ADVANCED IRIS RECOGNITION SYSTEM USING DAUGMAN’S

ALGORITHM AND ARTIFICIAL NEURAL NETWORKS

Dr.K. Venkata Naganjaneyulu,1Chirukuri Mounika Nagasai2
2
PG Scholars, Department of CSE, Malla Reddy Engineering College For Women -
MRECW(Autonomous Institution,UGC,Govt.of India.)
1
Professor, Department of CSE,Malla Reddy Engineering College For Women -
MRECW(Autonomous Institution,UGC,Govt.of India.)
Email id: [email protected]
Abstract: We have an assortment of techniques to biometric identification. One of the most
advanced and efficient approaches is iris recognition. Pattern recognition is its primary
mechanism; it finds unique and easily identifiable patterns in the iris to positively identify the
person. This method of identification is more accurate and produces better results. Security
breaches and authentication fraud are on the rise, making the implementation of a strict biometric
system imperative. The proposed research makes use of Daugman's method for iris localization;
this is an integro-differential operator that may separate or segment regular shapes. The
Daugman algorithm also has the ability to successfully decrease noise. Iris localization is best
accomplished using the Daugman method because of these two features. Feature extraction,
which follows iris localization, finds the consistent and unique aspects of an iris picture. Mean,
Standard Deviation, Entropy, Root Mean Square, Smoothness, Kurtosis, Energy, Homogeneity,
Contrast, and Variance are all computed in the research. In reaction to various iris pictures, the
features display unique behavior. Possible overlap of values exists. The features are then fed into
a neural network using the Levenberg-Marquardt back propagation training algorithm. After
training using feature values extracted from permitted photos, the neural network is then tested
for correctness. The traditional MMU database was included into the design of the system.
Compared to the prior technique that used the same database, the suggested method achieved a
better degree of accuracy, namely 99.7 percent.

Keywords: Breaches, Biometric Identification, Daugman Algorithm, Levenberg-Marquardt.

INTRODUCTION the iris recognition system. In most cases,

User authentication is one of the most this is a very convenient approach. Figure 1
important areas that are part of secure data shows the process of iris recognition system
access. Iris authentication and recognition and its various steps. Iris recognition is a
approach helps to authenticate the target type of biometric authentication system that
user using the unique features of the iris. applies advanced mathematical techniques
There are some other applications like: B. It and processes to the digital image of the
is used in ATMs and biometric eye's iris. It is mainly based on pattern
authentication systems. Typically, this recognition methods that identify clear and
system involves different functional phases. distinct patterns in the iris that can
The following diagram shows the process of accurately recognize the target user. This

1
detection system is highly accurate and also people around the world make use of these
performs well. With security breaches and iris recognition mechanisms to ensure the
other forms of authentication fraud on the authenticity of the authentication. It is
rise, it is crucial to have a rigorous biometric observed that they are widely distributed in
authentication system in place. This is financial and secret institutions as well as in
especially useful at key locations like banks places of national importance. This system
and ATMs. Different algorithms are used to is also incorporated in airports. This system
encode the different sets of patterns present has also been implemented in national level
in the iris localization scheme. It helps to examinations, where anyone can take the
know if the user is who they claim to be or if exam illegally. A very useful feature of this
someone else is impersonating the intended system is that it has a very robust and
person. The system manages a database. accurate pattern recognition matching
Typically, the database is large and consists procedure, with false matches being very
of numerous templates of iris features. rare. Furthermore, since the iris is a stable
These are matched by a search engine of part of the interior of the human eye, it is
models for recognition purposes. Many very reliable and useful for this purpose.

LITERATURE SURVEY localization and presentation attack

detection (PAD) using a CNN. Their multi-
Alaa S. Al-Waisy et al. proposed a solution task PAD (MT-PAD) approach computes
to overcome the limitations of single-mode the presentation assault probability from eye
biometric systems by introducing a photos and simultaneously predicts iris
multimodal biometric system. Their bounding box parameters, achieving state-
approach leverages deep learning of-the-art results on publicly available
representations of both left and right irises, datasets. This novel technique is the first to
utilizing ranking-level fusion for efficient detect presentation assaults and iris attacks
and real-time recognition. The IrisConvNet concurrently. Shabab Bazrafkan et al.
architecture employs a Convolutional Neural address the challenge of precise iris
Network (CNN) and a Softmax classifier to segmentation, crucial for embedded
automatically extract discriminative features authentication on mobile devices, by
from iris images. Their discriminative CNN developing an augmentation method for
training scheme, employing back- deep learning systems. This method aims to
propagation and mini-batch AdaGrad enhance iris segmentation performance,
optimization, surpasses previous methods particularly on challenging image datasets,
such as Wavelet transform, Scattering to a level competitive with mobile devices.
transform, Local Binary Pattern, and PCA, Preliminary evaluations indicate significant
achieving a Rank-1 identification rate of speed improvements compared to existing
100% on public datasets, with recognition methods. Abhishek Gangwar and Akanksha
times of less than one second per Joshi present DeepIrisNet, a deep learning-
person.Arun Ross and Cunjian Chen based method for representing iris data.
propose a solution that integrates iris DeepIrisNet addresses scalability and

2
resilience issues in iris detection under less- a deep convolutional neural network
than-ideal conditions by employing an designed for iris recognition systems to
extraordinarily deep architecture and identify contact lenses. Trained extensively
strategies from previous CNNs. It achieves on large datasets, ContlensNet outperforms
state-of-the-art accuracy in simulating iris state-of-the-art systems, yielding an average
microstructures and makes significant improvement in Correct Classification Rate
advancements in cross-sensor recognition (CCR%) of over 10%.
assessment.R. Raghavendra, Kiran B. Raja,
and Christoph Busch introduce ContlensNet,

AUTH TITLE TECHINQU DATASET PERFORMAN LIMITATION

OR E USED CE S
ANALYSICS
S. Improving iris Daugman’s CASIA Iris Accuracy: 98%, High
Kumar, recognition algorithm dataset Sensitivity: computational
A. accuracy using combined 95%, cost due to ANN
Sharma Daugman’s with ANN Specificity: training
(2024) algorithm and 96%
Artificial Neural
Networks (ANN)
P. Iris recognitionDaugman’s UBIRIS Accuracy: 97%, Requires large
Reddy, with enhanced
algorithm database Sensitivity: dataset for
M. feature extraction
with deep 94%, effective ANN
Gupta using Daugman’s ANN for Specificity: performance
(2023) algorithm and
feature 95%
deep learning extraction
R. Addressing noiseNoise- Accuracy: 95%, Struggles with
Singh, issues in iris resistant Sensitivity: heavily occluded
L. recognition using
Daugman’s 92%, iris images
Verma Daugman’s algorithm Specificity:
(2022) algorithm and
integrated 93%
ANN with ANN
M. Enhancing speed Daugman’s CASIA-V4 Accuracy: 96%, High training
Patel, and accuracy in iris Iris dataset Sensitivity: time due to
D. iris recognition
segmentation 93%, complex ANN
Shah systems using
algorithm Specificity: architecture
(2021) Daugman’s combined 94%
algorithm and
with ANN
ANN for
classification
T. Optimizing Daugman’s MMU Iris Accuracy: 97%, Struggles in
Zhang, feature extraction circular database Sensitivity: low-light and
Y. Li in iris recognition segmentation 95%, blurred iris
(2020) using hybrid with ANN Specificity: image conditions
Daugman-ANN for feature 96%

3
 model extraction

system processes should be used.

RELATED WORK

Most notably, the biometric security system

has run into a number of problems and
serious worries. Many people are concerned
about various privacy and security risks Fig1: Various Stages Of Fingerprint Iris
related to biometric authentication Recognition Process
procedures and related technology (Hamd &
Ahmed, 2018). Unfortunately, once Objective:
biometric data has been processed, there is The objective of the proposed work is to
no way to undo the harm or retrieve the develop a robust iris recognition system for
required information. Anyone with access to security-intensive applications. The system
a fingerprint, iris, or ear image effect may employs Daugman's algorithm for accurate
update a compromised password. iris localization and feature extraction from
Consequently, biometrics' fundamental digital images. These extracted features are
operation is still a privacy and security risk then utilized as input for an Artificial Neural
for all these reasons. There are a lot of Network trained using the Levenberg-
presentations about the iris recognition Marquardt backpropagation algorithm. The
system, and some of them have problems primary goal is to achieve a high level of
with the sensor module, the preprocessing accuracy and reliability in iris recognition,
module, and the feature extraction approach. with the reported accuracy reaching 99.7%.
With the correct technology and modern, By leveraging Daugman's algorithm for
state-of-the-art approaches, all of these precise iris segmentation and employing a
security and privacy issues may be properly neural network for feature classification, the
addressed. To further ensure the security proposed system aims to significantly
process, strong passwords and dependable enhance the accuracy and reliability of iris
recognition. The system's design specifically
targets security-intensive applications,
where stringent authentication measures are
required. Previous studies have explored
various iris recognition approaches,
including deep learning, convolutional

4
neural networks, machine learning, and system enhances iris recognition by
support vector machines, yet faced leveraging Daugman’s algorithm for
limitations in accuracy, complexity, and localization and neural networks for
adaptability to diverse conditions. These classification, detecting the pupil using
systems often struggled with complex iris Daugman’s integro-differential operator
patterns and relied on high-quality images (IDO) and training a neural network
and user cooperation, limiting real-world classifier for final classification
applicability. In contrast, the proposed
PROBLEM STATEMENT

The proposed system leverages Daugman's SYSTEM DESIGN

algorithm for precise iris localization and
Artificial Neural Networks (ANNs) for
enhanced classification accuracy in iris
recognition. By extracting and analyzing key
features such as Contrast, Correlation, and
Energy using the ANN, the system achieves
Fig2: Architecture of system.
superior performance in identifying irises.
Notably, it attains an impressive 99.7% The proposed system architecture for iris

success rate on the MMU iris database, recognition is meticulously designed to

showcasing its reliability and effectiveness. achieve precise identification in security-

This innovative approach surpasses previous sensitive applications. It comprises distinct

systems by integrating advanced image yet interconnected modules, each playing a

processing and pattern recognition crucial role in the recognition process.At the

techniques. Daugman's algorithm adeptly forefront is the Iris Segmentation module,

segments the iris, even under less optimal which employs Daugman's algorithm to

conditions, while the ANN ensures swift and accurately delineate the iris region within

precise classification. As a result, the digital images. This step ensures that only

proposed system represents a significant the pertinent iris data is extracted for further

advancement in iris recognition technology, processing. Following segmentation, the

offering improved accuracy and reliability Feature Extraction module comes into play.

for various application. Here, relevant features capturing the

5
intricate details of the iris, such as its texture Daugman’s Algorithm:
and structural patterns, are extracted. These
The Daugman's Algorithm is widely
features serve as the foundation for accurate
recognized as one of the most effective
identification.Once the features are
classifiers in iris recognition, primarily
extracted, the system moves to Pattern
because of its proficiency in accurately
Matching, where the extracted iris templates
segmenting regular shapes with distinct
are compared against a database containing
boundaries. It capitalizes on the inherent
a plethora of stored templates. This
property that regular shapes with clear
matching process involves intricate
boundaries can be efficiently segmented
algorithms to determine the closest matches
using this method. The algorithm entails
based on similarity metrics, ensuring robust
several steps, including Histogram
recognition performance.Subsequently,
Equalization, which enhances iris contrast to
Classification is performed utilizing an
facilitate improved segmentation. Another
Artificial Neural Network (ANN) trained
crucial step is Binarization, which
with the Levenberg-Marquardt
accentuates the disparities between the pupil
backpropagation rule. This neural network
and iris sections while eliminating any
analyzes the extracted features and assigns
interfering objects that might hinder
them to specific classes or identities,
separation performance. Binarization relies
enabling precise identification of the
on the application of an integro-differential
iris.Collectively, these modules form a
operator to precisely locate the contours of
cohesive system architecture that seamlessly
the iris and pupil. The mathematical
integrates iris segmentation, feature
formulation of the Daugman's algorithm
extraction, pattern matching, and
encapsulates its effectiveness in iris
classification to deliver exceptional accuracy
recognition.
and reliability in iris recognition. This
architecture is tailored to meet the stringent
demands of security-intensive applications,
offering robust biometric authentication
capabilities essential for safeguarding
sensitive information and secure
environments.

6
 reducing noise and enhancing the clarity of
the circular contours, which aids in
accurately locating the boundaries between
the iris and pupil.

In the context provided, 'a' represents the

peak of the distribution curve of intensity,
Fig3: Front View of Eye Image
determined by the formulation of a two-
In the Daugman's algorithm, where I(x,y) dimensional intensity function. In two
represents the input eye image, r denotes the dimensions, this intensity function is the
radius being searched, Gσ signifies a product of two Gaussians, each
Gaussian function for smoothing, and corresponding to a specific direction.

s denotes the circle contour determined by r,

x 0, and y 0, the operator conducts a pixel-
wise search across the entire image. It
operates akin to a partial derivative (blurred)
In the given context, 'x' and 'y' represent the
of the integral over circular contours, which
spatial coordinates, while σ2 denotes the
are normalized across different contours.
variance of the random process. The
The boundaries of the pupil, separating it
algorithm's effectiveness in localizing
from the iris, exhibit maximum contour
regular shapes stems from the fact that such
integral derivative—the point where there is
shapes exhibit a sudden intensity peak at
a sudden change in intensity values over the
their contours, maximizing contour
circular borders. The two-dimensional
derivative shapes. In the case of iris
Gaussian Filter employed in Daugman's
segmentation, the segmentation inequality is
𝑅1<𝑠<𝑅2, 𝑅1
algorithm is defined as follows:
governed by where
represents the inner radius of the iris, 𝑅2
denotes the outer radius of the iris, and 's'
Here, σ represents the standard deviation of denotes the region of the iris lying within the
the Gaussian distribution, and x and y region bounded by 𝑅1and 𝑅2. This
represent the coordinates of the filter kernel. inequality effectively separates the iris into a
The Gaussian filter smooths the image, strip, facilitating iris localization. The
7
segmented portion corresponds to a circular
ring patch enclosed within the region
defined by R1 and 𝑅2.

RESULT:

Fig:4 Iris Segmentation.

Fig: 1 Upload Iris Dataset

Fig:5 Matrix Graph by Ann

Fig:2 Image Processing

Fig:6 Iris Classification from Test Image.

Fig:3 Iris Segmentation

Fig:7;Iris images processing.

8
CONCLUSION the proposed iris recognition system, making
it a promising solution for security-critical
In conclusion, the necessity for robust
applications.
security measures in highly classified
environments underscores the importance of REFERENCES
reliable authentication systems to prevent
[1] Alaa S. Al-Waisy et al., “A multi-
unauthorized access. Iris recognition stands
biometric iris recognition system based on a
as a formidable biometric authentication
deep learning approach”, Springer 2018
method, leveraging advanced mathematical
techniques to analyze digital images of the [2] Cunjian Chen , Arun Ross et al., “A
iris. Central to this approach is Daugman's Multi-task Convolutional Neural Network
algorithm, renowned for its ability to for Joint Iris Detection and Presentation
accurately segment iris shapes while also Attack Detection”, IEEE 2018
effectively smoothing out noise. In this
[3] Shabab Bazrafkan et al., “Enhancing iris
proposed work, iris localization through
authentication on handheld devices using
Daugman's algorithm is coupled with feature
deep learning derived segmentation
extraction, capturing distinct iris patterns
techniques”, IEEE 2018
such as contrast, correlation, and entropy.
These features, though exhibiting unique [4] Ritesh Vyas, Tirupathiraju Kanumuri,
behaviors across different iris images, may Gyanendra Sheoran, Pawan Dubey,
occasionally overlap. Nonetheless, they “DeepIrisNet: Deep iris representation with
serve as valuable inputs for a neural network applications in iris recognition and cross-
trained via the Levenberg-Marquardt sensor iris recognition”,, IEEE, 2017.
backpropagation rule. By training the neural
[5] R. Raghavendra et al., “ContlensNet:
network with feature values from authorized
Robust Iris Contact Lens Detection Using
images and subsequently testing for
Deep Convolutional Neural Networks”,
accuracy, the proposed system achieves an
IEEE 2017
impressive 99.7% accuracy rate, surpassing
previous systems utilizing the same MMU [6] Nianfeng Liu et.al, “Accurate iris
database. This heightened accuracy segmentation in noncooperative
underscores the efficacy and reliability of environments using fully convolutional
networks”, IEEE 2016

9
[7] MariaDe Marsico et al., “Iris recognition [15] Navjot Kaur , Mamta Juneja, “A review
through machine learning techniques: A on Iris Recognition”, IEEE 2014.
survey”, IEEE 2016
[16] Heinz Hofbauer et al. , “A Ground
[8] Sushilkumar S. Salve et al., “Iris Truth for Iris Segmentation, IEEE 2014
recognition using SVM and ANN”, IEEE
[17] Daniela Sánchez et al., “Modular
2016
granular neural networks optimization with
[9] Shervin Minaee et al., “An experimental Multi-Objective Hierarchical Genetic
study of deep convolutional features for iris Algorithm for human recognition based on
recognition”, IEEE 2016 iris biometric”, IEEE 2013

[10] Shervin Minaee , AmirAli Abdolrashidi [18] Nuzhat Faiz Shaikh et al. , “Improving
et al., “Iris recognition using scattering the Accuracy of Iris Recognition System
transform and textural features”, IEEE 2015 using Neural Network and Particle Swarm
Optimization”, Citeseer 2013
[11] Pedro Silva et al., “An Approach to Iris
Contact Lens Detection Based on Deep [19] Ching-Han Chen et al., “Optimal fusion
Image Representations”, IEEE 2015 of multimodal biometric authentication
using wavelet probabilistic neural network”,
[12] Firoz Mahmud et al., “PCA and back-
IEEE 2013
propagation neural network based face
recognition system”, IEEE 2015 [20] Mrunal M. Khedkar et al., “Robust
human iris pattern recognition system using
[13] Vivek Srivastava et.al , “Biometric
neural network approach”, IEEE 2013
recognition by hybridization of evolutionary
fuzzy clustering with functional neural
networks”, Springer 2014 .

[14] Miloš Oravec, “Feature extraction and

classification by machine learning methods
for biometric recognition of face and iris”,
IEEE 2014

10