(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 13, No. 4, 2022

A Comprehensive Overview on Biometric

Authentication Systems using Artificial Intelligence
Techniques

Shoroog Albalawi1 , Lama Alshahrani2 , Nouf Albalawi3 , Reem Kilabi4 , A’aeshah Alhakamy5
Faculty of Computers and Information Technology,
Master of Artificial Intelligence at University of Tabuk, Saudi Arabia1,2,3,4,5
Industrial Innovation & Robotics Center (IIRC) and Faculty of Computers and Information Technology,
Department of Computer Science at University of Tabuk, Saudi Arabia5

Abstract—Biometric authentication is becoming more preva-

lent as it allows consumers to authenticate themselves without
entering a physical address or a personal identification number.
Thus, a simple finger gesture or a glance at a camera can
still prove one’s identity. In this review, we explain in detail
how the concept of authentication and the various types of
biometric techniques is used for user identification. Then, we
discuss the various ways these techniques can be combined to
create a truly multimodal authentication system. For a more
organized approach, our overview is classified into two main
categories based on human biometric traits. First, the physiolog-
ical traits include fingerprint, facial, iris/retina, hand, and finger-
vein. Second, the behavioral traits includes voice, signature, and
keystroke recognition systems. Finally, we offer a comprehensive
comparison of selected methods and techniques and focus on
three criteria: algorithms, merits, and drawbacks. Based on this
comparison, we provide insight into our future research in iris
recognition, by which we combine several artificial intelligence
algorithms to develop our system.
Keywords—Biometric authentication; physiological traits; be-
havioral traits; facial recognition; iris recognition; voice recogni-
tion; signature

I. I NTRODUCTION
Authentication is a security function that involves providing
and checking the proof of the person’s identity, the message Fig. 1. Classification of Biometric Traits.
sender, the software, the logical server, or the device. Different
identification methods have been proposed for exchanging
safety-related information [1], [2], [3], [4], faced with the
logical evolution of international regulations; new technolog- The motivations and potential benefits that drive our re-
ical solutions are gradually being implemented. Among these search are the convenient use of this approach in different
technologies, biometrics is the most relevant technology to systems around the world. Our future project involves using
reliably and rapidly identify and authenticate a person, based iris recognition methods in airports for fast and accurate
on their unique biological characteristics. authentication. The objective is to identify the human iris
and link the data together by using artificial intelligence (AI)
Biometric authentication uses a person’s biological char- techniques to extract a passenger’s information, such as flight
acteristics to verify their identity and ensure secure access to number, dedicated gate, seat number, departure and arrival
an electronic system. The biometric technologies are based times, travel bag weight, how many kilos are reserved for travel
on how each person can be identified distinctly through one bags, vaccination status, and chronic disease diagnosis (e.g.,
or more biological characteristics, such as fingerprint, hand diabetes, high blood pressure).
morphology, retina and iris, voice, DNA, or signatures. In
general, biometrics can be classified into two categories: Because different types of information can be extracted
physiological traits, and behavioral traits, as shown in Fig. 1. from the human iris, from our point of view, we believe the
Biometric authentication is the application of these biometric following methods better fit for our future work. (1) The edge
technologies to identify a person as part of a user validation detection algorithm is used to localize the iris and select the
process to access a system. most important features. The system can then be segmented
www.ijacsa.thesai.org 782 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

and localized by using the fuzzy algorithms and the edge enrollment and the file obtained during authentication. Several
detection. (2) statistical features extraction methods are then methods are used to recognize fingerprints, such as locating
used to extract the characteristics of human iris. (3) The feature minutiae and processing textures. The process of extracting
space is then divided into classes, and the extracted features are minutiae involves using the template comparing to the image
trained parallel to partition the features into these classes using digitization and minutiae extraction, as shown in Fig. 2.
Principal component analysis (PCA). (4) The Support vector
machines (SVMs) algorithm is then used to classify the new
images in the database according to their classes to pinpoint
the traveler’s identity.
In this work, we present a review of biometric authentica-
tion methods and techniques that ease the way people interact
with systems and how their identity influences governmen-
tal and private systems. Section III introduces physiological
biometrics, and the behavioural biometrics are presented in
Section IV. In these sections, we briefly discuss and list the
most common biometrics, such as the facial, iris, fingerprint,
Fig. 2. Fingerprint Authentication Process: (1) Image Digitization, (2)
voice, and signature recognition techniques. We also depict, Minutiae Extraction and (3) Comparison to the Templates.
the typical design steps of each system and its applications to
provide a thorough description.
The image digitization step consists of digitizing the
II. B IOMETRICS R ECOGNITION S YSTEMS fingerprint, filtering unnecessary features (e.g., scars), and
Biometric technologies are used to secure a wide range determining information useful to the system. To detect the
of electronic communications or to connect to a computer endpoints and crossing points of ridges, known as minutiae,
or smartphone. Biometric authentication systems compare the a fingerprint skeleton is created using complex algorithms in
human biometric data to be authenticated with the biometric order to make each line of the imprint, with a length from 5 to
database. If the two samples are matched, the authentication 8 pixels and a thickness of 1 pixel. Fig. 3 shows the creation
is then confirmed and access is granted. This process is of a fingerprint skeleton.
sometimes part of a multifactor authentication system. Thus,
the smartphone user can connect using their secret code such
as a personal identification number [PIN] and add an iris scan.
Generally, biometric identifiers are classified into physiological
or behavioral characteristics; see Fig. 1.
Human biopsies are related to the various physiological
characteristics of the human body such as fingerprints, facial
features, iris, retina recognition, and DNA. Behavioral biomet-
rics are related to a behavioral pattern, like the rhythm of a
person typing, or how they use their fingers or look at the
Fig. 3. Fingerprint Skeleton Creation.
camera.

III. P HYSIOLOGICAL B IOMETRICS The minutiae allows the extraction step to determine the
Physiological or static biometrics use physical character- signature of the fingerprint using different algorithms. The
istics, such as fingerprint or facial recognition, etc. to unlock template used to characterize the fingerprint is based on a
cell phones, log into bank accounts, or complete transactions. minimum number ranging from 12 to 14 minutiae, although
However, the main types of static biometrics used to verify a this number can be higher as needed to establish a reliable
person’s identity are fingerprint recognition, facial recognition, comparison. In fact, it is possible to identify one fingerprint
iris recognition, etc. among several million fingerprints with this minimum of
correctly identified and localized minutiae.
A. Fingerprint Recognition Finally, the comparison of two minutiae’s, corresponding to
The fingerprint is one of the oldest forms of biomet- two fingers to be compared, constitutes the identity verification
ric authentication, and mobile platforms use this technology system. To determine whether two minutiae’s extracted from
widely. It was originally popularized by Apple’s Touch ID. A two images correspond to the same fingerprint, it is necessary
fingerprint reader analyzes a person’s fingerprint and compares to adopt a comparison system that is insensitive to any trans-
it to the finger’s stored digital pattern during authentication. lations, rotations, or deformations, which systematically affect
Fingerprint recognition may change if the finger is wet or dirty. the fingerprints. From two extracted minutiae’s, the system
An attacker cannot replicate a person’s fingerprint because of should be able to give a similarity or correspondence index
its vividness; however, it can be used to create a 3D model or of 0% if the fingerprints are totally different, and 100% if the
a fake image. fingerprints are from the same image.
Fingerprint authentication is based on the concordance Fingerprint recognition algorithms are sensitive to the im-
between the registration or signature file obtained during age quality. The pretreatment step is therefore necessary before
www.ijacsa.thesai.org 783 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

performing the following steps. The quality of fingerprint

images depends on several factors, such as contact with the
probe, quality of the probe, and depth of ridges/bifurcations.
Generally, preprocessing consists of smoothing, contrast en-
hancement, spatial/frequency domain filtering. Today, several
techniques are used to solve the problems associated with
fingerprints recognition.
A comprehensive overview of the patterns and techniques
used in fingerprint recognition depends on the minutiae-based
technique. [5]. Peralta, et al. [6] described the various aspects
of fingerprint authentication and identification with respect to
the minutiae-based matching algorithm. Fingerprint authenti-
cation utilizing minutiae extraction technique are discussed
by Sharma, et al. [7] who covered all related systems and
processes. Unimodal and multi-modal biometrics techniques
were summarised by Delac and Grgic [8] including pros and
cons for each model.

B. Facial Recognition
facial recognition is a technique used to identify/verify
human identity based on their facial features. facial recognition
can be performed from photos or video recordings. To develop
a system with robust facial recognition, four steps are taken
under consideration: (1) facial detection, (2) feature extraction,
and (3) feature classification, and (4) feature matching. The Fig. 4. Facial Recognition Structural Steps: (1) Facial Detection, (2) Feature
face detection step involves identifying the human face in the Extraction and (3) Facial Recognition.
image, and the feature extraction step involves extracting the
feature vectors for the identified facial. The feature extraction
step is considered the most crucial step in the facial recognition
process. The results can then be compared and classified A study conducted by Anusha, et al. [10] showed that the
according to a certain criterion to identify the features of the weight of the image when compared with the test images can
image. These steps are illustrated in Fig. 4. be used to identify the facials. With the same objective, Shen,
et al. [11] pa method that uses both the, (1) PCA algorithm
There are three major approaches to automatic face recog- to get the feature space of the training set and then trains to
nition by computer: global feature approach (facial-based identify the projected facial and (2) Fisher linear discriminant
recognition), local feature approach (constituent-based recog- (FLD) algorithm to obtain the fusion feature space and then
nition), and hybrid approaches. to train and recognize the projected facial in the feature space.
Their study proved that the algorithm based on the FLD and
1) Global methods: The global approach is commonly used PCA features can outperform the existing facial recognition
to identify facials using the entire image without taking into methods.
account the a face’s local physiological features, such as the
eyes and mouth. Luabibi, et la. [12] proposed a four-phase approach for
facial recognition. The first step is the localization of the
The global algorithms are based on statistical properties facials, and the second involves extracting the features, and
and are usually quick to implement. However, they are sen- the third one is the classification, and the fourth phase is the
sitive to various factors such as lighting conditions and facial back propagation algorithm to identify the facials.
expression. Some of these algorithms are principal component
analysis (PCA), linear discriminant analysis (LDA), support The LDA method is a numerical technique that can be
vector machine (SVM), and neural network (NN). used to classify and improve the data representation. The
resulting combinations are usually used as linear classifiers.
In this context, Sugandi, et al. [9] have proposed a facial This algorithm cuts each facial into a linear combination before
recognition method based on PCA and backpropagation NN. classifying it. The resulting sets of new dimensions are called
Each facial image in the training is represented exactly by a fisherfaces after the LDA method. Due to its complexity, the
linear combination of eigenfaces. This method is performed LDA algorithm is not as effective as other methods.
in three stages. In the first step, facial detection is performed
using Haar-like features. In the second step, the authors use Bhattacharyya, et al. [13] proposed a method that involve
the AdaBoost learning algorithm to select the most important grouping images of the same class and different classes. The
features. Finally, the backpropagation NN is used for the resulting set of facials is then classified according to the closest
recognition process. In their work, the authors demonstrated training images. The study’s results revealed that the proposed
that using 5 data facial images with each data are taken algorithm is significantly better than the existing methods. Lu,
100 times than usual, the experimental result showed the et al. [14] combined both the direct LDA and Fractional LDA
satisfactory result with 87.5% recognition rate. techniques, and algorithms can be used for small sample sizes.
www.ijacsa.thesai.org 784 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

It can also classify the facials according to the closest training lack of precision in the extraction phase of the points, which
images. Results of the experiments on two databases supported constitute their major drawback. Specifically, the methods start
the effectiveness of the proposed algorithm. by extracting the local facial features such as the nose, eyes,
and mouth; and the method then use their geometry and/or
SVM’s are commonly used in machine learning to solve appearance as input to the classifier. Hence, we can distinguish
problems related to discrimination and regression. Although two practices:
they are not as efficient as some of the most popular algo-
rithms, their potential is still very promising. In their study, Jin, • The first practice is based on the extraction of entire
et al. [15] proposed an algorithm based on a modified SVM regions of the facial; it is often implemented with a
learning scheme that uses the SVM and particle swarm op- global facial recognition approach.
timization techniques. The results of the experiments showed
that the proposed method has better accuracy than the existing • The second practice extracts particular points from
methods. different characteristic regions of the facial, such as
the corners of the eyes, mouth, and nose.
In addition, Jose, et al. [16] presented a review of the
techniques used for 2D facial recognition using the SVM tech- Among these approaches, we can list the hidden Markov
nique. The authors analyzed the recognition results according model (HMM) and the elastic bunch graph matching algorithm
to the techniques used to extract information such as facial (EBGM). An HMM is a statistical model in which the modeled
features, pattern classifiers, and databases. A new method that system is assumed a Markov process with unknown parame-
uses the SVM algorithm to detect the facials in grayscale ters. Unlike a classic Markov chain, where the transitions have
images was presented by Ignas et al. [17]. They then identifed taken are unknown to the user but the states of execution are
the faces using the sizes and positions of the eyes and lips. known, in an HMM, the states of execution are unknown to the
user. HMMs are currently among the most widespread models
Also, Javed [18] used the PCA and SVM to build for build- of form recognition. They then were established in speech
ing a facial recognition model. The authors [19] presented an recognition and written recognition.. However, when using
overview of the effective use of machine learning, particularly HMMs, the structure of the facial is considered as distinct
regarding using SVMs in facial recognition. Therefore, the regions, described by characteristic vectors.
authors give an extensive survey of facial recognition and its
applications. In this context, Alhadi, et al. [22] studied three different
methods to extract feature vectors from HMMs: discrete cosine
Artificial Neural Networks (ANNs) are systems comprising transform, discrete wavelet transform, and PCA. The results
of several interconnected processing units. They can perform of the experiments revealed that combining these methods
various computational tasks based on the input data. In com- improved the models’ recognition performance. In addition, a
puter vision, an NN is composed of several processing units state-of-art HMM model applied to facial recognition problems
known as neurons. These components can learn and adapt to in the review [23]. In this work, the authors have presented
different tasks in order to classify the data. the evolution of HMM use from the early 1990s to the present
Deep learning (DL) is an AI method that learns by itself. It day; this makes it easier for new researchers to understand and
is inspired by the human brain. The goal of DL is to learn and adopt the HMMs more easily for facial recognition.
recognize different words and facials in an image. For example, In 1997, Wiskott, et al. [24] proposed a graph matching
DL can detect the letters in text before recognizing a face. In algorithm commonly used in computer vision. This algorithm
their work, Hassan, et al. [20] detailed the various approaches makes it possible to recognize objects in an image by using
used in facial recognition and provided a thorough analysis of a graphic representation extracted from other images. This
their results. They also introduced hybrid algorithms that can approach extracts a set of characteristics using a data structure
be used for extracting and classifying facial features. called a packet graph [25]. Furthermore, Jaiswal [26] presented
Convolutional Neural Networks (CNNs) are commonly a method for recognizing human facials that uses the concept
used in facial recognition. They are typically complex and of a graph matching algorithm known as EBGM. The proposed
require a high amount of processing power and storage space method achieves high recognition rates for both the facial and
to perform their intended applications. In this area, Liu, et al. the image graphs.
[21] tried to improve the performance of CNNs by introducing 3) Hybrid methods: Hybrid methods pair the advantages
a block called the squeeze-and-excitation algorithm. The pro- of global and local methods by combining the detection of
posed algorithm has fewer parameters and can be more suitable geometric (or structural) characteristics with the extraction of
for various applications. local characteristics. They make it possible to increase the
2) Local methods: These are also called the geometrical, stability of the recognition performance during changes in
local characteristics, or analytical methods. This approach pose, lighting, and facial expressions. Local feature analysis
involves in applying transformations in specific places of the and Gabor wavelet extracted features (such as EBGM) are
image, most often around the characteristic points (i.e., corners typical hybrid algorithms [27], [28], [29].
of the eyes, mouth, and nose). Attention is given to small local
details, and the approach avoids the noise generated by aspects C. Iris / Retina Recognition
such as hair, eyeglasses, hats, and beards.
The ability to identify the facial using the iris is one of the
However, the difficulty of these methods arises when it most accurate and secure methods for biometric identification.
taking into consideration several views of the facial and the Unlike the hands and facial, the iris is a protected internal
www.ijacsa.thesai.org 785 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

organ and is therefore less, and therefore less likely to be 1) Iris Recognition: A new approach for the recognition of
damaged. The user must fix a digital camera that scans the the human iris was developed by Fernando, et al. [30]. It uses
iris of a person from a distance of 30 to 60 cm and directly the SIFT feature transformation to extract the characteristic
acquires iris drawing. Then, this drawing is compared to a features and then matches two images. Experiments with the
computerized personal identification file to identify the person. BioSec database show that the SIFT approach achieves better
performance than the existing matching methods.
An iris recognition system is a type of biometric system A novel approach for deep learning that aims to improve
that uses images of human irises to identify people. It features the accuracy of the recognition of the iris using a more
two main processes: localization and segmentation. Images of simplified framework was presented by wang and Kumar [31].
the iris are determine and analyzed to extract its biometric sig- It utilizes residual network learning and dilated convolutional
nature. However, the image processing operations are divided kernels to improve the training process. An unsupervised
into four stages: localization, segmentation, encoding, and network approach can greatly simplify the network and provide
classification. The diagram in Fig. 5 shows the methodology better performance than state-of-the-art algorithms for iris
steps and sequential processing of the proposed iris recognition recognition. It also eliminates the need for upsampling and
system. downsampling layers. The results of our experiments demon-
strate the applicability of our approach to improve the accuracy
of iris recognition.
Furthermore, a new generalizability method for the recog-
nition of the iris is introduced by Adamović, et al. [32],
which performed on the two CASIA databases. The results
of the experiments demonstrated the system performs as ex-
pected. The generalizability method can significantly reduce
the computational costs of the system, which in turn makes
the method suitable for practical applications. The goal is to
achieve a classification accuracy of almost perfect. The method
also eliminates the possibility of generating an image from a
template.
2) Retina Recognition: The retina is a region of the eye that
consists of four layers of cells. The arrangement of these cells
is unique and provides a high level of recognition. This tech-
nology is well-suited for high-security applications because it
can achieve a recognition rate of around 90% compared to
other methods. The idea for retinal identification and unique
vascular pattern was first introduced by two ophthalmologists,
Dr Carleton Simon and Dr. Isodore Goldstein in 1935 [33].
A new method for retina recognition based on a fractal
dimension was also presented by Sukumaran, et al. [34]. The
authors of the study compared the accuracy of this technology
with the commercially available ones. The experimental results
of the method revealed that it produces high accuracy and low
computational cost.
With the same objective, Tuama and George [35] proposed
a personal identification system using the vascular diagram of
the human retina. This system is composed of four stages.
First, the prepocessing technique is used to extract the retinal
image from the background and to remove noisy areas from
the retinal image. Then, wavelet transforms 2D and adaptive
thresholding were used to extract the blood vessels. Next, the
system performs feature extraction and filtration. Finally, the
matching step is used for the retina recognition. Experimental
results on three publicly available databases (DRIVE, STARE,
and VARIA) have demonstrated that the proposed method is
better than several existing techniques.

Fig. 5. The Architecture of the Iris Recognition System which Include; (1)
IV. B EHAVIORAL B IOMETRICS
Image Acquisition, (2) Iris Localization, (3) Segmentation and (4) Feature Behavioral biometrics analyze a person’s unique habits and
Extraction.
movements to create a behavioral pattern. Like static biomet-
rics, behavioral biometrics adds another layer of security to
www.ijacsa.thesai.org 786 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

verify a person’s identity. Accordingly, this technology uses sequences by dynamic time warping. This technique represents
motion sensors and AI to identify unique behaviors, such as the most recent work in manual signature verification
how a phone is held. These technologies are is widely regarded
as the last frontier in security [36], [37]. C. Voice Recognition

A. Keystroke Dynamics Recognition Voice recognition is a process that uses the sounds pro-
duced by a person’s vocal tract and the shape of their nose,
Stroke rhythm analyzes a person’s typing rhythm on digital mouth, and larynx to identify their voice. Analyzing a person’s
devices (e.g., smartphones) to create a form of human digital voice is a strong authentication method, but illness (e.g., the
footprint or signatures. Gaines, et al. [38] first proposed this common cold, bronchitis) and background noise can distort
technology when they created the first automated dynamic the voice and disrupt authentication. The architecture of an
keystroke recognition systems [39].Major contributions have automatic voice recognition system is represented in Fig. 7.
notably evaluated the fuzzy logic [40], NNs, [41], [42], and
different pattern recognition techniques (e.g., Bayes classifier)
[43], [44].

B. Signature Recognition
The recognition of a signature can be accomplished by
analyzing a large number of discriminative variables: (i) global
characteristics such as writing time or the number of touches
on the tablet with the pen, and/or (ii) local characteristics such
as the position of certain curvatures or the instantaneous speed.
A pressure-sensitive pencil-shaped reader and a digital tablet
usually acquire a signature.
Developing a robust facial recognition system involves
three basic steps: (1) preprocessing, (2) feature extraction,
and (3) signature matching and classification. Before the
algorithm can be used for facial recognition, the preprocessing
step involves removing background noise and refining on Fig. 7. Voice Recognition System Architecture that Includes Feature
Extraction, Feature Matching and Decision-making based on the Voice Print
the signature. The signature recognition system architecture Dataset.
is presented in Fig. 6. Feature extraction is the next step
in the recognition process. It involves extracting the various
features of the human signature, such as the Walsh coefficient, However, the applications of speech recognition are diverse
grid, and texture. The signature recognition step matches the and each system has its own architecture and operation mode.
various features of the human signature to template signature In 1987, the Worlds of Wonder Company first marketed
databases. speech recognition via a doll named Julie [49]. For its voice
recognition algorithm, Texas Instruments used a digital signal
processing system that could recognize eight different sen-
tences.
In addition, voice recognition has been studied for many
years to help people with disabilities [50], [49], [51]. People
without mobility can control an electric wheelchair through
voice commands [52].

V. D ISCUSSION AND C OMPARISON

In this section, we provide a summary and discussion
of the review and outline different methods of biometric
authentication. We select and compare some of these methods
Fig. 6. Signature Recognition System Architecture that begins with Feature
Extraction, then Matches these Features with a Reference Template and and observe each method’s algorithms, advantages and disad-
Finally Obtains the Recognition Result. vantages. The study of limitations is key to improving in the
future work, and we must highlight them to build our own
methodology.
The best-known comparison techniques use a HMM [45], It is crucial to acknowledge that each method is not
or a dynamic programming approach [46], [47].The Inter- excluded from the others. These techniques can be combined
national Graphonomic Society (IGS) research community, with one another or used in different parts of the whole process
particularly the Scribens team, proposed the most important to achieve the desired realism.
contribution to handwritten signature verification [48]. In this
area, Jain, et al. [46] proposed to use a new measure of To explore each technique and compare these methods, we
dissimilarity based on the alignment of characteristic vector established criteria to recognize the major differences and how
www.ijacsa.thesai.org 787 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

they affect the final outcome. The major differences in aim,

accuracy, and robustness among methods make concluding the
compression challenging.
Therefore, based on the information provided in each study
we attempted to distill the most useful information to write
this section. We believe this effort will be helpful as a future
reference for our work and that of others regarding usability,
performance and more. Thus, the criteria are as follows:

• Method. The fundamental calculations and algorithms

essential for the system’s methodology and structure.
• Merit. The advantages the system provides, including
the accuracy of the result under the previous criteria.
• Drawbacks. The system’s limitations based on the re-
quired data, assumptions, and outcomes. Future work
usually starts by discussing and discovering these
drawbacks and then attemptting to solve them.

We could cover more criteria, but doing so requires insight

into and evaluation of each system, which is outside the scope
of the current review. We might approximate for the time and
effort for the rendering and processing, but the result would
be untrustworthy without any quantitative data.
Fig. 8. Methodology Steps for Our Future Implementation in Iris
An overview of these criteria to compare the previous Recognition System.
methods from selected papers are presented in Table I. These
papers were chosen based on two factors: (1) significance in
their field, (2) ability to offer insights for our future work. will be refined using the PCA reduction technique and will be
the input to the SVM algorithm for classifiying human iris.
VI. O PEN P ROBLEM AND P ROPOSED M ETHODOLOGY Once the iris’s image is acquired, image-processing tech-
Biometric identification is a process that uses sensors to niques are used to extract the iris, construct biometric signa-
measure a person’s biological characteristics. The data these ture, and finally find its identity. However, the acquired image
devices collect then can be compared to information stored in a must have a minimum resolution and quality to ensure the
database. In addition to fingerprints and eyes, biometrics such characteristics necessary for representation and identification
as facial recognition and hand geometry have been studied and have been obtained. Therefore, efficient segmentation of these
used. Eye biometrics offer the highest level of accuracy and regions is necessary.
individuality. Iris localization is one of the most important steps in an
Iris recognition is a biometric technique that allows recog- iris recognition system because it determines the matching
nition of a person by observing their iris. Generally, the accuracy. This step mainly localizes the boundaries of the iris,
iris recognition system includes a series of steps: (i) image which are the inner and outer boundaries of the iris and upper
acquisition; (ii) iris preprocessing, including localization and and lower eyelids. To do this, edge detection techniques can be
segmentation; (iii) feature extraction; and (iv) matching and used to localize the iris in the original image. After acquiring
classification. The diagram in Fig. 8 shows the methodology the image, the iris should be isolated. As a first step, a series
steps and sequential processing of the proposed iris recognition of image enhancement operations such as filtering, contrast
system. enhancement, and histogram equalization type can be applied.
The goal is to enhance the quality of the image and then apply
For passenger identification at an airport, we propose an iris iris segmentation operations.
recognition system that performs the feature extraction method
and the PCA algorithm to extract and select the most important The processing set necessary to extract the iris from
statistical features of the human iris. In addition, this system its environment defines the iris segmentation: pupil, sclera,
performs a supervised classification process using the SVM eyelids, eyelashes, and specular reflections. This is the most
algorithm to identify a person’s identity. The proposed system difficult step in the recognition system because it affects the
will be applied to a collection of human iris databases, such system’s performance.
as the CASIA iris database. Segmentation approaches are based on detection by fuzzy
algorithms such as the fuzzy c-means algorithm. In this step,
This system consists of two main phases: the preprocessing
the iris is segmented in details. The encoding consists of
phase and the classification phase. In the preprocessing phase,
extracting the iris’s most discriminating and relevant charac-
the system will acquire the required human iris image. The
teristics, which are required for its identification.
iris’s features will be processed to obtain information using
the statistical feature extraction method. Next, these features To do this, the statistical features are extracted from the iris
www.ijacsa.thesai.org 788 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

TABLE I. C OMPARISON OF P REVIOUS M ETHODS FROM THE S ELECTED PAPERS THAT WE C OVERED F OCUSING ON THREE C RITERIA : A LGORITHMS ,
M ERIT AND D RAWBACKS
Research Method Merit Drawbacks
PHYSIOLOGICAL BIOMETRICS
Fingerprint Recognition Methods
Borra, et al. [5] -Minutiae based approach. High accuracy rate. -No noisy/encrypted images,
-Pattern recognition. -Slow performance due to three levels of split
-Wavelet. texture
-Fails to determine real humans sometimes.
Peralta, et -Minutiae-based local matching -Simplicity Expensive computation, slow and depend on the
al. [6] -Correlation-based matching -Distortion tolerance. skin situation.
-Indexing
Sharma, [7] Minutiae based matching. Widely used and familiar. Affected with wet or dry skin.
Delac, et -Unimodal biometric systems Reliability due to use the combination of different -Noisy Scanned data.
al. [8] -Multimodal biometric system. biometric strength. Difficulty in the data gained from humans.
-Biometric sign can expose to forgery.
Facial Recognition
Budiet, et -PCA -No Correlated Features, Independent variables become less interpretable,
al. [9] -Back Propagation Neural Network. -High Performance, -Data standardization is must before PCA,
-Reduce Overfitting, -Information Loss.
-Improved Visualization,
-No user action: not very intrusive, No physical
contact.
Anusha, et PCA -Not intrusive, done from a distance, -More suited for authentication than for identifi-
al. [10] -Inexpensive technique. cation purposes.
-Several characteristic features -User perceptions and civil liberty.
Shen, et -PCA -Reduced Overfitting. -Data standardization before PCA.
al. [11] -FLD . -Improved Visualization. -Information Loss.
Morooj, et -PCA -Removes Correlated Features. -Independent variables become less interpretable,
al. [12] -Backpropagation NN -Improves Algorithm Performance -Information Loss
Bhattacharyya, LDA -Objective evaluation. -Singular within-class scatter matrix due to small
et al. [13] -small set of features for classification purposes. size sample.
-Overcomes the limitation of PCA by applying the -Difficulty differentiating identical twins.
linear discriminant criterion. Sensitive to changes such as beard and glasses.
Jin, et al. [15] -PSO Effective in high dimensional spaces. -Not suitable for large data sets.
-SVM -Premature convergence of PSO leading to stag-
nate in local optimum.
-Environmentally sensitive technology.
Jose, et al. [16] SVM Effective in high dimensional spaces. Not suitable for large data sets.
Ignas, et -SVM Effective in high dimensional spaces -Not suitable for large data sets.
al. [17] -Gaussian kernel for grayscale. -No Noisy data.
-No Overlapped classes.
Mahnoor, [18] -PCA Effective in high dimensional spaces -Not suitable for large data sets.
-SVM -No Noisy data.
-No Overlapped classes.
Jaiswal, et EBGM. Insensitive to lighting variation More complicate procedure.
al. [26] -Rigid, and deformable matching.
Iris/Retina Recognition
Fernando, et SIFT - Stable in lighting and perspective variations. -Complicated Mathematically.
al. [53] -Locality. .High computation cost.
-Resistant to occlusion and crowding .Not effective for low powered devices.
-No prior segmentation.
- Distinctiveness.
-Quantity, generate many features.
Poursaberi, et -Wavelet-based texture -Efficiency: close to real-time performance. -Acquiring an image requires proper alignment
al. [54] Hamming distance of minimum and har- -Extendable and positioning.
monic mean -Robust High accuracy. -Result affected by pupil size change.
-Iris does not change over time
Félix, et Facial key-point detection, -No intimate contact with the reader. -Hard to use.
al. [55] Integro-differential operator (IDO) -More robust than voices. -Difficulty integrating with other systems
Mathematical morphology
Vahid, et Pattern Recognition Approach. -Higher average for matching performance. -The eye position can be problematic.
al. [56] -Convenient for people who wear glasses. -Expensive specialized devices.
-Low chances of a false positive.
Sukumaran, et Fractal dimension using box counting -Most reliable biometric technology -Serious health risk, infrared light beam.
al. [57] Pattern Recognition Approach -Unique data points -Very large false rejection rate
Saba, et -Vascular diagram of the human retina -Very accurate. -Inconvenient for people who wear eyeglasses.
al. [58] -Pattern Recognition Approach -Impossible to forge a retina. -Uncomfortable for some users.
-Low error rate. -Retina biometric devices.
-Low false rejection rate
-Low false acceptance rate.
BEHAVIORAL BIOMETRICS
Keystroke Dynamics Recognition
Gaines, et Examine the probability distribution of -Ergonomic. -Person’s physical condition is important.
al. [38] time to pressed/relaxed each key while -Uniqueness. -Lower Accuracy.
typing on keyboard. -Low Implementation and Deployment Cost. -Lower Permanence .
-Transparency and Noninvasiveness.
-Increase Password Strength and Lifespan.
-Replication Prevention
-Additional Security.
-Continuous Monitoring and Authentication

www.ijacsa.thesai.org 789 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

Signature recognition
Jain, et al.[46] Behavioral biometrics based on person’s -Ergonomic. -Depend on the emotional state.
handwriting -Highly resistant to impostors. -Difficult to use.
-Enrollment is intuitive and fast. -Large template.
-Fast response -High Cost
-Low storage needed. -Vary over time.
-Native language don’t matter.
-Little time of verification.
Voice recognition
Simpson, et Arduino, HM2007 Voice recognition -More natural and easier- -Not reliable.
al. [52] module and Motors. -Faster than typing Sensitivity to environment variations.
-Can be done over the phone-
No specialized equipment
-Not intrusive,
Low computing power

image, and the linear dimensionality reduction technique PCA [2] A. Sumalatha and A. B. Rao, “Novel method of system identification,”
is used to refine the features. Therefore, the iris encoding pro- in 2016 International Conference on Electrical, Electronics, and Opti-
cess constitutes representing the iris’s signature. The encoding mization Techniques (ICEEOT). IEEE, 2016, pp. 2323–2328.
or feature extraction step is unique for each iris, regardless [3] M. Ortega, M. G. Penedo, J. Rouco, N. Barreira, and M. J. Carreira,
“Retinal verification using a feature points-based biometric pattern,”
of dimensional variations or rotations created during the iris EURASIP Journal on Advances in Signal Processing, vol. 2009, pp.
acquisition. This information then will be used to classify the 1–13, 2009.
iris. [4] C. D. Byron, A. M. Kiefer, J. Thomas, S. Patel, A. Jenkins, A. L.
Fratino, and T. Anderson, “The authentication and repatriation of a
VII. C ONCLUSION AND F UTURE W ORK ceremonial tsantsa to its country of origin (ecuador),” Heritage Science,
vol. 9, no. 1, pp. 1–13, 2021.
Biometrics involves studying physical characteristics to [5] S. R. Borra, G. J. Reddy, and E. S. Reddy, “A broad survey on
identify human. With information technology and digitization, fingerprint recognition systems,” in 2016 International Conference on
these techniques have undergone major development. Gen- Wireless Communications, Signal Processing and Networking (WiSP-
erally, biometrics can be applied to individuals who have NET). IEEE, 2016, pp. 1428–1434.
voluntarily shared information with the biometric system, [6] D. Peralta, M. Galar, I. Triguero, D. Paternain, S. Garcı́a, E. Bar-
which must control them based on characteristics such as renechea, J. M. Benı́tez, H. Bustince, and F. Herrera, “A survey on
fingerprint minutiae-based local matching for verification and identifi-
their fingerprint or voice. This step represents the enrollment cation: Taxonomy and experimental evaluation,” Information Sciences,
procedure. This biometric information is then transformed into vol. 315, pp. 67–87, 2015.
a digital file called a signature or template, using a specific [7] M. Sharma, “Fingerprint biometric system: a survey,” International
algorithm that retains and encrypts the characteristic elements Journal of Computer Science & Engineering Technology (IJCSET),
of the digitized image. This signature is then archived and vol. 5, no. 7, pp. 743–747, 2014.
compared with the person’s characteristics during the control. [8] K. Delac and M. Grgic, “A survey of biometric recognition methods,” in
Proceedings. Elmar-2004. 46th International Symposium on Electronics
In this paper, we presented a detailed review of biometric in Marine. IEEE, 2004, pp. 184–193.
systems and compared them to find the best methods and [9] B. Sugandi, I. Dewita, and R. P. Hudjajanto, “Face recognition based
algorithms for our future work. We summarized the notion on pca and neural network,” in 2019 2nd International Conference on
of authentication and authentication technologies and focused Applied Engineering (ICAE). IEEE, 2019, pp. 1–5.
on biometric authentication methods. In addition, we briefly [10] P. Anusha, K. L. Prasad, G. R. Kumar, E. L. Lydia, and V. S. Parvathy,
“Facial detection implementation using principal component analysis
discussed the most common biometrics (i.e., face, iris, voice, (pca),” Journal of Critical Reviews, vol. 7, no. 10, pp. 1863–1872, 2020.
signature, and fingerprint recognition techniques) and the dif- [11] S. Shen, C. Zhang, R. Xiao, W. He, and N. Zhang, “Research on face
ferent ways to combine them to obtain multimodal systems. recognition based on pca and fld,” in 2019 3rd International Conference
Finally, we presented the typical design steps of each system on Data Science and Business Analytics (ICDSBA). IEEE, 2019, pp.
and biometric applications. Consequently, the process automa- 479–481.
tion, particularly the measure of characteristics such as the iris, [12] M. K. Luaibi and F. G. Mohammed, “Facial recognition based on dwt–
the shape of the face or hand, the voice, the speed of typing hog–pca features with mlp classifier,” Journal of Southwest Jiaotong
University, vol. 54, no. 6, 2019.
or pressing keys, the dynamics of signature represent some
challenges that can be develop in the authentication system. [13] S. K. Bhattacharyya and K. Rahul, “Face recognition by linear dis-
criminant analysis,” International Journal of Communication Network
After collecting all this information, we will present a hy- Security, vol. 2, no. 2, pp. 31–35, 2013.
brid approach that combines SVM-based classification and AI [14] J. Lu, K. N. Plataniotis, and A. N. Venetsanopoulos, “Face recognition
techniques to perform advanced iris recognition. The first step using lda-based algorithms,” IEEE Transactions on Neural networks,
vol. 14, no. 1, pp. 195–200, 2003.
involves developing an edge detection algorithm that selects
[15] J. Wei, Z. Jian-Qi, and Z. Xiang, “Face recognition method based
the most important features. The fuzzy and edge detection on support vector machine and particle swarm optimization,” Expert
algorithms are then used to segment and localize the system. Systems with Applications, vol. 38, no. 4, pp. 4390–4393, 2011.
The extracted features are then divided into classes using the [16] J. A. C. Moreano, N. L. S. Palomino, and A. C. L. Casa, “Facial
PCA algorithm. The proposed methodology will be used to recognition techniques using svm: A comparative analysis,” Enfoque
identify passengers at an airport in our future work. UTE, vol. 10, no. 3, pp. 98–111, 2019.
[17] I. Kukenys and B. McCane, “Support vector machines for human
R EFERENCES face detection,” in Proceedings of the New Zealand computer science
research student conference. Citeseer, 2008, pp. 226–229.
[1] C. Otti, “Comparison of biometric identification methods,” in 2016
IEEE 11th International Symposium on Applied Computational Intelli-
gence and Informatics (SACI). IEEE, 2016, pp. 339–344.

www.ijacsa.thesai.org 790 | P a g e
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 13, No. 4, 2022

[18] M. Javed, Building a Facial Recognition Model using PCA & SVM [38] R. S. Gaines, W. Lisowski, S. J. Press, and N. Shapiro, Authentication by
Algorithms. Medium, 2020. keystroke timing: Some preliminary results. Rand Corp Santa Monica
[19] Y. M. Riyazuddin, S. MahaboobBasha, J. K. Reddy, and S. Naseera- CA, 1980.
Banu, “Effective usage of support vector machine in face detection,” [39] S. Bleha, C. Slivinsky, and B. Hussien, “Computer-access security
International Journal of Engineering and Advanced Technology, vol. 9, systems using keystroke dynamics,” IEEE Transactions on pattern
no. 3, pp. 1336–1340, 2020. analysis and machine intelligence, vol. 12, no. 12, pp. 1217–1222, 1990.
[20] R. J. Hassan, A. M. Abdulazeez et al., “Deep learning convolutional [40] W. G. De Ru and J. H. Eloff, “Enhanced password authentication
neural network for face recognition: A review,” International Journal through fuzzy logic,” IEEE Expert, vol. 12, no. 6, pp. 38–45, 1997.
of Science and Business, vol. 5, no. 2, pp. 114–127, 2021. [41] T. Ord and S. M. Furnell, “User authentication for keypad-based devices
[21] W. Liu, L. Zhou, and J. Chen, “Face recognition based on lightweight using keystroke analysis,” in Proceedings of the second international
convolutional neural networks,” Information, vol. 12, no. 5, p. 191, network conference (INC-2000), 2000, pp. 263–272.
2021. [42] M. S. Obaidat and B. Sadoun, “Verification of computer users using
[22] M. W. F. F. H. Alhadi, “Support vector machines for human face keystroke dynamics,” IEEE Transactions on Systems, Man, and Cyber-
detection,” in IASTED International Conference on Computational netics, Part B (Cybernetics), vol. 27, no. 2, pp. 261–269, 1997.
Intelligence. Citeseer, 2005, pp. 4–6. [43] F. Monrose and A. D. Rubin, “Keystroke dynamics as a biometric for
[23] P. Corcoran and C. Iancu, Hidden Markov Models in automatic face authentication,” Future Generation computer systems, vol. 16, no. 4,
recognition-A review. IntechOpen, 2011. pp. 351–359, 2000.
[24] L. Wiskott, N. Krüger, N. Kuiger, and C. Von Der Malsburg, “Face [44] J. A. Robinson, V. Liang, J. M. Chambers, and C. L. MacKenzie,
recognition by elastic bunch graph matching,” IEEE Transactions on “Computer user verification using login string keystroke dynamics,”
pattern analysis and machine intelligence, vol. 19, no. 7, pp. 775–779, IEEE transactions on systems, man, and cybernetics-part a: systems
1997. and humans, vol. 28, no. 2, pp. 236–241, 1998.
[25] ——, “Face recognition by elastic bunch graph matching,” Intelligent [45] R. S. Kashi, W. Turin, and W. L. Nelson, “On-line handwritten signature
Biometric Techniques in Fingerprint and Face Recognition, vol. Chapter verification using stroke direction coding,” Optical Engineering, vol. 35,
11, pp. 355–396, 1999. no. 9, pp. 2526–2533, 1996.
[26] S. Jaiswal, “Comparison between face recognition algorithm-eigenfaces, [46] A. K. Jain, F. D. Griess, and S. D. Connell, “On-line signature
fisherfaces and elastic bunch graph matching,” Journal of global re- verification,” Pattern recognition, vol. 35, no. 12, pp. 2963–2972, 2002.
search in computer science, vol. 2, no. 7, pp. 187–193, 2011. [47] B. Wirtz, “Stroke-based time warping for signature verification,” in
[27] H. Cho, R. Roberts, B. Jung, O. Choi, and S. Moon, “An efficient hybrid Proceedings of 3rd International Conference on Document Analysis and
face recognition algorithm using pca and gabor wavelets,” International Recognition, vol. 1. IEEE, 1995, pp. 179–182.
Journal of Advanced Robotic Systems, vol. 11, no. 4, p. 59, 2014. [48] I. G. S. (IGS), “chemin de polytechnique montréal (québec),” http://
[28] S. Shan, P. Yang, X. Chen, and W. Gao, “Adaboost gabor fisher www.scribens.polymtl.ca/, accessed: 2021-11-30.
classifier for face recognition,” in International Workshop on Analysis [49] K. Tang, R. Kamoua, V. Sutan, O. Farooq, G. Eng, W. C. Chu, and
and Modeling of Faces and Gestures. Springer, 2005, pp. 279–292. G. Hou, “Speech recognition technology for disabilities education,”
[29] H. Cho, R. Roberts, B. Jung, O. Choi, and S. Moon, “An efficient hybrid Journal of Educational Technology Systems, vol. 33, pp. 173 – 184,
face recognition algorithm using pca and gabor wavelets,” International 2004.
Journal of Advanced Robotic Systems, vol. 11, no. 4, p. 59, 2014. [50] J. Noyes and C. Frankish, “Speech recognition technology
[30] V. R.-A. J. O.-G. Fernando Alonso-Fernandez, Pedro Tome-Gonzalez, for individuals with disabilities,” Augmentative and Alternative
“Iris recognition based on sift features,” International Conference on Communication, vol. 8, no. 4, pp. 297–303, 1992. [Online]. Available:
Biometrics, Identity and Security (BIdS), pp. 1–7, 2009. https://doi.org/10.1080/07434619212331276333
[31] K. Wang and A. Kumar, “Toward more accurate iris recognition using [51] J. Schönbächler, “Le traitement de la parole pour les personnes handi-
dilated residual features,” IEEE Transactions on Information Forensics capées,” Travail de séminaire, 2003.
and Security, vol. 14, no. 12, pp. 3233–3245, 2019. [52] R. Simpson and S. Levine, “Voice control of a powered wheelchair,”
[32] S. Adamović, V. Miškovic, N. Maček, M. Milosavljević, M. Šarac, IEEE Transactions on Neural Systems and Rehabilitation Engineering,
M. Saračević, and M. Gnjatović, “An efficient novel approach for vol. 10, no. 2, pp. 122–125, 2002.
iris recognition based on stylometric features and machine learning [53] F. Alonso-Fernandez, P. Tome-Gonzalez, V. Ruiz-Albacete, and
techniques,” Future Generation Computer Systems, vol. 107, pp. 144– J. Ortega-Garcia, “Iris recognition based on sift features,” in 2009 First
157, 2020. IEEE International Conference on Biometrics, Identity and Security
[33] C. Simon, “A new scientific method of identification,” New York state (BIdS). IEEE, 2009, pp. 1–8.
journal of medicine, vol. 35, no. 18, pp. 901–906, 1935. [54] A. Poursaberi and B. N. Araabi, “Iris recognition for partially occluded
[34] M. P. S. Sukumaran, “Retina recognition based on fractal dimension,” images: methodology and sensitivity analysis,” EURASIP Journal on
International Journal of Computer Science and Network Security (IJC- Advances in Signal Processing, vol. 2007, pp. 1–12, 2006.
SNS), pp. 1–7, 2009. [55] F. Fuentes-Hurtado, V. Naranjo, J. A. Diego-Mas, and M. Alcañiz, “A
[35] S. A. Tuama and L. E. George, “Retina recognition based on texture hybrid method for accurate iris segmentation on at-a-distance visible-
analysis: Building a system for individual recognition based on vascular wavelength images,” EURASIP Journal on Image and Video Processing,
retina pattern,” LAMBERT Academic Publishing (LAP), pp. 1–7, 2016. vol. 2019, no. 1, pp. 1–14, 2019.
[36] J. Liebers, M. Abdelaziz, L. Mecke, A. Saad, J. Auda, U. Gruenefeld, [56] V. Nazmdeh, S. Mortazavi, D. Tajeddin, H. Nazmdeh, and M. M. Asem,
F. Alt, and S. Schneegass, “Understanding user identification in virtual “Iris recognition; from classic to modern approaches,” in 2019 IEEE
reality through behavioral biometrics and the effect of body normaliza- 9th Annual Computing and Communication Workshop and Conference
tion,” in Proceedings of the 2021 CHI Conference on Human Factors (CCWC). IEEE, 2019, pp. 0981–0988.
in Computing Systems, 2021, pp. 1–11. [57] S. Sukumaran and M. Punithavalli, “Retina recognition based on fractal
[37] I. Stylios, S. Kokolakis, O. Thanou, and S. Chatzis, “Key factors driving dimension,” IJCSNS Int J Comput Sci and Netw Secur, vol. 9, no. 10,
the adoption of behavioral biometrics and continuous authentication pp. 66–7, 2009.
technology: an empirical research,” Information & Computer Security, [58] S. A. Tuama and L. E. George, Retina Recognition Based on Texture
2022. Analysis: Building a system for individual recognition based on vascu-
lar retina pattern. LAP LAMBERT Academic Publishing, 2016.

www.ijacsa.thesai.org 791 | P a g e