Content uploaded by Jarina Mazumdar
Author content
All content in this area was uploaded by Jarina Mazumdar on Jun 21, 2018
Content may be subject to copyright.
DOI: http://dx.doi.org/10.26483/ijarcs.v9i1.5322
Volume 9, No. 1, January-February 2018
International Journal of Advanced Research in Computer Science
REVIEW ARTICLE
Available Online at www.ijarcs.info
© 2015-19, IJARCS All Rights Reserved 711
ISSN No. 0976-5697
RETINA BASED BIOMETRIC AUTHENTICATION SYSTEM: A REVIEW
Jarina B. Mazumdar
Department of ECE Gauhati University
Guwahati, India
S. R. Nirmala
Department of ECE Gauhati University
Guwahati, India
Abstract: A biometric system provides an automatic person authentication based on some characteristic features possessed by the individual.
Among all other biometrics, the retinal biometric system is unique as well as stable. The retina is a secure and reliable source of person
recognition as it lies behind the eye and is unforgeable. The process of recognition mainly includes pre-processing, feature extraction and then
features matching. The features generally used in this process are either blood vessel features or non-blood vessel features. In this paper,
different methods available in the literature for retina based person authentication system are described, discussed and compared.
Keywords: Biometric; retinal recognition; blood vessel segmentation; vascular feature; non-vascular feature.
I. INTRODUCTION
With the advancement in technology, the world is
exposed to more threats and insecurities. In many fields, such
as police or military environments and banking affairs,
security is a critical issue. This requires a highly dependable
and precise person authentication system. Traditional
authentication systems are either information-based such as a
password, a pin or token based like a card, a key. In many
conditions, these systems are not sufficiently reliable due to
their common impotence to individuate between a true
authorized user and a user who illicitly procured the privilege
of the authorized user. These problems can be solved by
employing a biometric based authentication approach.
A biometric system is nothing but a pattern recognition
method, which recognizes a person depending on some
specific features, derived from the physiological or
behavioral characteristic that a person possesses [1].
Physiological characteristics include face; fingerprints, iris,
and hand and finger geometry [2] are defined below:
Face: Face authentication system analyzes facial
characteristics. A digital camera is used to capture a facial
image and some features are extracted from it which is then
used for authentication.
Fingerprint: A fingerprint is an impression of the
friction ridges of all or any part of the finger.
Iris: Iris is a colored ring of tissue that surrounds the
pupil. The feature extracted from these regions are used in
Iris based biometric.
Hand and finger geometry: Hand Geometry involves
scanning and measuring the shape of the hand. These
techniques include the estimation of length, width, thickness
and surface area of the hand.
The behavioral characteristics are ear shape, voice, gait,
signature, and key stroking [2].
Ear shape: The ear shape and the structure of the tissue
of the pinna are unique to every individual. The approaches
for ear recognition are based on matching the distance of
prominent points on the pinna from a landmark location on
the ear.
Voice: Voice is also a physiological trait because every
person has a different pitch, but voice recognition is mainly
based on the study of the way a person speaks, commonly
classified as behavioral.
Gait: Gait biometric is the particular way that a person
walks and is an intricate spatiotemporal biometric. Although,
gait is not supposed to be very distinctive but is favourable in
some low security applications to allow verification.
Signature: Signature verification checks the way a user
signs his/her name. Like signature’s static shape, some other
signing features such as speed, velocity, and pressure are also
countable for this verification.
Keystroke Dynamics: Keystroke dynamics is a
biometric characteristic to verify the identity of a person by
their typing rhythm, which can handle with the trained typists
and the amateur two-finger typist as well.
Although there are some authentic systems based on
different biometrics but all of these can easily be forged.
However, several spoof attacks can also be carried against
many types of biometrics, like fingerprint, face, and iris.
Under spoof attacks replica of the genuine user’s fingers are
created using some artificial material like silicon or wood
glue, etc. to circumvent a system [3]. Sometimes the
attackers use an artificial finger, a mask over a face or a
contact lens on an eye [4]. As one’s signature may alter over
a period and it is not as unique as Iris, Face, Fingerprint [5].
In voice recognition, the main obstacle is speech features are
sensitive to background noise. Gait biometric also may vary
due to change in body weights. It may not remain fixed for
long period of time, owing to major injuries involving leg or
brain or due to inebriety. Hence, it is required to consider
such a biometric characteristic, which cannot be forged
easily but is unique, stable and measurable. This leads to
retina based biometric authentication system.
As the retina of human eye lies at the rare end and is not
directly attainable, it is very difficult to forge. The retinal
vasculature is unique to every individual and it remains same
throughout a person's life unless the person undergoes some
severe eye surgery or injury [2].
The first stage of a biometric system is enrollment phase,
where each user has to register his or her biometric trait into
the system database. Then feature is extracted from the
enrolled data and a template model is been made for each
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 712
user and stored in the database. A biometric authentication
system is generally employed in verification and
identification mode depending on the application
environment. The vital difference between these two modes
of operation is in the testing phase. In the verification mode,
a user's requested identity is been authorized by comparing
the extracted biometric features with the stored biometric
template of the person in the database. In the identification
mode, the system validates an individual by checking the
templates of all the users in the database for a match [7].
For a biometric authentication system the general block
diagram is shown in Fig.1. The sensor module takes an
image of the biometric characteristic then features are
extracted from the image. The extracted feature information
is called template. This template is then sent to a matcher,
where the matcher compares the newly presented biometric
information with the previously stored template to make a
decision.
Figure 1. General block diagram of biometric authentication system
II. REVIEW OF DIFFERENT RETINA BASED
BIOMETRIC AUTHENTICATION METHODS
The retinal authentication system is mainly based on the
unique blood vessel pattern in the retina. The uniqueness of
retinal vascular pattern was introduced by two
ophthalmologists, Dr. Carleton Simon and Dr. Isodore
Goldstein in 1935 [8]. And later in 1950, Dr. Paul Tower
discovered that even among identical twins the vascular
patterns are unique [8]. EyeDentification 7.5, the first retina
identification system using commercial retinal scanner was
proposed by EyeDentify Company in 1976. Another retinal
scanning device named ICAM 2001 is also manufactured by
this company, which can store up to 3000 templates with a
storage capacity of up to 3300 history transactions [9]. Some
companies like Retica System Inc. working on a multi-modal
system using Retina and Iris for identification. A retinal
authentication system has basically three main steps: image
acquisition, feature extraction, and feature matching. These
are explained below:
The first stage of a retinal authentication system is image
acquisition. To capture the image of retina a fundus camera
is used, where the user must position his eye very close to the
lens and the user must also remain perfectly still at this point.
Moreover, if the person is wearing glasses then these has to
be removed to avoid signal interference.
The second stage is feature extraction. Different features are
extracted from the blood vessel structure of the retina image,
and/or the retinal image itself. These features are unique for
each individual and stored as a reference pattern.
At the last stage, the features of test image are matched
with the pre-stored template in the database depending on
some criteria. If the criteria are satisfied then the person is
authenticated.
The retina based authentication procedures can be broadly
classified into two different categories. Few algorithms
depend on the vascular features of retinal images. These
processes require blood vessel segmentation, which is a very
time consuming. Hence, the computational time of these
methods is high. The different blood vessel features are listed
below:
Branch point: The vascular structure of retinal
image is of tree shape. So, the main features of the
blood vessel pattern are branch point where a
vessel bifurcated into two vessel branches [10].
End point: The end points of the blood vessel
patterns, where each vessel ends, is considered as
another feature for retina based authentication
system [10].
Crossover point: The points of intersection
between two blood vessels are termed as crossover
points and are taken as another feature set [11].
Blood vessels in and around Optic Disc (OD):
The blood vessel patterns around the OD region are
also considered as a feature set because of their
stability and unique variation inside that region for
a particular person [12].
However, some other algorithms are based on non-
vascular features extracted from retinal images. They are
structural features of the retinal image like luminance,
contrast, and structure. Some algorithm computed the retinal
edge dissimilarity score for authentication. Whereas others
used Fourier transform of retinal image to extract Fourier
spectrum phase feature (FSPF). Corners of retinal image are
also considered as one of the features. The implementation
time of these methods is low, as this algorithm does not
require blood vessel segmentation for feature extraction.
A. RETINAL AUTHENTICATION BASED ON
VASCULAR FEATURES
Most of the retinal authentications systems are based on
blood vessel features, as these a unique to each and every
individual. The block diagram of this kind of authentication
system is shown in Fig.2. Different techniques exist in
literature for enhancement and segmentation of retinal blood
vessels that are used to extract the feature. Few of them are
discussed below:
Ortega et al. [10] have extracted the blood vessel using
MLSEC (multi-local level set extrinsic curvature) which has
useful invariance properties. Then they have segmented the
OD using fuzzy circular Hough transform and extracted the
ridge endings and ridge bifurcation from vessels inside the
optical disc as their feature points. For feature matching, the
affine transformation parameters are obtained to associate the
query image and its best corresponding enrolled image. The
computational time of this method is low as they have
extracted the feature points inside the OD but, the main
limitation of their algorithm is the use a very small database
with 14 subjects only. In their consecutive work [11], they
have considered one more feature point which is the
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 713
Figure 2. Block diagram of retinal authentication system based on
vascular features
They have done a deep analysis of similarity metric to
identify the most suitable matching pair. But this method
works only for the verification of a person’s identity. Farzin
et al. [13] have proposed a novel recognition method based
on the features of retinal images. The main steps of the
method are: blood vessel segmentation, feature extraction,
and feature matching. The OD of the image is detected and a
circular region around the OD is selected from the segmented
blood vessel image. Then, a rotation invariant template is
created, using a polar transformation as shown in Fig.3.
Figure 3. Polar transformation: (a) Region of interest in Cartesian
coordinates, (b) polar representation of the image [13].
Figure 4. (a) Multiscale analysis of polar image: wavelet approximation
coefficients in scale 3 (left up), 2 (left middle), and 1 (left bottom); (b) vessel
separation result: large (right up), medium (right middle), and small (right
bottom) vessels [13].
Next, these templates were tested in three different scales
using wavelet transform to separate the vessels according to
their diameter sizes shown in Fig. 4. In the last stage, vessel
location and direction in each scale are used to form a feature
vector for each subject in the database. For feature matching,
they have introduced a modified correlation measure to
obtain a similarity index for each scale of the feature vector.
These steps made the computational time of the system high,
and this method is used only in identification mode.
Figure 5. a) Original retinal image, b) Enhanced image, c) Segmented
blood vessels, d) Thinned blood vessels [14].
In papers [14], [15] and [16], the authors have used Gabor
wavelet to enhance the blood vessels of retinal images as it
has the capability of selecting oriented features [17].
However, Akram et al. [14] have used an adaptive
thresholding technique to segment the blood vessels. But it is
hard to find one optimal threshold value for accurate vascular
extraction, especially in case of PDR (Proliferative Diabetic
Retinopathy) as new abnormal blood vessels are normally
very thin. So, Qamber et al. [15] and Fatima et al. [16] have
presented a supervised multilayered thresholding based
method for accurate vascular segmentation where by varying
the threshold value they can segment the thin blood vessels
also. They have shown that their blood vessel segmentation
technique outperforms existing methods [15]. The blood
vessel segmentation result is shown in Fig.5. Then for feature
extraction, all the authors have used vessel end points and
branch points. Then they validated their extracted blood
vessel features to remove false endings and bifurcations due
to small breaks and spikes. For feature matching, Akram et
al. [14] have used Euclidian distance; Qamber et al. [15] and
Fatima et al. [16] have used Mahalanobis distance. For
recognition [14], [15] they have taken 354 images from
DRIVE, STARE and VARIA databases and got a quite good
recognition rate for DRIVE (100%) and VARIA (99.5%). In
STARE database, the recognition rate is 96.29% because of
the presence of different retinal lesions but, in [16], 333
images were used from VARIA and RIDB database. Other
than the recognition rate they have also used False
Acceptance Rate (FAR), False Rejection Rate (FRR) and
calculated Equal Error Rate (ERR) (0.0557%) to further
check the validity of their proposed system.
The main difficulty in the retinal recognition system is
due to the head or eye movement. To overcome those
difficulties, it is required to design an authentication system,
which will be invariant to scale, rotation and translation. In
[18], Kose et al. introduced a scale, rotation and translation
invariant person recognition system using retinal vascular
pattern. To tolerate these factors the proposed method used a
technique utilizing the distances between vessels. The
method counted the number of corresponding vessels
encountered along the processing line on the current and
reference images stored in the database. For similarity
measurement, they have employed two measures. The first
one counted the matching vessel and calculated a similarity
value for the reference and sample images; here, the
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 714
maximum number of matching are found for each sample
lines and then, the total number of matching for all sample
lines are calculated for the identification. The second one
used a rational similarity measurement of reference and
sample images. Here the maximum numbers of
corresponding vessels are counted. Finally, by considering
these two similarity measurement techniques, they calculated
the final similarity value for the images. The maximum value
of comparisons between sample and reference images shows
the best matching between them. Their method can identify a
person with degenerated retina, most of the degeneration
occurs due to Diabetic Retinopathy (DR) or Age-related
Macular Degeneration (AMD). The proposed method was
found to be very fast (milliseconds) in person identification
and without any user intervention.
Ekka et al. [12] proposed a retinal verification system
based on point set matching. In their method, they have
observed that the blood vessels inside OD region are more
stable and exhibit unique variation for each individual and
for reason they segmented the vascular pattern inside that
region only. The edge map of the OD blood vessels is used
as a feature set for similarity measurement and for this; the
authors have used a Hausdorff distance measure, which is an
edge dissimilarity measure. The main difficulty of using
Hausdorff distance technique is when a portion of an object
is missed due to occlusion the mismatch is large. To
overcome this, a partial Hausdorff distance method is
introduced to calculate the similarity between two feature
maps of the vascular structure of retina inside the OD
region. The performance of their system was measured in
terms of FRR, FAR and calculated EER as 9.79% for an
optimum threshold. They have got the highest accuracy of
90.21%, which is not quite satisfactory, and also their
method works only for retinal verification.
Khakzar et al. [19] have proposed person identification
system using retinal images. Their method is rotation-
invariant based on a tessellation of frequency spectrum.
They have used Discrete Fourier Transform (DFT)
technique to make the features invariant to translation and
rotation. By analyzing the Fourier spectrum of the retina
image it has been observed that the density of useful
information in retina image is reduced from low to high
frequency. Hence, they have implemented a multi-resolution
tessellation algorithm, where a sequence of radii has
tessellated the Fourier domain into concentric circles. Then
the Fourier transform of retina image is divided into
different sub band and the energy of each sub band was
taken as a feature vector. For decision-making, they have
used Euclidean distance. Though they have obtained a quite
good accuracy rate of 99.29% but the performance of the
algorithm has been evaluated using a small data set.
Lajevardi et al. [20] have introduced a Biometric Graph
Matching (BGM) algorithm to make a retina verification
system. They have generated a retinal template based on the
spatial graphs derived from the blood vessel structure of the
retina. For these, the vascular structures of the retinal image
are segmented first by using a family of 2D matched filters.
The features like vessel branch and crossover point are
extracted. Then a retina graph is formed by connecting these
feature points and it consists of vertex set, edge set, vertex
labeling function and edge labeling function and is shown in
Fig.6. In the graph matching technique the graph is first
registered and then error-tolerant graph matching technique
differentiates between genuine and imposter. To measure
the performance of the algorithm, a Kernel Density
Estimation (KDE) technique is used to obtain the authentic
and imposter score distributions of the training set as the
accessible data set is small. Finally, to increase the matching
performance of the verification system, a combination of
SVM classification and KDE model is used. This method
have used multiple characteristics of the scores of retina
graphs in the course of decision making whereas previous
state-of-the-art retinal verification architectures have used a
single measure alone to distinguish the retinal models [11],
[13]. The validation of their algorithm is evaluated in terms
of Equal Error Rate (ERR) for both automatic and manual
retina graphs and these were 0.005 and 0.001 respectively.
Figure 6. (a) Binarised image of the Retinal image (b) The skeleton of the
retinal image with spurious feature points which have short paths ending in
termination points (red marks) and true feature points (yellow marks). (c)
Retina graph of (b) after removal of spurious points [20].
Hussain et. al. [21] have proposed a person identification
based on the retinal vascular network features. These are
branch point and crossover point. To extract these features
they have segmented the blood vessels using multi-scale line
detector algorithm. They have applied a vessel width
measurement technique to extract the major vessels then
identified the vessel branch and crossover points for the
major vessels. To make the extracted retinal features
invariant to scale, rotation and translation they have used a
geometric hashing technique. This approach uses indices
based on local Geometric features, which remained invariant
to object transformation. Their method is invariant to scale,
rotation and translation and also achieved a detection
accuracy of 100%. Condruache et. al. [22] have proposed a
novel feature extraction process that includes the
segmentation of feature point using scale invariant feature
transform (SIFT). To make the feature set they have also
considered vessel branch point and crossover point. At first,
they have segmented the vascular structure of the retinal
image using a hysteresis classifier. Then they have evaluated
their corresponding SIFT descriptors. Their feature vector is
associated to the statistical properties of the SIFT descriptor
sample in the feature-point cloud which exhibit all the
invariance properties needed for the retina based person
authentication. They have also shown that in their
approaches it is not required to align the two images to
check whether they have come from the same eye, which
clearly reduces the computational burden of the system.
After feature extraction, the authentication is conducted with
the help of sparse classifier.
Zahedi et al. [23], authors have used the Radon
transform on retinal image for designing a person
identification system. The blood vessel features around the
OD region of the retinal image are considered as a template.
For these, they have localized the OD first, using a template
matching technique then to make the template rotation
invariant polar transformation has been used. Then they
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 715
have segmented a circular ring around the optic disc and
extracted features using radon transform. For matching they
have used 1D-DFT (Discrete Fourier Transform) and
Euclidean distance. They tested their method on 200 images
by generating 5 images for every 40 subjects from DRIVE
database and achieved an identification rate of 100%. After
going through all these literature it has been seen that all
these methods have considered different sets of images from
different databases and evaluated their results using different
performance metrics. These methods are compared
depending on the number of images, number of subjects,
authentication mode (Identification or verification),
processing time, the rate of EER and Accuracy. These
comparison result using blood vessel feature of the retina are
shown in Table 1.
Table I. Comparison Table of different Recognition methods using Vascular feature
Method. No. of images No. of subjects Database used Identification or
Verification Processing time (sec) EER Accuracy
(%)
Ortega et al. [10]
Farzin et al. [13]
Akram et al. [14]
Qamber et al. [15]
Fatima et al. [16]
Kose et al. [18]
Ekka et al. [12]
Khakzar et al. [19]
Lajevaedi et al. [20]
Hussain et al. [21]
Condurache et al. [22]
Zahedi et al. [23]
90
300
354
354
333
392
84
280
135
184
513
200
83
60
260
260
159
----
59
40
57
----
159
40
Collected
from
hospital
DRIVE and
STARE
DRIVE,
STARE and
VARIA
DRIVE,
STARE and
VARIA
VARIA and
RIDB
STARE
VARIA
DRIVE
DRIVE
164(Own
database), 20
Staal
VARIA and
DRIVERA
DRIVE
Verification
Identification
Identification
Identification
Identification
Identification
Verification
Identification
Verification
Identification
Both
Identification
and Verification
Identification
0.155
----
----
----
----
6464
----
----
----
----
6
----
----
0.006
----
----
0.0557
----
9.79
----
0.005 for
automatic
and
0.001
for
manual
----
----
----
100
99
98.30
98.87
99.57,97
99.5
90.21
99.29
----
100
99.29
100
B. RETINAL AUTHENTICATION BASED ON
NONVASCULAR FEATURES
Few researchers have carried out retinal authentication
based on non-vascular features of the retina. The block
diagram of this type of retinal authentication system is
shown in Fig.7. After preprocessing the features of the
retinal image are extracted from retina without undergoing
blood vessel segmentation, hence we name them non-
vascular features as these do not depend on blood vessels of
retina only. Therefore, it decreases the execution time of the
system while preserving its good performance.
Waheed et al. [24] proposed a non-vascular based retina
recognition system. It computes similarity measure using
Figure 7. Block diagram of retinal authentication system based on non-
vascular features.
features based upon structural information such as
luminance, contrast and structural features of an image. To
extract the features they have taken two images, which are
to be compared at the same time. Luminance l(x,y) is a
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 716
measure of mean intensity, for two candidate images x and x
and is given as,
Contrast c(x,y) function is a measure of standard deviation
of images x and y and is given by,
Then they have combined luminance and contrast
functions to obtain structure measurement. For matching an
empirically optimized function is used to generate a
similarity score between two candidate images. They have
used only 34 subjects for their experiment and obtained an
identification rate of 92.5%. Ong et al. [25] presented a
retina biometric system consisting of two stages: feature
points matching and the measurement of edge dissimilarity.
To obtain the feature points they have developed a graph-
based method where they have used scale-invariant feature
transform (SIFT). The graphs comprise of vertex set and
edge set. For matching, the feature points they have
employed a sub-graph matching algorithm and pruned the
putative match feature points so that wrong feature points
get removed. To achieve this, a robust transformation
estimator is used which is based on the Least-Median-
Squares (LMedS) estimator followed by a Weighted Least
Squares (WLS) estimator. If a sufficient number of matched
feature points have been detected, then they have performed
an image registration using the estimated affine
transformation parameters to register the template image
with the test image. They have extracted the edges from the
test retinal image and registered template retina image using
Canny’s edge detector and performed the edge dissimilarity
measurement with the help of Robustified Hausdorff
Distance (RHD) measure for decision making. To evaluate
their experiment they have created DRIVERA (DRIVE for
Retinal Authentication) dataset consisted of 280 images
from 20 images of DRIVE database. The performance of
their identification system is evaluated in terms of FAR
(False Acceptance Ratio) and FRR (False Rejection Ratio),
which are 0% and 3.169% respectively. They have
presented that their retina verification technique surpasses
two of the state-of-the-art approaches [22] and [26] when
tested on the same dataset.
Sabaghi et al. [27] have introduced a human
identification system based on Fourier transform (FT) and
angular partitioning of the spectrum of retinal images. For a
robust system, the rotation of retinal image should be
compensated. To achieve this first they have localized the
OD using template matching technique, and then determined
the center of the OD and the center of mass to obtain the
rotation angle of the retinal image as shown in Fig.8. For
extracting features, they have used FT of the retinal image.
The angular partitioning of the Fourier spectrum is
performed to calculate the energy of Fourier spectrum. Then
the sum of the phase angle per partition is computed. They
have named this feature vector as Fourier Spectrum Phase
Feature (FSPF) and are shown in Fig.9. They have also
shown that their system is robust to noise.
Figure 8. The result of regulate retinal image to reference position: (a)
Retinal image after localized center of mass and optical disk; (b) Obtain the
angle of rotate; (c) Compensated image by rotation [27].
Figure 9. Complete flow diagram for feature extraction process in the
proposed system. (a) Retinal image after rotation compensation (b) Fourier
spectrum (c) phase angle (d) partitioning the Fourier spectrum and phase
angle (e) calculate energy of Fourier spectrum and sum of the phase angle
per partition (e) future vector construction [27]
Later in [28], they have introduced another method,
which is a combination of wavelet transform along with
Fourier transform for FSPF extraction. With Fourier
transform they extracted Fourier energy feature and with
wavelet transform, they have extracted wavelet energy
feature. For matching they have used Euclidian distance in
both the papers. They have shown that the accuracy of the
identification system is 95.4% with Fourier transform
features and 97.2% with wavelet transform features but it
increases to 99.1% by combining both the methods.
Dehghani et. al. [29] have proposed a person authentication
system using retinal features. Their method consists of
feature extraction, phase correlation, and feature matching.
They have observed that the corners of retinal images are
rotation invariant and for this, they have used Harris corner
detector. After feature extraction, they have applied the
phase correlation method to determine the rotation of the
head and eye movement in front of a retinal scanner.
Finally, for feature matching they have used a similarity
function to calculate the similarity between different retinal
images. Their method perform in both identification and
verification mode and the system is more efficient as the
rate of accuracy in both the mode is 100% and the
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 717
processing time is also very less. The retinal recognition
using different non-vascular features is discussed above. All
the authors have carried out their experiments using a
different database and have obtained percentage of EER and
accuracy as shown in Table 2.
Table II. Comparison Table of different Recognition methods using non-vascular feature
Method. No. of images No. of subjects Database used Identification or
Verification Processing time (sec) EER Accuracy
(%)
Waheed et al. [24]
Ong et al. [25]
Sabaghi et al. [28]
Dehghani et al. [29]
100
42
280
400
480
20
14
20
40
80
RIDB
AFIO
DRIVERA
DRIVE
DRIVE and
STARE
Identification
Identification
Verification
Identification
Both
Identification
and Verification
----
----
----
----
5.3s for Identification
0.07 for Verification
0.26
0.2857
0.170
----
----
92.50
85.75
----
99.10
100
III. STRENGTH AND WEAKNESS OF RETINAL
AUTHENTICATION
After going through the literature we observed that retina
based biometric authentication system possesses its own set
of strength and weakness, just like all other biometric
technologies. They are mentioned below:
The strengths can be described as follows:
1) The blood vessel pattern of the retina hardly ever changes
during a person’s life.
2) As the retina is located at back end of the eye, it is not
exposed to the threats posed by the external environment, as
other organs such as fingerprint, hand geometry, face etc.
3) The stable and distinctive features of retina can be
extracted from the vascular structure.
4) In comparison to other biometric characteristics, the
average feature vector size of retina biometric is very small,
which could result in faster verification and identification
processing times. Whereas, the larger sized feature vectors
could decelerate the processing times.
The weaknesses can be discussed as follows:
1) If a person is affected with some eye diseases such as
hard glaucoma, cataracts, and so on then the identification
process will be very complicated.
2) The image acquisition involves the cooperation of the
subject, entails contact with the eyepiece, and also the
retinal scanning technology cannot accommodate people
wearing contact lenses.
3) The blood vessel structure of retina can divulge some
medical conditions of that person which are another factor
obstructing the public acceptance of retina based biometric
authentication system.
4) The devices used to capture the retinal images are very
expensive to procure and implement.
IV. CONCLUSION AND DISCUSSION
In this paper, the different methods used for retinal
authentication are reviewed. They are mainly categorized
based on vascular and non-vascular feature extraction.
Among all other biometric traits, retina is the most stable
characteristic for person authentication because of the
uniqueness in blood vessel pattern and it consistent during
one’s life. In this paper, a comparative study is done on the
recognition rates, equal error rates, processing time and the
numbers of images taken in different methods. From the
study it is found that the blood vessel patterns are the
foundation for retinal recognition, therefore it is important to
segment it accurately. In the survey, it is observed that most
of the authors used bifurcation point and end point as a
feature vector. In future, technologies must be developed to
extract number of features, which will increase the
efficiency of the system. It is also seen that very few authors
have taken retinal images suffered from diseases like AMD,
DR etc. for their experiment, so in future system must be
designed so that it can identify not only a normal image but
also a degraded diseased image. The main difficulty in
retina based authentication system is the movement of head
or eye in front of the retinal scanner. Hence, to make the
system more robust it should tolerate scale, rotation, and
translation for accurate identification. It has been observed
that most of the methods can operate in either identification
or verification mode, but it is very important to develop a
method which will operate in both modes for security
application. It is also required to optimize the processing
time to speed up the process and to make this a real-time
biometric identification system for high security.
S. R. Nirmala et al, International Journal of Advanced Research in Computer Science, 9 (1), Jan-Feb 2018, 711-718
© 2015-19, IJARCS All Rights Reserved 718
V. REFERENCES
[1] S. Prabhakar, S. Pankanti, and A. K. Jain, “Biometric
recognition: Security and privacy concerns,” IEEE security
& privacy, vol. 99, no. 2, pp. 33–42, 2003.
[2] A. K. Jain, A. Ross, and S. Prabhakar, “An introduction to
biometric recognition,” IEEE Transactions on circuits and
systems for video technology, vol. 14, no. 1, pp. 4–20, 2004.
[3] S. K. Sahoo, T. Choubisa, and S. M. Prasanna, “Multimodal
biometric person authentication: A review,” IETE Technical
Review, vol. 29, no. 1, pp. 54–75, 2012.
[4] K. A. Nixon, V. Aimale, and R. K. Rowe, “Spoof detection
schemes,” in Handbook of biometrics, pp. 403–423,
Springer, 2008.
[5] L. R. Kumar and A. S. Babu, “Genuine and forged offline
signature verification using back propagation neural
networks,” IJCSIT) International Journal of Computer
Science and Information Technologies, 2011.
[6] S. Shankar, V. Udupi, and R. D. Gavas, “Biometric
verification, security concerns and related issues-a
comprehensive study,” 2016.
[7] I. Bhardwaj, N. D. Londhe, and S. K. Kopparapu, “A novel
behavioural biometric technique for robust user
authentication,” IETE Technical Review, pp. 1–13, 2016.
[8] R. B. Hill, “Retina identification,” Biometrics, pp. 123–141,
2002.
[9] R. Das, “Retinal recognition: Biometric technology in
practice,” Keesing Journal of Documents & Identity, vol. 22,
pp. 11–14, 2007.
[10] M. Ortega, M. G. Penedo, J. Rouco, N. Barreira, and M. J.
Carreira, “Personal verification based on extraction and
characterisation of retinal feadture points,” Journal of Visual
Languages & Computing, vol. 20, no. 2, pp. 80–90, 2009.
[11] M. Ortega, C. Marino, M. Penedo, M. Blanco, and F.
Gonzalez, “Biometric authentication using digital retinal
images,” in Proceedings of the 5th WSEAS international
conference on Applied computer science, pp. 422–427, 2006.
[12] B. K. Ekka, N. Puhan, and R. Panda, “Retinal verification
using point set matching,” in Signal Processing and
Integrated Networks (SPIN), 2015 2nd International
Conference on, pp. 159–163, IEEE, 2015.
[13] H. Farzin, H. Abrishami-Moghaddam, and M.-S. Moin, “A
novel retinal identification system,” EURASIP Journal on
Advances in Signal Processing, vol. 2008, no. 1, pp. 1–10,
2008.
[14] M. U. Akram, A. Tariq, and S. A. Khan, “Retinal
recognition: Personal identification using blood vessels,” in
Internet Technology and Secured Transactions (ICITST),
2011 International Conference for, pp. 180–184, IEEE, 2011.
[15] S. Qamber, Z. Waheed, and M. U. Akram, “Personal
identification system based on vascular pattern of human
retina,” in Biomedical Engineering Conference (CIBEC),
2012 Cairo International, pp. 64–67, IEEE, 2012.
[16] J. Fatima, A. M. Syed, and M. U. Akram, “A secure personal
identification system based on human retina,” in Industrial
Electronics and Applications (ISIEA), 2013 IEEE
Symposium on, pp. 90–95, IEEE, 2013.
[17] A. Arneodo, N. Decoster, and S. Roux, “A wavelet-based
method for multifractal image analysis. i. methodology and
test applications on isotropic and anisotropic random rough
surfaces,” The European Physical Journal B-Condensed
Matter and Complex Systems, vol. 15, No. 3, pp. 567–600,
2000.
[18] C. K¨ose, C. ˙Iki, et al., “A personal identification system
using retinal vasculature in retinal fundus images,” Expert
Systems with Applications, vol. 38, no. 11, pp. 13670–
13681, 2011.
[19] M. Khakzar and H. Pourghassem, “A rotation invariant
retina identification algorithm using tessellation-based
spectral feature,” in Biomedical Engineering (ICBME), 2014
21th Iranian Conference on, pp. 309–314, IEEE, 2014.
[20] S. M. Lajevardi, A. Arakala, S. A. Davis, and K. J. Horadam,
“Retina verification system based on biometric graph
matching,” IEEE transactions on image processing, vol. 22,
no. 9, pp. 3625–3635, 2013.
[21] A. Hussain, A. Bhuiyan, A. Mian, and K. Ramamohanarao,
“Biometric security application for person authentication
using retinal vessel feature,” in Digital Image Computing:
Techniques and Applications (DICTA), 2013 International
Conference on, pp. 1–8, IEEE, 2013.
[22] A. P. Condurache, J. Kotzerke, and A. Mertins, “Robust
retina-based person authentication using the sparse
classifier,” in Signal Processing Conference (EUSIPCO),
2012 Proceedings of the 20th European, pp. 1514–1518,
IEEE, 2012.
[23] A. Zahedi, H. Sadjedi, and A. Behrad, “A new retinal image
processing method for human identification using radon
transform,” in 2010 6th Iranian Conference on Machine
Vision and Image Processing, pp. 1–4, IEEE, 2010.
[24] Z. Waheed, A. Waheed, and M. U. Akram, “A robust non-
vascular retina recognition system using structural features
of retinal image,” in 2016 13th International Bhurban
Conference on Applied Sciences and Technology (IBCAST),
pp. 101–105, IEEE, 2016.
[25] E. P. Ong, Y. Xu, D. W. K. Wong, and J. Liu, “Retina
verification using a combined points and edges approach,” in
Image Processing (ICIP), 2015 IEEE International
Conference on, pp. 2720–2724, IEEE, 2015.
[26] H. Oinonen, H. Forsvik, P. Ruusuvuori, O. Yli-Harja, V.
Voipio, and H. Huttunen, “Identity verification based on
vessel matching from fundus images,” in 2010 IEEE
International Conference on Image Processing, pp. 4089–
4092, IEEE, 2010.
[27] M. Sabaghi, S. R. Hadianamrei, A. Zahedi, and M. N. Lahiji,
“A new partitioning method in frequency analysis of the
retinal images for human identification,” Journal of Signal
and Information Processing, vol. 2, no. 04, p. 274, 2011.
[28] M. Sabaghi, S. R. Hadianamrei, M. Fattahi, M. R. Kouchaki,
and A. Zahedi, “Retinal identification system based on the
combination of Fourier and wavelet transform,” 2012.
[29] A. Dehghani, Z. Ghassabi, H. A. Moghddam, and M. S.
Moin, “Human recognition based on retinal images and
using new similarity function,” EURASIP Journal on Image
and Video Processing, vol. 2013, no. 1, pp. 1–10, 2013.