupil center. Any region
outside this circle is deleted and then finally we get the segmented
image as shown:
Fig no. 3.2.1 Fully Segmented Iris Image
21

22. CHAPTER 4
NORMALIZATION
The main aim of normalization is to
transform the iris region into fixed dimensions, in order to null
the effect of dimensional inconsistencies. These inconsistencies
are the result of changes that occur in image distance and
angular position with respect to the camera. Other sources of
inconsistencies include stretching of the iris caused by pupil
dilation from varying levels of illumination, head tilt and
rotation of eye within the eye socket. Normalization is a very
important step in that success of subsequent steps depends
largely on the accuracy of this step.
4.1 BASIC MOTIVATION
Daugman`s Rubber Sheet Model
This method is based on remapping each point within the iris
region to a pair of polar coordinates (r, θ) where r is on the
interval [0 1] and θ is angle in [0 2Π]. The proposed method is
capable of compensating the unwanted variations due to
distance of eye from the camera (scale) and its position with
respect to the camera (Translation). The Cartesian to polar
transform is defined as:
Where,
I(x, y) is the iris region image,(x, y) are the original Cartesian
coordinates ,(r, θ) are the corresponding normalized polar
22

23. coordinates and , and , are the pupil
and iris boundaries along θ direction. Pictorial Depiction of this
model is as shown
Fig no. 4.1.1 Daugman`s rubber sheet model
This Rubber sheet model takes into account pupil dilation and
size inconsistencies in order to produce a normalized with
constant dimensions. However, this model does not compensate
for rotational inconsistencies. In Daugman`s system, rotation is
accounted for during matching by shifting the iris template in θ
direction until iris templates are aligned.
4.2 OUR APPROACH:
Our algorithm involves the polar conversion of the iris
image. The polar conversion of the iris image accounts for all
the rotational inconsistencies in the iris image. In this algorithm
we map only the annular iris region coordinates into a fixed
dimension vector thus eliminating the unwanted and redundant
pupil & other non iris regions.
The algorithm can be explained as follows:
1. Finding out incremental values of the radius and the
rotation angle.
23

24. 2. Creating a pseudo polar mesh grid of fixed dimensions
using the incremental values and the radius of the iris
region.
3. Separating only the annular iris ring thus eliminating pupil
& other regions.
4. Conversions of each coordinate of the iris region into its
equivalent polar coordinate using the relations.
5. Mapping of the each of the iris coordinate onto the polar
grid a using linear interpolation.
6. The iris region is mapped into a vector of fixed dimensions
of size 100x360 which can be seen as shown below
Fig no.4.2.1 Fully and enhanced Normalized Image
24

25. CHAPTER 5
FEATURE EXTRACTION
The normalized image is used to extract the unique features in
the iris image. Each iris is characterized by its unique collarette
pattern, iris variation patterns. Hence it is very essential to extract
these features to represent each iris distinctively. Most Iris
recognition systems make use of a band pass decomposition of iris
image to create a biometric template. This step is responsible of
extracting the patterns of iris taking into account the correlation
between the adjacent pixels.
5.1 BASIC TECHNIQUE
a. THEORY OF WAVELETS
Wavelets are the basis functions wjk(t) in continuous
time. A basis is a set of linearly independent functions that can be
used to produce all admissible functions f(t).
F(t)=combination of basis functions = ∑bjk wjk(t)
The special feature of the wavelet is that all the functions are
constructed from a single mother wavelet w(t).
Wavelet transform overcomes the resolution problem of the
traditional Fourier transform techniques by using a variable length
window. Techniques like short time fourier transform used to divide
the signal into short time domains. The fourier transform of the
signal was then computed in each domain around a particular center
frequency. However this led to the time resolution problem which
led to the wavelet approach. Analysis windows of different lengths
are used for different frequencies:
Analysis of high frequenciesè Use narrower windows for
better time resolution
Analysis of low frequencies è Use wider windows for
better frequency resolution
25

26. This works well, if the signal to be analyzed mainly consists of
slowly varying characteristics with occasional short high
frequency bursts.
The function used to window the signal is called the wavelet
ψ ψ 1 ∗ t −τ 
CWT x (τ , s ) = Ψ (τ , s ) =
x ∫ x( t )ψ  s dt
s t  
~
x[n yhigh [k ] = ∑ x[n] g[− n + 2k ] ∑ yhigh [k ] ⋅ g[−n + 2k ] x[n]
k
] n
~
G G +
2 2
~ 2 ~ 2
H G 2 2 G + H
~ 2 2
~
ylow[k ] = ∑ x[n] h[− n + 2k ] H H ∑ yhigh [k ] ⋅ g[−n + 2k ]
n k
Decomposition Reconstruction
G 2
H 2
Fig no. 5.1.1 2 Level Decomposition of Wavelet
b. Discrete Wavelet Transform
The DWT analyzes the signal at different frequency bands
with different resolutions by decomposing the signal into coarse
approximation and detail information. DWT employs two sets of
functions, called scaling functions and wavelet functions, which
are associated with low pass and high pass filters, respectively.
The decomposition of signal into different frequency bands is
simply obtained by successive high pass and low pass filtering of
the time domain signal. The original signal x[n] is first passed
through a half band high pass filter g[n] and low pass filter h[n].
After the filtering, half of the samples can be eliminated
26

27. according to the Nyquist`s rule, since the signal now has a highest
frequency of Π/2 radians instead Π. The signal can therefore be
sub sampled by discarding every other sample. This constitutes
one level of decomposition and can mathematically be expressed
as follows:
Yhigh[k] =Σx[n].g[2k - n]
Ylow[k] =Σx[n].h[2k - n] for all n;
Where Yhigh[k] and Ylow[k] are the outputs of the high pass and
low pass filters, respectively, after sub sampling by 2.
This decomposition halves the time resolution since only half the
number of samples characterizes the entire signal. However, this
operation doubles the frequency resolution since the frequency
band of signal now spans only half the previous frequency band,
effectively reducing the uncertainty in the frequency by half. The
above procedure, which is also known as the sub band coding,
can be repeated for further decomposition. At every level, the
filtering and sub sampling will result in half the number of
samples (and hence half the time resolution) and half the
frequency bands spanned (and hence double the frequency
resolution).
c. VARIOUS TYPES OF WAVELETS
1. Haar Wavelet
This uses the wavelet transform to extract features from the
iris region. Both the Gabor transform and the Haar wavelet are
considered as the mother wavelet. From multi-dimensionally
filtering, a feature vector with 87 dimensions is computed. Since
each dimension has a real value ranging from -1.0 to +1.0, the
feature vector is sign quantized so that any positive value is
represented by 1, and negative value as 0. This results in a
compact biometric template consisting of only 87 bits.
27

28. 2. Daubechies Wavelet
These are a family of orthogonal wavelets defining a DWT
and characterized by a maximal number of vanishing moments
for some given support. With each wavelet type of this class,
there is a scaling function (also called as Father Wavelet) which
an orthogonal multi resolution analysis. An orthogonal wavelet is
a wavelet where the associated wavelet transform is orthogonal
i.e. the inverse wavelet transform is ad joint of the wavelet
transform.
In general the Daubechies wavelets are chosen to have the
highest number ‘A’ of vanishing moments (This does not imply
the best smoothness)for given support width N=2A and among
the 2^(A-1) possible solutions, the one is chosen whose scaling
filter has external phase. These wavelets are widely used in
solving broad range of problems, example similarity property of a
signal, signal discontinuities, etc. Daubechies orthogonal wavelet
D2-D0 is commonly used. The index number refers to the
number N of coefficients. Each wavelet has a number of zero
moments or vanishing moments equal to half the number of
coefficients.
5.2 OUR APPROACH:
We have applied a 5- level Wavelet transform and have
analyzed the results for all types of Wavelets like Haar, Daubechies
and Bior. We found the best results for Daubechies that to for db2
type. In each case, feature size varied depending upon the type
wavelet applied. Normally, the final template size should not be a
function of type of wavelet, but since our image size is not exactly
in terms of powers of two (i.e it is 100x360 ) and so subsequent
desampling will result in loss of some pixels. This causes template
size to be a function of type of Wavelet and also on the number of
levels. Following figure shows 4- level DB2 Wavelet applied to the
eye image.
28

29. Fig no. 5.2.1 Application of db2 Wavelet to Eye image
The results obtained by applying different types of Wavelet are
shown separately in Results section.
29

30. CHAPTER 6
MATCHING
For comparing the templates obtained by feature
extraction process, there are number of design metrics available.
Some of them are explained:
6.1 Different Distance Metrics
If x and yare two d-dimensional feature
vectors of database image and query image respectively, then the
distance metrics are defined as:
a. Euclidean or L2 metric :
Euclidean distance is not always the best metric. The fact that the
distances in each dimension are squared before summation,
places great emphasis on those features for which the
dissimilarity is large.
b. Weighted Euclidean distance metric:
The weighted Euclidean distance (WED) can be used to
compare two templates, especially if the template is composed of
integer values. The weighting Euclidean distance gives a measure
of how similar a collection of values are between two templates.
This metric is specified as
30

31. where is the feature of the unknown iris, and is the
feature of iris template, ,and is the standard deviation of the
feature in iris template . The unknown iris template is found to
match iris template , when WED is a minimum at .
c. The Manhattan or LI metric:
The Manhattan distance metric uses using sum of the absolute
differences in each feature, rather than their squares, as the
overall measure of dissimilarity. It is obvious that the distance
of an image from itself is zero. The distances are then stored in
increasing order and closest sets of patterns are then retrieved.
In ideal case all the top 16 retrievals are from same large image.
The performance is measured in terms of the average retrieval
rate, which is defined as the average percentage of patterns
belonging to the same image as the query pattern in the top 16
matches.
d. Hamming Distance metric:
The Hamming distance gives a measure of how many bits
are the same between two bit patterns. Using the Hamming
distance of two bit patterns, a decision can be made as to
whether the two patterns were generated from different irises or
from the same one. In comparing the bit patterns X and Y, the
Hamming distance, HD, is defined as the sum of disagreeing
bits (sum of the exclusive-OR between X and Y) over N, the
total number of bits in the bit pattern.
31

32. Since an individual iris region contains features with high
degrees of freedom, each iris region will produce a bit-pattern
which is independent to that produced by another iris, on the
other hand, two iris codes produced from the same iris will be
highly correlated. If two bits patterns are completely
independent, such as iris templates generated from different
irises, the Hamming distance between the two patterns should
equal 0.5. This occurs because independence implies the two bit
patterns will be totally random, so there is 0.5 chance of setting
any bit to 1, and vice versa. Therefore, half of the bits will agree
and half will disagree between the two patterns. If two patterns
are derived from the same iris, the Hamming distance between
them will be close to 0.0, since they are highly correlated and
the bits should agree between the two iris codes.
e. Canberra distance metric:
The Canberra distance metric is given by
In these metric equations the numerator signifies the
difference and denominator normalizes the difference. Thus
distance values will never exceed one, being equal to one
whenever either of the attributes is zero. Thus it would seem to
be a good expression to use, which avoids scaling effect.
32

33. 6.2 Our approach:
Running the experiment with different distance metric
on same set of images we were able to find which distance
metric gives best result.
The Canberra distance metric performed exceptionally
well than other distance metric. The fact is that in this metrics
equation the numerator signifies the difference and denominator
normalizes the difference. Thus distance values will never
exceed one, being equal to one whenever either of the attributes
is zero.
33

34. CHAPTER 7
ENROLLMENT AND IDENTIFICATION
7.1 ENROLLMENT PHASE
a. PROBLEM STATEMENT
The enrollment phase in any biometric system is required to
validate its users. If an authentic person supposed to use the system
does not have his images registered in the database then every time
the person tries enter the system the biometric system rejects him
and refrains from providing access to the system to that person.
Hence a person worthy of access is denied the service. In such cases
it becomes necessary to first enroll the eligible persons into the
database for their future identification. Our enrollment algorithm is
basically aimed at solving this problem.
Fig no. 7.1.1 General Iris identification system
b. STEPS:
The enrollment phase of the project involves checking whether
any of the subject is already enrolled in the database or not and if
not getting the subject enrolled. Here a basic graphic user interface
window is implemented. Of the available seven image s of a subject
three are considered as principle images and rest four are considered
34

35. for testing the enrollment. The basic steps are highlighted as
follows:
• A test image of a subject is fed to the algorithm.
• The image is subjected to series of steps of segmentation,
normalization, and feature extraction to form a feature vector
template.
• This is used to compute the distance between the template and
combined pattern of all the subjects using the Canberra distance
metric.
• The distances are stored in a vector.
• This is then sorted in ascending order.
• The minimum distance indicates the class of images to which
this test image belongs to.
• The test image is then once again compared with the principle
images of the class and the distance between them is computed.
• If the above distance is less than the one obtained for the
combined pattern of that class then we can establish that the image
indeed belongs to that class and hence is already enrolled.
• If the distance computed in the step seven is greater than the
combined distance we conclude that the image does not belong to
that class and is not enrolled.
• In case the test image is not enrolled then it is enrolled by
storing all the patterns of the subjects in the database.
7.2 IDENTIFICATION PHASE
a. PROBLEM STATEMENT
This phase verifies the claimed identity of the individual. In this
phase a person who wants to get access to a system claims his
identity as being a particular authentic person. The biometric system
verifies for the claim and establishes whether he is indeed the person
35

36. who he claims to be or an imposter.The following algorithm solves
this problem.
b. STEPS
This algorithm is implemented using a graphic user interface.
• A test image is input through the graphic interface.
• The image is subjected through the processing steps such as
segmentation, normalization and feature extraction to the retrieve
the feature vector.
• This is used to compute the distance between the template and
combined pattern of all the subjects using the Canberra distance
metric.
• The distances are stored in a vector.
• This is then sorted in ascending order.
• The minimum distance indicates the class of images to which
this test image belongs to.
• The test image is then once again compared with the principle
images of the class and the distance between them is computed.
• If the above distance is less than the one obtained for the
combined pattern of that class then we can establish that the image
indeed belongs to that class and hence the claimed identity of the
person is established.
• If the distance computed in the step seven is greater than the
combined distance we conclude that the image does not belong to
that class and hence the person is an imposter.
CHAPTER 8
36

37. CONCLUSION
8.1 Summary of work
This project work presents an Iris recognition system, which
was tested on CASIA Iris Image Database, in order to verify the
claimed performance of iris recognition technology. Analysis of the
developed system has revealed a number of interesting conclusions.
Accuracy of Biometric identification systems is specified in terms of
FAR (False Acceptance Rate) and FRR (False Rejection Rate) .
FAR measures how often a non –authorized user, who should not be
granted access, is falsely recognized, while FRR measures how
often an authorized user, who should have been granted access is not
recognized.
a. Results for applying different types of Wavelets
Following is the table which shows the effect of applying various
types of the Wavelet
Wavelet Vector Size FAR FRR No. of Faulty
Type Subjects
Db3 764X1 5 130 74
Db1 418X1 17 87 52
Db4 1020X1 15 133 71
Bior 4.4 1047X1 4 87 52
Bior 1.1 418X1 17 87 52
Db2 477X1 1 73 30
Thus we can conclude that db2 gives the best possible results and so
we have selected this type of Wavelet.
b. Results for Overall Database
37

38. Thus after analyzing the results over complete database we
found following results.
False Acceptance Rate=1
False Rejection Rate=73
Number of Faulty Subjects= 41
Thus, our project has very small FAR which is what a
reliable biometric system should possess. However,
comparatively our FRR is quite high and some measures can be
taken to reduce it.
8.2 Suggested Improvements
1. Segmentation is a very crucial step in Iris Recognition Systems
and so to whatever extent its accuracy can be increased will result in
more improved results. To improve segmentation algorithm, a more
elaborate eyelid and eyelash detection system could be
implemented.
2. Presently, our Template size varies with the type of Wavelet
applied. So, some measures can be taken to avoid this effect.
Our aim of project was mainly to have a very low
FAR, which we have achieved to a large extent. Also, our
segmentation part has a very high accuracy and has worked
satisfactorily over entire CASIA database. Since, we had
restricted ourselves to software part implementation and not
taken too much hardware implications in too concern, our
efforts have resulted in efficient Recognition System.
CHAPTER 9
38

39. GUI Using MATLAB
Since MATLAB provides a very easy way of
implementing a GUI (Graphical User Interface), we have
prepared a GUI which gives a general idea of the work we have
done. Following are some of its details:
1. Starting Window:
This is starting window of our GUI which provides link to
various phases of our project. The Results for overall can be
obtained by clicking on 1st pushbutton. To have Individual
analysis of the database, we have provided one option in the form
of 2nd pushbutton. Different steps in pattern formation can be seen
by clicking on 3rd pushbutton.
2. Results for Complete Database
39

40. The above window shows the overall
database results. The update facility is provided so as to again run
the code over entire database, in case any changes made in the
original code.
3. Individual Enrollment
40

41. This window allows us to individually analyze each person by
enrolling that person separately and then check if we get proper
results against each image of that person in database. Here, we
have used four images for pattern formation and then keep
remaining three images as test images.
4. Steps in Pattern formation
41

42. This window provides a detailed analysis of pattern formation.
Here, we have made provisions to see Pupil separation, Iris
segmentation and also Normalized Image for any image in
database.
42

43. CHAPTER 10
REFERENCES
1. JOURNALS AND CONFERENCE PAPERS
• Daugman, J., How Iris Recognition Works, IEEE
Transactions on Circuits and Systems for Video Technology,
Vol. 14, Number 1, January 2004.
• Masek L., Recognition of Human Iris Patterns for Biometric
Identification,
[http://www.csse.uwa.edu.au/~pk/studentprojects/libor].
• CASIA iris image database, Institute of Automation, Chinese
Academy of Sciences, [http://www.sinobiometrics.com]
• R. Wildes. Iris recognition: an emerging biometric
technology. Proceedings of the IEEE, Vol. 85, No. 9,1997.
• Daugman J., Biometric Personal Identification System based
on Iris Analysis,United States Patent, Patent No. 5,291,560,
March 1994.
• J. Daugman. High confidence visual recognition of persons
by a test of statistical independence. IEEE Transactions on
Pattern Analysis and Machine Intelligence, Vol. 15, No. 11,
1993.
• W.W.Boles, A security system based on human Iris
Identification using Wavelet Transform ,Engineering
Applications of Artificial Intelligence, 11:77-85,1998
2. OTHER REFERENCES
• A book on Digital Image Processing using MATLAB by
Rafael C. Gonzalez, Richard E. Woods and Steven L.Eddins.
• Wavelet tutorial By Robi Polikar.
43

44. 44