Research Article

Optimized Algorithms Based Iris

Recognition System
Amanpreet Kaur1, Anujpreet Singh2

Abstract
Human Being’s eyes are recognized as the most efficient biometric trait. Because of its great
accuracy, reliability as well as originality iris recognition is included in numerous fields associated
with access control as well as security. For efficient working of the iris recognition system the
researchers have to face many challenges in designing an effective system. In this article the author
discusses the new way of individual identification depending upon the iris recognition developed by
many researchers. It includes various techniques used by them and their effectiveness in iris
recognition process.

Keywords: Artificial neural, Gabor filter, Genetic algorithm, K-Mean clustering, Log Gabor filter,
Network

Introduction
Growth of information technology along with risks upon security has increased the need of a robust as well as
extremely accurate human recognition system. Authentication of any human can be carried out by means of
password, biometrics or even smart cards. Keeping in mind password and handling smart card is tough, therefore
biometrics dependent authentication system is extremely recommended. Biometrics means the scientific discipline
which includes identifiers such as fingerprints, iris, face and also voice that are unique measurable features utilized to
tag or explain a person’s identity. The biometric traits are unique to a person because they are hereditary in nature
thus turned out to be more accurate as compared to traditional human authentication schemes.12 This system collects
the biometric features of an individual, elicit-ate the characteristics with the gathered biometric traits and matches
the elicited features together with the one saved in the database, in order to confirm the claimed identity. Iris is
recognized as more stable as well as distinctive as the design and structure involving iris left constant in course of life.
The possibility that two different individual possess similar iris is 1 in 1078 (Morimoto et al., 2005). There are multiple
problems throughout entire procedure of Iris Recognition in the form of obtaining high quality picture, segmentation
of the concentric boundaries, scalability, and appropriate characteristic representation as well as matching. John
Dougman proposed first iris recognition system; afterwards various techniques had been proposed to enhance this
biometric system. The procedure of iris recognition includes some pre-processing phases like localization,
segmentation as well as normalization and also some filtering methods are employed to produce iris templates. The
final binary template (iris code) can be generated by quantizing the phase details included in the filtered normalized
image. The acquired binary template is then saved in the database. Bit-based metrics such as hamming distance is
utilized for assessment of iris codes during iris recognition Inverse Biometrics refers to the procedure of reconstructing
or maybe re-engineering the biometric template to acquire details regarding the identification related to the
template.5
1
Department of Computer Engineering and Technology, Guru Nanak Dev University, Amritsar, Punjab, India.
2
Department of Electronics and Communication Engineering, DAV Institute of Engineering and Technology, Jalandhar, Punjab,
India.
Correspondence: Ms. Amanpreet Kaur, Guru Nanak Dev University, Amritsar, Punjab, India.
E-mail Id: [email protected]
How to cite this article: Kaur A, Singh A. Optimized Algorithms Based Iris Recognition System. J Adv Res Image Proc Appl 2017;
4(1&2): 8-15.

© ADR Journals 2017. All Rights Reserved.

 J. Adv. Res. Image Proc. Appl. 2017; 4(1&2) Kaur A et al.

In today’s situation in which the biometric systems are for forensic applications, etc. Within this article, a
taking over the traditional authentication techniques, different scheme for iris recognition has been explained
inverse biometrics provides for a critical security threat. using multiple classifiers. Performance of the every
Since iris recognition is utilized in offering single employed selection technique is calculated using
authentication or personal identification for extremely parameters like false real rate, false acceptance rate,
confidential as well as secure systems including airports, etc.
Literature Survey
Table 1.Comparison of Different State of Art
Authors Year Reference No. Parameters Techniques Strengths Limitations
John D January [1] Pupillary and Band pass 2-D filter For badly focused Not so
2004 the outer kernel, 2-D wavelet images of eye, accurate
(IEEE) (limbus) demodulation Demodulation because of
boundary of phase sequence false matches
iris bits are set
High speed
performance
Araghi March [2] Covariance Discrete wavelet Simulation results Less amount
LF, et 2010 matrix transform by are very promising of data set
al. (IMECS) utilizing This method offers used
Competitive Neural good class
Network (LVQ) discriminacy
Sibai FN 2010 [3] RGB values Feed forward High accuracy up Only a single
et al. (Elsevier) extracted artificial Neural to 93.33% Iris was
from the iris network with the processed
jpeg image back propagation instead of
algorithm Whole pair
Raja SV 2013 [4] Accuracy rate, Genetic algorithm High accuracy rate In the starting
et al. (IJCA) average time Neural network and reduced stage, the
learning time is network
achieved started
mugging up
rather than
learning as
well as
predicting
Galbally 2013 [5] Motion blur, Genetic algorithm High level of Not much
J et al. (Elsevier) contrast, matching among accurate
occlusion real samples
Rai H et 2013 [6] Zigzag Support vector Efficient for both Small dataset
al. (Elsevier) collarette area machine, hamming identification as used
of iris distance, parabola well as verification
detection, trimmed
median filter
De 2014 [7] Mean, relative Histogram, FIRME provides a Behave
Marsico (Elsevier) error Hamming distance solution for both negatively
M et al. CSUM algorithm face spoofing because of
detection, and for bad
optimal sample acquisition,
selection, by sensor
utilizing iterative problem,
entropy evaluation incorrect
device
orientation

9
 Kaur A et al. J. Adv. Res. Image Proc. Appl. 2017; 4(1&2)

Raja KB 2014 [8] Equal error Deep sparse Real life dataset Not so much
et al. (Elsevier) rate (EER), filtering usedProvide high accurate since
contrast, degree of Robust accuracy is
diameter features 75%
Sánchez 2015 [9] Percentage of Modular neural Optimizes the Division of
D et al. (Elsevier) complexity network, genetic multiple modular tasks cannot
algorithm neural networks Be done
which utilize
different data
points in the
training phase,
providing multiple
results in the same
evolution
Bhateja 2015 [10] Sparse k-nearest distance Highly accurate
AK et (PATREC) Concentration classifier, and efficient
al. Index, cumulative sparse method
Confidence concentration Combines the
Index index based strengths of three
classifiers, sector classifiers
based classifier,
genetic algorithm,
canny edge
detection, hough
transform
Ibrahim 2016 [11] False Libor Masek and Accuracy rate is Error rate
AA et (JCC) acceptance genetic algorithms, higher increases with
al. rate, and Principal the
correct Component increase in
recognition Analysis (PCA) number of
Rate, False samples
Rejection Rate and the
recognition
rate
decreases
when large
samples are
utilized
Kaur R 2016 [12] False Genetic Algorithm Provides high Small amount
et al. (IJCET) acceptance accuracy rate and of data set
rate, false real low error used
rate

Figure 1.The Components of an Eye

10
 J. Adv. Res. Image Proc. Appl. 2017; 4(1&2) Kaur A et al.

Basic Techniques picture is employed with edge detector in order to

produce edge map. Gaussian filter is utilized to choose
Iris Recognition Process the appropriate scale of edge analysis as well as smooth
the image.15 Accuracy depends relies on the number of
Iris Recognition process includes pre-processing accumulator cells, Amount of memory needed raise up
measures such as Acquisition, Segmentation, as well as due to increase of cells.
Normalization and the encoded normalized images are
matched with the claimed identity. In iris recognition K-mean Clustering Technique16
various steps involved are-
It split the eyes picture within 3 different parts
Image Acquisition dependent on the values of intensity. 1st region includes
iris along with pupil as well as eye lashes possessing
An eye’s picture is obtained by using a sensor or little value of intensity. 2nd area contains bigger values
perhaps camera or some other image acquisition tool. with intensity composed of sclera region as well as
luminance reflections. 3rd is the skin area in between 2
Image Segmentation/Localization areas. This technique is beneficial because primary
Captured image of an eye is segmented to differentiate dilemma is within the dark area. The benefit of
particular iris region from remaining section of the eye. employing this technique is that the useful regions in
the iris region are not lost because of using arc.
Canny Edge Detector13
Normalization
This technique was adopted in order to identify edges of
iris therefore three primary performance associated The annular-like iris pictures of the eye is transformed
into rectangular normalized images Fig. 4, 5 which are
standards were recognized.
the representation of segmented iris picture from
i. Superior Recognition: The algorithm must identify Cartesian into pseudo-polar coordinates.
all possible actual edges within image.
ii. Superior Segmentation: Within the actual picture Feature Extraction
the edges must be as near as it can be to the circled
It is the essential phase of recognition of iris utilized to
edges.
find out iris representation.
iii. Minimal multiplicity response to single edge as well
as image must not produce fake edges. Log-Gabor Filter17
Circular Hough Transform14 Numerous modifications are included in the
complicated texture of iris. Iris characteristic details are
It is extremely efficient algorithm therefore it is
acquired through break down of the picture among
employed to discover the iris boundaries. Large
complicated-valued phase coefficients of iris. One
calculations represents the three stage iteration
dimension filter is used to each and every portion of
problem within Hough transform, each (p, q) point
normalized design of iris so as to achieve suitable
edges in the image space votes for point (s, t, r) for
characteristic details.
every feasible circle moving from it within the
parameter space, here s, t are the coordinates of the The whole iris recognition process can be summarized
center of the circle and r’ is its radius. Gray scale iris using Fig. 4.

11
Kaur A et al. J. Adv. Res. Image Proc. Appl. 2017; 4(1&2)

Figure 2.Normalization of Annular-like Segmented Iris Image

Figure 3.Normalization Process

12
J. Adv. Res. Image Proc. Appl. 2017; 4(1&2) Kaur A et al.

Figure 4.Flowchart of Iris Recognition Process

Artificial Neural Network Selection

Artificial neural network (ANN) is a self-learning The selection operator allows selecting a parent
approach that imitates the properties of biological chromosomes based on their related fitness values. The
nervous systems. It is a system that changes its chromosomes having the best fitness values are
parameters based on external or internal information forwarded to the next generations. This sort of selection
that flows through the network during the learning is referred as elitism. There are various selection
phase. operators available for usage among which three main
selection operators utilized in this article are: 2.1 Rank
Genetic Algorithms Selection: In this, the population is categorized based on
the fitness values. The fitness allocated to every single
A fundamental genetic algorithm consists of three chromosome relies upon its location among the
operators: Selection, Crossover, and Mutation. The individual’s rank and not on the actual fitness value. The
algorithm begins by considering a set of chromosomes probability of every single chromosome getting selected
(usually the elements or units on which we want to for mating is actually their fitness normalized with the
apply genetic algorithms), called population. The overall fitness of population.
chromosomes of the present population are employed
to produce offspring which is forwarded to the next Roulette Wheel Selection
generation or population hoping that newly generated
offspring chromosomes will be better in comparison to The chromosomes tend to be mapped to contiguous
parent chromosomes.18 sections of a line in a way that each portion is equal in
size to its fitness. Any arbitrary number is produced, and
Fitness Evaluation the chromosome whose section spans the random
number is selected. Till the needed number of
While producing offspring chromosomes, initially the chromosomes is obtained (called mating population),
fitness of all the chromosomes within consideration is the procedure is repeated another time.20
determined by means of predefined fitness function.

13
Kaur A et al. J. Adv. Res. Image Proc. Appl. 2017; 4(1&2)

Stochastic Universal Sampling (typically Log Gabor filtering technique) before

quantizing the phase response of the output that is
The chromosomes are mapped to contiguous sections of filtered and representing the iris images in binary
a line, in a way that each chromosome’s portion of the form.19
segment is equal to its fitness value.
Matching
Encoding and Matching
In this process two iris codes i.e. the iris code of target
Encoding iris image and that saved in the database are compared
using bitwise operators like hamming distance. The
Encoding procedure maps the normalized iris image to encoding and matching process can be summarized in
binary representation. Many encoding methods are Fig. 5.
available and most of them utilize filtering techniques

Figure 5.Encoding and Matching Process

Conclusion 2. Araghi LF, Shahhosseini H, Setoudeh F. IRIS

recognition using neural network. International
The colored ring of eye is called iris which made by Multiconference of Engineers and Computer
textured tissue that fills area of the pupil of the eye used Scientists, Hong Kong. 2010; 1.
for identify a person. No two single persons have same 3. Sibai FN, Hosani HI, Naqbi IM, et al. Iris recognition
iris patterns .Iris of both the eyes are different. Iris using artificial neural networks. Expert Systems with
based recognition system provides the accurate and Applications 2011; 38(5): 5940-6.
authentic result. Its working speed is also high. Iris 4. Raja SV, Rajagopalan SP. IRIS recognition system
recognition system has similar working with retinal using neural network and Genetic Algorithm.
recognition system. The only difference between two is International Journal of Computer Applications
that retinal recognition system uses input acquisition 2013; 68(20): 49-53.
subsystem. Iris recognition takes only few seconds to 5. Galbally J, Ross A, Gomez-Barrero M, et al. Iris
produce record and store the result in database so that image reconstruction from binary templates: An
it may work as cache. Before recognizing the new iris, efficient probabilistic approach based on genetic
we match the entries in database. For future recognition algorithms. Computer Vision and Image
and verification, database must be maintained. Contact Understanding 2013; 117(10): 1512-25.
lens and spectacles does not affect the quality of iris 6. Rai H, Yadav A. Iris recognition using combined
image. 256 bytes are used for storing the value of iris support vector machine and Hamming distance
code. approach. Expert Systems with Applications 2014;
41(2): 588-93.
References 7. De Marsico M, Galdi C, Nappi M, et al. FIRME: Face
and iris recognition for mobile engagement. Image
1. John D. How iris recognition works. Transactions on and Vision Computing 2014; 32(12): 1161-72.
Circuits and Systems for Video Technology 2004; 8. Raja KB, Raghavendra R, Vemuri VK, et al.
14(1): 21-30. Smartphone based visible iris recognition using

14
J. Adv. Res. Image Proc. Appl. 2017; 4(1&2) Kaur A et al.

deep sparse filtering. Pattern Recognition Letters and computationally efficient approach for
2015; 57: 33-42. unconstrained iris segmentation. Image and Vision
9. Sánchez D, Melin P, Castillo O. Optimization of Computing 2010; 28(2): 261-9.
modular granular neural networks using a 15. Fatt Ng RY, Tay YH, Mok KM. A review of iris
hierarchical genetic algorithm based on the recognition algorithms. Transactions Information
database complexity applied to human recognition. Technology 2008; 2: 1-7.
Information Sciences 2015; 309: 73-101. 16. Sahmoud SA, Abuhaiba IS. Efficient iris
10. Bhateja AK, Sharma S, Chaudhury S, et al. Iris segmentation method in unconstrained
recognition based on sparse0 representation and k- environments. Pattern Recognition Letters 2013;
nearest subspace with genetic algorithm. Pattern 46(12): 3174-85.
Recognition Letters 2016; 73: 13-8. 17. Rankin DM, Scotney BW, Morrow PJ, et al. Iris
11. Ibrahim AA, Khalaf TA, Ahmed BM. Design and recognition failure over time: the effects of texture.
implementation of iris pattern recognition using Pattern Recognition Letters 2012; 45(1): 145-50.
wireless network system. Journal of Computer and 18. Chen R, Lin X, Ding T. Liveness detection for iris
Communications 2016; 4(7): 15. recognition using multispectral images. Pattern
12. Kaur R, Kaur P. Optimizing the genetic algorithm to Recognition Letters 2012; 33(12): 1514-9.
resynthesize iris patterns from iris Code Template. 19. Zuo J, Schmid N, Chen X. On generation and analysis
International Journal of Current Engineering and of synthetic iris images. Transactions on Information
Technology 2016; 6. Forensics and Security 2007; 2(1): 77-90.
13. Canny J. A computational approach to edge 20. Cui J, Wang Y, Huang J, et al. An iris image synthesis
detection. Transactions on Pattern Analysis and method based on PCA and super-resolution.
Machine Intelligence 1986; 8(6): 679-86. International Conference on Pattern Recognition,
14. Chen Y, Adjouadi M, Han C, et al. A highly accurate UK. 2004. DOI: 10.1109/ICPR.2004.1333804.

15