[Link].

in
International Journal Of Engineering And Computer Science
Volume 12 Issue 07 July 2023, Page No. 25774-25800
ISSN: 2319-7242 DOI: 10.18535/ijecs/v12i07.4741

A framework for Modified Firefly Algorithm in Multimodal

Biometric Authentication System
Agbaje M.O.1, Ayankoya F.Y2, Ogu E. C 3and Oyelakun T. A.4
1,2,4
Department of Computer Science
Babcock University, Illisan-Remo, Ogun State. Nigeria
3
Department of Information Technology,
Babcock University, Illishan-Remo. Ogun State. Nigeria

Abstract
Many end users are turning to multimodal biometric systems as a result of the limitations of conventional
authentication techniques and unimodal biometric systems for offering a high level of accurate
authentication. When high accuracy and security are required, multimodal biometrics are the best choice
because to the utilization of numerous identification modalities. It is difficult to identify the best features
that contribute to the recognition rate/accuracy and have a high redundancy of features since different
features are acquired at the feature level fusion from a variety of physiological or behavioral variables. At
the feature selection level, the utilization of meta-heuristic algorithms will reduce the number of redundant
features while keeping critical feature sets that are important to biometric performance, accuracy, and
efficiency. The study demonstrated a multimodal biometric authentication system that used the features of
the face and both irises. In order to avoid being stuck at the local optimum and hasten convergence, the
Firefly Algorithm (FFA) was modified by including a chaotic sinusoidal map function and a roulette wheel
selection mechanism as deterministic processes. The results of the study demonstrated that in terms of
sensitivity, precision, recognition accuracy, and time, the proposed MFFA with multimodal outperformed
the MFFA for unimodal, bi-modal, and bi-instance. In addition to being computationally faster, more
accurate, and suitable for real-time applications, the modified method, known as MFFA, proved effective in
integrating multimodal data sets.

Keyword: Biometric, Firefly Algorithm, Multimodal, Bi-modal, meta-heuristic

1.0 Introduction
Biometrics is the science of assessing personal characteristics such as the iris, face, fingerprints, retina, palm
print, hand geometry, voice, or signatures to secure authentication [1]; [2] and is becoming the technology of
the future in the field of security [3]. The biometric system identifies people based on their physiological
and/or behavioral features. These features are sometimes known as attributes in literature [4]. Fingerprint,
face, ear, retina, palm print, iris, hand geometry, inner knuckle print, and so on are physiological traits,
whereas voice, gait, keystrokes, and signature are behavioural characteristics. Among the biological traits of
identification, the face-based approach such as iris scan, retina, lip and face recognition have been
considered to be generally used and acceptable techniques of biometric [5]; [6].
A unimodal biometric system uses only one property to perform a recognition operation, such as a
fingerprint, nose, gait, voice, face, iris, or ear [7]. However, [8] observed that existing biometric systems
have to deal with a variety of problems with the use of a single trait, such as a fingerprint image with a scar
or poor illumination of the subject in face recognition. In real-world scenarios, the majority of biometric
technologies are unimodal [9]. The limits of unimodal biometric systems have prompted researchers to focus
their efforts on multimodal biometric systems, as the biometric source may become unreliable owing to a

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25774

variety of factors such as sensor or software failure, noisy data, non-universality, and so on [10]. Combining
two or more biometric systems is a promising solution to provide more security according to [11] and,
avoiding the falsification of several biometric traits at the same time [12]. A multimodal biometric system
recognizes people using data from many biometric sources [13] and created by combining two or more
biometric features to create a recognition system. In multimodal biometric systems, information fusion is a
crucial step [14]. At different levels, fusion of different modalities can take place[15]. A multimodal
biometric must consider a fusion of features to be unique, in which at several phases of a recognition system,
biometric features can be fused [16] either at fusion-before-matching, that involves integrating biometric
data before matching templates i.e. sensor level and feature level; or fusion-after-matching, which involves
integrating data after the matcher/classification step i.e. score level, match level, rank level and decision
level [17].
The application of meta-heuristics optimizations approach to feature selection has substantially increased as
a result of the exponential rise of real-world problems and the value of having quick access to solutions [18]
which have grown to be more common. To choose the best subset of a dataset while maintaining the model's
accuracy, meta-heuristic algorithms are therefore particularly effective and efficient [19]. This work
concentrates on feature selection issues using meta-heuristic approach based on its strength.
An efficient optimization strategy created by Yang is the firefly algorithm (FA) [20]. The three well-known
meta-heuristic methods, particle swarm optimization (PSO), differential evolution (DE) and simulated
annealing (SA) have been combined to create the FA according to [21] examined many facets of FFA in
their study and came to the conclusion that it is a promising optimization approach when it comes to
resolving the most challenging NP-hard numerical optimization issues[22] in both continuous and discrete
areas.
FFA suffers from early convergence and inadequate global exploration when dealing with challenging high-
dimensional issues, despite the benefit of avoiding the local optimal trap [18]. Due to improvements in the
other techniques, hybridizing FFA applications and modifications appear to be advantageous to standard
FFA in terms of processing speed and performance. The standard FA may need a lot of time at times to
obtain the ideal parameter values. As a result, FA should be modified or hybridized with other techniques to
speed up processing [24] and also a good idea to archive high-quality solutions by making the randomness
step length decrease with iteration [25]. This serves as the impetus for the current work, which added a
chaotic sinusoidal map function and deterministic process to the standard or already existing FA.
This study focuses on improved feature selection method in multimodal biometric systems through handling
high features dimensionality and selecting salient features which are the major problems identified in feature
level fusion that will continue to be a subject of intensive research due to its potential to improve biometric
recognition accuracy according to [20] by introducing meta-heuristic optimization techniques into the
feature selection phase of a multimodal biometric system before the classification phase to select the most
relevant features from the two biological traits. Finally, to classify a person as authorized or unauthorized,
the fused feature vectors were fed into a Support Vector Machine (SVM) classifier.
In this study, the physiological biometrics of the face and iris are used to support the finding of [26] that iris
recognition is one of the most accurate biometrics while face recognition is the most natural and acceptable
for use in identity verification.
In order to address the issues raised above, particularly high redundancy and irrelevant features, various
meta-heuristic optimization techniques, such as Genetic Algorithm (GA), Ant Colony Optimization (ACO),
Simulated Annealing (SA), Particle Swarm Optimization (PSO) and Firefly Algorithm (FFA) have been
used in the literature as feature selection techniques to choose the best subset of the original features.
However, it still has significant limitations. Though, FA has been widely used for dimensionality reduction
technique due to its effectiveness, simplicity, and eases of implementation but still suffers from premature
convergence, an imbalance between exploitation and exploration, and a significant risk of becoming stuck in
a local optimum, especially when applied to high-dimensional optimization problems like fusion [27].
Therefore, in order to further improve the performance of multimodal biometric authentication system, this
research introduced a meta-heuristic optimization approach using a modified firefly algorithm for feature
Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25775
level fusion as an efficient feature selection algorithm to select optimal features, reduce redundant features
in the feature space and speed up convergence rate for better classification and employed Support Vector
Machine (SVM) as the classifier. The study combined the two face-based (iris and face) biometric
modalities.
2.0 Review Of Literature
2.1 Firefly Optimization Algorithm
The firefly optimization algorithm is a swarm intelligence or a meta-heuristic algorithm inspired by the
flashing behaviour of fireflies and the phenomenon of bioluminescent communication [29]. Firefly
Algorithm (FFA) is a reliable and efficient meta-heuristic algorithm capable of solving many real-world
problems such as scheduling, optimization problems in dynamic environments, and economic load dispatch
problem [30]. This algorithm is influenced by the flashing behaviour of fireflies to attract one another [31].
The firefly attraction process was modeled by the algorithm for solving optimization problems [32].
Brightness and attraction are the two key components of the Firefly algorithm. The brightness of the firefly
shows its supremacy in terms of location and determines its movement direction. The travelling distance of
the firefly is determined by the degree of attraction [33]. The brightness and degree of attraction are adjusted
on a regular basis to meet the optimization goal. In the firefly algorithm sorting of flies is achieved by
sorting algorithm. It is constructed based on three rules:
(i) All fireflies are unisex so that one firefly is attracted to all other fireflies.
(ii) The attractiveness of a firefly is proportional to its brightness. For any two fireflies, the dimmer one
is attracted by (and thus moves towards) the brighter one. However, if there are no fireflies brighter
than a given firefly, the fireflies will move randomly.
(iii) The brightness of a firefly decreases as the distance from it increases. This is because the light is
absorbed when it passes through the medium.

Furthermore, because the parameters are constant during all iterations, the search behaviour remains
consistent for any condition. As a result, one of the study questions has been to improve the typical firefly
algorithm's performance as discovered by [18].

2.2 Classification of Firefly Algorithm

Different researchers have proposed various FA iterations. There is also a multi-objective optimization
variant of the firefly algorithm. For the purpose of enhancing performance, FA might introduce chaos. For
the sake of performance improvement, hybrid firefly varieties are also created. There are two most efficient
ways to classify the firefly algorithm generally:
i. Parameter Tuning: Before performing an algorithm, parameter tuning involves determining the
optimal values of the utilized parameters and modifying them over the course of iterations.
ii. Parameter Control: During the execution process, parameter control modifies the value of the
parameter.
The behaviour of the Firefly algorithm depends on both the appropriate parameter value and the
characteristics of the method's constituent parts [34]. Therefore, the following factors should be taken into
account while classifying FFA:
- What has been changed or modified?
- How are these modifications made?
- What is the extent of the modification?
2.3 Components of Firefly Algorithm (FFA)
The light intensity and the attractiveness of the fireflies are two key components of the firefly algorithm.
i. Light Absorption: The brightness of the firefly, which is expressed and quantified using a sort of
fitness function, determines how much light is produced by each source.
ii. Attractiveness: it is determined by brightness, which depends on light intensity. In the literature, the
fitness function was applied as the attractiveness function.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25776

2.4 Standard Firefly Algorithm
The FFA is modelled after how fireflies communicate by flashing their lights. The algorithm assumes that
all fireflies are unisex, meaning that any firefly can be attracted to any other firefly; a firefly's attraction is
proportionate to its brightness, which is determined by the objective function. A brighter firefly will attract
it. Furthermore, according to the inverse square law, the brightness diminishes with distance, as seen in Eqn.
(1) below.

. . . Eqn. (1)

When light passes through a medium with a light absorption coefficient γ, the light intensity at a distance of
r from the source can be calculated using Eqn. (2).

. . . Eqn. (2)

Where, IO is the intensity of light at the source. In the same way, the brightness β, can be calculated using
Eqn. (3).

. . . Eqn. (3)

In Eqn, (4), there is a generalized brightness function for ω ≥ 1. In fact, one can use any monotonically
declining function.

. . . Eqn. (4)

Each firefly will follow fireflies with higher light intensity after the intensity or brightness of the solutions is
assigned. The brightest firefly will conduct a local search by traveling around in its immediate vicinity at
random. If firefly j is brighter than firefly i then firefly i will migrate towards firefly j using the updating
method for two fireflies in Eqn. (5).

. . . Eqn. (5)

Where,
β0 is the attractiveness of xj at r = 0,
γ is an algorithm parameter which determines the degree in which the updating process depends on the
distance between the two fireflies
α is an algorithm parameter for the step length of the random movement and
ε( ) is a random vector from uniform distribution with values between 0 and 1.
For the brightest firefly, xb, the second expression in Eq. (6) will be omitted, as given in Eq. (6).

. . . Eqn. (6)
In FA, the form of attractiveness function of a firefly is depicted by the following:
(7)
where,
r = The distance between any two fireflies
β0 = The initial attractiveness at r = 0
γ = An absorption coefficient which controls the decrease of the light intensity.
The distance that exist in-between any two fireflies i and j, at a particular position xi and xj, can be defined
respectively as a Cartesian or Euclidean distance as show below:
Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25777
 (8)
where,
d is the dimensionality of the given problem.
The pattern of movement of a particular firefly i that is attracted by another firefly j that is brighter can be
represented by the following equation:
(9)
(10)
Advantages of Firefly Algorithm
(i) It is not complicated which makes it easy to understand and implement.
(ii) It can be used in different application for different discipline.
(iii) It is highly effective and efficient.
(iv) It accepts modification to boost its performance and makes it suitable for a problem at hand.
Disadvantages of Firefly Algorithm
(i) It is prone to premature convergence
(ii) It can easily get trapped in local optimum for multimodal biometrics
(iii) Its update depend on current performance
(iv) No memory of previous best solution and performance
2.5 Modification of Firefly Algorithm
The firefly algorithm, like any other meta-heuristic algorithm, is vulnerable to parameter values. It has been
discovered that altering the parameters based on the state of the search is effective. As a result, modifying
the settings is a straight-forward way to improve the performance of the firefly algorithm according to [35].
There are three (3) classes of modification for Firefly algorithm namely: Parametric Modification,
Modification to Formulas and Modifications on the Search Space.
Class 1: Parametric Modification - The parameters of the algorithm are changed in this category but, the
same updating techniques or formulas are used.

Class 2: Modifications on the Search Space - It may be easier to transit to another 'easy-to-search' space,
as well as changes in the probability distribution while generating random numbers, if the same updating
technique is used.

Where, is the existing firefly parameter;

is the chaotic gauss mapping firefly parameter,
where is considered as chaotic map parameters.
Class 3: Modification to Formulas - It covers changes such as adding mutation operators, changing part or
all of the updating formulas, and so on.
2.6 Theoretical Framework
The theoretical framework of this study shows the relationships that exist among different types of fusion in
multimodal biometrics system. Biometric fusion is a mechanism which can combine the outcome from all
the biometric modalities. The primary motivation behind this fusion is to make the biometric systems more
secure. However, matching-score level fusion and decision level fusion are more popular in the literature
and there is not much research on feature level fusion. Currently, feature level fusion is more effective than
other alternative levels of fusion because it provides greater information on the biometric input information
than the coordinating score or the classification output of a classifier but, are more prone to irrelevant data
and high dimension vector space. In order to address this issue, the feature level fusion is subsequently
provided with a meta-heuristic optimization approach as a feature selection technique for
eliminating redundant features and selecting the best features to improve the classification performance.
Figure 2.22 below shows the theoretical framework.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25778

Figure 2.23: Theoretical Framework (Researcher‟s design)

2.7 Review of Related Works

Numerous studies on optimization techniques have been conducted utilizing various meta-heuristic
algorithms in domains like engineering design, mathematics, computing, economics, and route planning.
Time and resources are limited in the real world. The objectives of the optimization algorithms are to
maximize these resources that are already available. There are several promising meta-heuristic algorithms.
One of the most current meta-heuristic algorithms, the firefly algorithm (FFA) has been employed in
numerous applications and has undergone different modifications and hybridization for better performance
of the system.

Recently, [38] proposed a return-cost-based binary FA, a powerful feature selection technique based on the
Firefly algorithm (FA) (RcBBFA). By using binary variables, the suggested method expands on the FFA
concept. The new algorithm, which is particularly successful in handling the feature selection (FS)
problems, used three novel strategies: return-cost attractiveness, Pareto dominance-based selection, and
binary movement with adaptive leap. To gauge a firefly's attractiveness compared to other fireflies, a return-
cost indicator is first defined. The most appealing option is then offered for each firefly using a Pareto
dominance-based technique. In order to update a firefly's position, a binary movement operator based on
return-cost appeal and adaptive leap is created. The suggested strategy was demonstrated to be a highly
competitive option for resolving feature selection issues by experimental findings on a number of open
datasets.

The attraction of fireflies (FA) meta-heuristic-based optimization technique was used by [39] to propose a
solution to the balancing curriculum problem. In order to evaluate the effectiveness of the proposed solution,
a series of tests and actual situations were run. The study aimed to provide a system that would streamline
the process of establishing a curricular network in higher education institutions. The experimental results

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25779

demonstrated that the new method finds the known optimum in the majority of the tested cases and
converges rather quickly.

In another study, two types of improved firefly algorithms, the inertia weight based firefly algorithm and the
chaos based firefly algorithm, were compared in [40] research. Each algorithm's principle is examined. The
three algorithms' optimization performances were simulated and compared to five common two-dimensional
or multi-dimensional benchmark functions. The results reveal that CSFA has the best accuracy and stability,
and that it can effectively balance global and local search capabilities while overcoming the flaws of the
classic firefly algorithm.

[40] used evolutionary optimization approaches to construct microstrip antennas with different goals.
Particle Swarm Optimization, Genetic Algorithms, and the Firefly Algorithm were among the biologically
inspired algorithms included in the new software, Antenna Optimizer, which combined the electromagnetic
design environment of CST Microwave Studio with the technical computing and programming environment
of MATLAB. For this uni-planar antenna design method, FA performed better than PSO and GA; the study
then proposed modified FA to create optimal parameters that match the given design requirements.

Recently, [41] presented a novel immune multi-population firefly algorithm (IMPFA) to solve multimodal
function optimization problems. The suggested approach combines a multi-population clonal selection
technique with a genetic algorithm (NUMCSA). The MPFA based on multi-population learning mechanism
is used to search globally in the feasible region, followed by the NUMCSA to search locally to improve the
accuracy of the sub-optimal solutions found with MPFA. The IMPFA is particularly effective and boosts the
precision of solutions, according to simulation data. When tackling high-dimensional and complicated
optimization problems, a higher iterative count of recurrent search is necessary, this takes longer time and
leaves potential for algorithm adjustment to strike a balance between exploration and exploitation.

Also, the ability of the firefly algorithm to achieve the best results for optimization problems (maximization
or minimization) in a bottling company that manufactures a variety of different kinds of soft drinks with
various flavours was introduced by [42]. The problem is developed as a linear programming model. The
model is run through Lindo software, and the output from both techniques is compared in order to gauge the
firefly algorithm's effectiveness. The business can determine whether products can still be manufactured
based on the availability of raw resources by using the firefly algorithm. It is also expanded to compare each
algorithm's result and select the best one. This study can be expanded to create a model based on other
inventory-related aspects.

[43] applied the distributed computing concept to an optimized version of the Firefly Algorithm (FA) and
proposed a parallel version of the Firefly algorithm to an MLTP problem with natural images to analyse the
real speed of a serial firefly algorithm adapted for MLOTP purposes and determine if distributing it is a
viable option and then to analyse the performance of the distributed algorithm. There was no discernible
difference in results between the serial and parallel versions. Furthermore, employing a collection of
computer nodes, the time might be greatly decreased.

Several researches in the literature employed the Firefly Algorithm for feature selection. [21] proposed a
variation on the Firefly Algorithm (FA) for the selection of discriminative features in regression and
classification models to aid in the support of decision-making processes utilizing data-based learning
techniques. The FA variant uses Simulated Annealing (SA)-enhanced local and global promising solutions,
chaotic-accelerated attractiveness parameters, and diversion mechanisms of weak solutions to avoid falling
into the local optimum trap and to address the issue of premature convergence in the original FA algorithm.
The statistical test results showed that the proposed FA variant is successful at improving classification and
regression models to aid decision-making processes.

Another research [44] proposed the use of Genetic Algorithm (GA) and Firefly algorithm (FFA) for
estimating the amplitudes of the Cancellation subcarriers (CCs) that were inserted on either side of the used
Non Contiguous orthogonal frequency division multiplexing NCOFDM signal to cancel the side lobes. The

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25780

suggested methods produce greater side lobe reduction than current strategies in the literature, and the FFA's
overall performance is superior, according to simulation data.

In [45] an improved maximum power point tracking (MPPT) algorithm for photovoltaic (PV) systems in
partial shadowing conditions (PSCs) was presented. It is based on the fusion firefly algorithm (FFA) and
uses a novel simplified propagation process (SPP). The proposed FFA was capable of accurately tracking
the global maximum power points (GMPPs) by combining the neighbourhood attraction firefly algorithm
(NaFA) and simplified firefly algorithm (SFA). Furthermore, the suggested SPP approach decreases
sampling events by removing duplicate propagations, speeding up tracking and minimizing energy loss and
oscillations during the sample process. A hardware assessment device was used to mimic the planned FFA's
performance and the speed enhancement brought forth by the SPP procedure. The suggested FFA algorithm
provides good accuracy and efficiency with quick tracking speed.

The study by [37] focused on the functions of diabetes prediction, where an issue with data imbalance
makes it difficult to forecast the correctness of diabetes data. The Proposed Tailored Firefly Algorithm and
Map Reduce are utilized to increase the efficacy and precision of prediction. A variety of categorization
techniques are applied with the goal of enhancing the performance of the new enhanced Firefly by
comparing it to several benchmark algorithms. The suggested EFA approach is utilized to enhance the
accuracy of the diabetic data set, together with fuzzy sets and map reduce. The new approach speeds up
prediction, increased accuracy, and decreased time.

In recent study, meta-heuristic algorithm was introduced by [46] to tackle and reduce the demand side
management problem and provide the optimum outcome. Particle Swarm Optimization, Firefly Algorithm
and Salp Swarm Algorithm were the three evolutionary algorithms that were employed. This study's main
goal is to enhance the use of renewable resources for electricity generation while reducing overall load
demand, operating costs, and utility costs across three separate sectors; residential, commercial, and
industrial. In all the three instances, it is discovered that the load after load shifting curve is more closely
related to the project's objective curve, proving that the firefly algorithm was successful in achieving the
project's goals by lowering costs and reducing peak demand. In comparison to PSO and SSA, the Firefly
algorithm is discovered to have the best cost and better accuracy.

2.8 Application of Firefly Modification

[47] proposed a modified version of the firefly algorithm based on the existing firefly algorithm and
enhanced particle swarm optimization. The increased velocity notion of particle swarm optimization is used
in the suggested modified method to improve the conventional algorithm's searching performance. Through
simulations, the firefly algorithm and the modified firefly algorithm are compared for a few common
benchmark functions. The proposed improved algorithm converges more quickly and produced the best
answer while improving the performance of the original Firefly algorithm.

In order to enhance the original Firefly algorithm, [41] hybridized the Firefly algorithm by adding mating
behaviour and the GA crossover operator (FA). The suggested method is shown to perform noticeably better
than canonical FA and a variety of other comparison techniques. The technique will still find a local
minimum for some functions.

Another modified Firefly Algorithm-based feature selection method was proposed by [48]. The classic
Firefly Algorithm is modified by the use of the K-Nearest Neighbourhood (K-NN) classifier and an
additional feature selection stage. The proposed approach is tested on four different datasets containing
diverse forms of attacks. The proposed modified Firefly Algorithm successfully reduced the dimension of
data by picking features based on the accuracy rates of the K-NN approach. The proposed modified FFA
lowers the dimension, which results in a more than 50% reduction in memory usage. The outcomes
demonstrate a method that conserves both time and memory. Though, the distance formula for large
dimensional data had a problem.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25781

In another study, [49] proposed an improved firefly algorithm (IFA) by combining two specialized
operations: sigmoid-based attractiveness to increase the likelihood of finding a viable solution and, the local
refinement capacity to strengthen the use of a dynamical step parameter tuning technique. These two
customized activities are anticipated to complement one another and achieve a good balance between global
exploration and local exploitation. Twelve well-known benchmark functions were used as the basis for six
algorithms' numerical trials and statistical analyses. IFA is a strong and competitive algorithm that surpasses
the majority of other FA variations, as shown by the numerical results and statistical comparisons.

An updated firefly technique was proposed in [50]. In this study, fireflies are modelled by simulating their
position by altering the cell chi-square value after each movement and their intensity by computing a set of
different fitness functions as a weight for each characteristic. K-Nearest Neighbour and Discriminant
Analysis were used as classifiers to assess the suggested firefly strategy for selecting features. The research
suggested either developing a firefly algorithm based on parallel processing to process all fireflies
simultaneously, thereby speeding up processing, or developing a comprehensive model using firefly
approaches, in which both feature selection and classification procedures are based on fireflies.

In order to boost diversity, a craziness operator is added to the existing firefly algorithm in [51] paper to
create the crazy firefly algorithm. The updated method makes use of the crazy factor to broaden the scope of
the normal algorithm's search behaviour. The feasibility and effectiveness of the proposed algorithm were
confirmed by testing the benchmark functions. Results show that when the proposed method is used with a
certain set of control settings, it performs better than the traditional approach. The suggested modification
algorithm improves the standard firefly algorithm's performance convergence even more quickly to the best
solution while requiring less time to produce it.

3.0 Methodology
Due to the drawbacks of a single biometric solution, researchers are becoming interested in multimodal
biometric user recognition system which combines information from multiple biometric traits of an
individual at several levels. In feature level fusion, features from several biometric modalities are fused to
create a total feature vector with a large dimension feature space, redundant and irrelevant data increasing
computing cost and impairing the performances of the classifiers. This can be addressed by using feature
selection to obtain a subset of optimal features. By lowering the dimensionality of datasets, we can reduce
the computing time, memory usage and classification error an increase the system accuracy.
Meta-heuristic optimization techniques are appropriate for the selection of features because feature subset
representation is direct and the evaluation is easily accomplished. Here, a promising meta-heuristic approach
Firefly Algorithm (FFA) because of its simplicity and efficiency is being used as a feature selection
technique for feature level fusion in multimodal biometric recognition system. The selection of features is a
general problem common to large datasets. Fewer researchers utilize swarm intelligence algorithms for
feature selection than those that use statistical techniques. As a result of its capacity to choose the most
suitable features used for classification, the application of swarm intelligence algorithms has become a
motivator for researchers to address dimensionality issues. The Firefly Algorithm (FFA) is regarded as a
meta-heuristic algorithm that was inspired by the behavior of the flashing lights of actual fireflies. It is one
of the most recent proposed swarm intelligence (SI) algorithms. With its enhanced search capabilities, it can
solve multimodal optimization problems and allow fireflies with low light intensities to travel toward nearby
fireflies with higher light levels. To improve the feature selection process and potentially reach the best
classification accuracy possible.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25782

 Iris Face Face, Iris dataset

- Filtering
Pre-processing - Detection (ROI)
- Cropping
Segmentation - Noise removal
- Normalization
Feature dimensionality reduction,
Feature Extraction Using Principal
extraction and separation
Component Analysis

Feature Concatenation Feature concatenation using serial method

Normalization of extracted feature Feature scaling Using Min-max

Modified Firefly Algorithm for

Fusing Normalized Features Using
fusing normalized features
MFFA

Generate matching Score Using Support Vector Machine for

‘ Recognition/Classification
Support Vector Machine

Identification of users
- Accept or Reject User

Result Evaluation

Figure 3.1: MFFA Conceptual Design (Researcher‟s Design)

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25783

3.1 Meta-heuristic Approach for Feature Level Fusion for Multimodal Biometric Authentication
System
The approach that was employed in the proposed study to achieve the objectives of this work is discussed
here. Data collection, feature extraction, feature selection, and classification are the four processes that make
up the proposed multimodal biometric authentication system.

i. Image Acquisition
Face and iris images acquisition refers to the capture of both face and iris images simultaneously using an
iris camera. A CMITECH IRIS Camera device was used to accomplish this. The subjects involved some
interested Ladoke Akintola University of Technology, Ogbomoso (LAUTECH) students‟ and staff within
the campus. The study took into consideration 840 subjects with three different expressions each for the two
biometric traits. The total datasets captured was 7560; 70 percent of the dataset were used for training and
30 percent for testing.
ii. Image Preprocessing
Face and iris images were preprocessed to extract only the parts of the image that contain useful
information. Preprocessing techniques were performed differently on face images and iris images in their
datasets. The preprocessing phase of facial images involved were image cropping, image resizing and image
enhancement using Histogram Equalization. The preprocessing phases of iris images involved were iris
localization/segmentation and iris normalization.
iii. Feature Extraction
Through the process of feature extraction, enormous amounts of redundant data are minimized and reduced
computational complexity of the system is made possible. Using Principal Component Analysis, the feature
values from the preprocessed data are extracted for this study's feature extraction.
iv. Feature Concatenation
This is the feature concatenation at the feature extraction level that aids in the consideration of those features
that have the greatest impact on verification accuracy. This function maximizes the performance of the
biometrics system by picking features extracted in the previous steps. In the study, the extracted features of
both face and iris were combined to have a single feature vector using serial rule method. The feature
extracted string size for face is 70 x 70, 60 x 60 for left iris and 60 x 60 for right iris. After the application of
serial method to the initial string size; the concatenated string size is 170 x 170.
V. Feature Normalization
Features extracted from the face and irises are high dimensional, due to the variation in distribution and
range. The high dimensional issue is overcome by normalizing the features. The features extracted by the
PCA from each biometric trait were heterogeneous. The normalization of the face and iris features was
achieved by using Min-max method. The Min-Max technique was used to retain distribution and map the
features into a common range. Normalization maps the raw biometric features to the interval [0, 1] and
retains the original distribution of the features. The normalization of the features of the biometric traits by
the min-max rule is given in Equation 3.1,
(3.1)
Where,
is the original value,
is the normalized value,
is the maximum weight and,
is the minimum weight.

v. Feature Level Fusion

In the feature-level fusion process, features that are extracted from the raw data are fused with one another
to produce a single feature set that is sent to the subsequent classification stage. After the extracted features
from face and both irises images have been normalized, we move to the feature level fusion phase to
combine the extracted feature vectors. The two main stages of feature level fusion are feature fusion and
feature selection:

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25784

a. Feature Fusion
A single vector that boosts the feature vector's discriminating strength has been generated by fusing the
normalized features of the face and both iris images. The most important/optimal features are chosen from a
vast range of features using certain significant feature selection strategies.

b. Feature Selection
The feature selection technique was developed to choose the most important features from a dataset,
increase prediction accuracy, and eliminate redundant and unimportant features for a better understanding of
the dataset.
vi. Feature selection Using Modified Firefly Algorithm
To enhance performance and reduce extracted feature dimensions for better classification in this study, the
best features were selected using a feature selection technique, modified firefly algorithm (MFFA).
a. The Existing Firefly Algorithm (FFA)
FFA is a type of meta-heuristic algorithm that draws inspiration from the nighttime illumination behavior of
fireflies. The primary function of a firefly flash is to act as a signaling mechanism to draw in additional
fireflies. The following are the three guiding concepts behind firefly movement.
1) Fireflies are all genderless. Any firefly can draw the attention of another firefly.
2) There is an inverse relationship between firefly distance and attractiveness. With more gaps between
fireflies, attractiveness will diminish. It will move randomly throughout the search area if there are no
fireflies that are brighter than it.
3) The objective function that needs to be optimized determines how bright a firefly will be.
The attractiveness of the FFA can be calculated by using the equation (1)

(1) [50]
where, r = Distance between any two fireflies, β0 = Initial attractiveness at r = 0
γ = Fixed light absorption coefficient which controls the decrease of the light intensity
As presented in the equation (2) below, the distance between any two firefly, i and j, at a given position, xi
and xj, can be characterized as either a Cartesian distance or a Euclidean distance.

(2) [50]
where, d is the dimensionality of the given problem.
The attractiveness formula that indicates new position of less bright firefly to move to the brighter one is
calculated using equation (3)
(3) [50]
where,
= the first term is the present position of a firefly,

= movement of firefly towards the most attractive of the fireflies by the intensity
of light.

= random movement of a firefly when it lacks the brighter ones

α = a parameter for randomization.

[48]

𝑥𝑖 represents the current value of each feature,

(𝑥𝑖) is the probability of 𝑥𝑖 taking 1

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25785

 [48]

b. Derivation of Modified Firefly Algorithm (MFFA) as a feature selection technique.

Modification was made to the existing Firefly Algorithm at the movement of the brightest fireflies and at the
attractive phase of the fireflies, to allow a balance of exploitation and exploration which gives room for
selecting well-balanced and relevant features that is expected to reduce the false positive rate (FPR),
computational time and increase the accuracy of the system. The modified firefly algorithm (MFFA) was
developed from the existing firefly algorithm by introducing roulette wheel selection as a deterministic
process instead of a random process and a chaotic sinusoidal map function in order to resolve the problem of
premature convergence. The algorithm is shown in Figure 3.8. Figure 3.9 described the flowchart for
modified Firefly Algorithm. We derived the firefly modification by introducing a deterministic process into
the existing FFA instead of random process and integrating a chaotic theory into the attraction phase of FFA
using a sinusoidal map function.
Modified Firefly Algorithm (MFFA)
In the existing Firefly, the procedure starts from an initial population of randomly generated individuals. The
quality of each individual is calculated using equation 3.2 and the best solution among them is selected. In
FFA, the form of attractiveness function of a firefly is depicted by the following:
(3.2)
where,
r = The distance between any two fireflies
β0 = The initial attractiveness at r = 0
γ = An absorption coefficient which controls the decrease of the light intensity
The distance that exist in-between any two fireflies i and j, at a particular position xi and xj, can be defined
respectively as a Cartesian or Euclidean distance as shown below:
(3.3)
where,
d is the dimensionality of the given problem.
The pattern of movement of a particular firefly i that is attracted by another firefly j which is brighter can be
represented by the following equation:
(3.4)
(3.5)
In equation 3.4, the term which is the first term is the present position of a firefly;
The term which is the second term is meant for movement of firefly towards the
most attractive of the fireflies by the intensity of light and;
The third term is meant to cater for the random movement of a firefly (random part), when it
lacks the brighter ones. The α coefficient is a parameter for randomization, its value depends on the problem
that is to be solved, while „rand‟ is consistently distributed in the space (0, 1) as it is a random number
generator. In equation 3.5, the movement of the best candidate is done randomly.
The modified Firefly Algorithm was formulated using Equation 3.6 to model the pattern of movement of
firefly as a deterministic process instead of a random process in the existing firefly. The pattern of
movement of a particular firefly i that is attracted by another firefly j that is brighter was modified by
roulette wheel selection ( as expressed in Equation 3.3.
(3.6)
Where, is the objective function value of the firefly. Given the modified pattern of movement of
firefly as in Equation 3.5
(3.7)

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25786

 (3.8)
In equation 3.7, the term which is the first term is the present position of a firefly, the term
which is the second term is meant for movement of firefly towards the most
attractive of the firefly by the intensity of light and the third term is meant to cater for the random movement
of a firefly (random part), when it lacks the brighter ones. The α coefficient there is a parameter for
randomization, its value depends on the problem that is to be solved, while „ ‟ is consistently distributed
using roulette wheel selection. In equation 3.8, the movement of the best candidate is done randomly.
The challenges of imbalance between exploration and exploitation experienced in the existing firefly
algorithm were resolved in this study by modifying the attractiveness of the firefly with the application of
chaotic theory using a sinusoidal chaotic map function. This describes the chaotic absorption coefficient
( ) which controls the decrease of the light intensity, but not limiting these fireflies to search space
boundaries but count on the nature of the chaotic system that generates random and unpredictable outputs
from preceding conditions. The new attractiveness of the firefly was expressed in Equation 3.9, Equation
3.10 and Equation 3.11.

(3.9)

(3.10)

(3.11)
Where,
is the initial attractiveness at r = 0,
r is the distance between any two fireflies,
γ is an absorption coefficient which controls the decrease of the light intensity,
is the existing light intensity update;
is the chaotic sinusoidal mapping, where as chaotic map parameter.
is calculated to transform the .
is the modified updated light intensity of the firefly. The modified firefly algorithm (MFFA) is thus
established in this section.

Objective/Fitness Function of fused features selected by Modified Firefly Algorithm

In this work, the broad formulation of the problem of selected fused features was given:

Subject to: C1:

C2:

= accuracy - (k / )
Where, k is a penalty term that discourages the use of too many features and
is the total number of features in the dataset.
The desired trade-off between accuracy and simplicity can influence the penalty term.
Where,
are the vectors of the normalized feature state variables, respectively.
is the mean square error for .
The entire state vector is denoted as ,
Where, is the set of the feature vector of .
The problem was defined on the feature‟s horizon .

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25787

Where, consists of the original feature of of y and final feature selected from the normalized
feature.
The decision variables for optimization include the local and global best position feature-dependent control
variables as well as potentially the final feature . Finding the ideal set of decision variables to
minimize the fitness function , that is, is the aim of the optimization.
The search space for the optimum is limited by constraints that specify the proper fitness error and feature
parameter requirements to be met during feature selection at the feature selection stage.

In this study, fitness constraint and feature constraint were treated as C1 and C2, respectively. C1 made sure
that the feature values fall between 0 and 1.
C2 validated that the fitness value for the features to be chosen was marked as 1 and the irrelevant features
was labelled 0.
3.2 String Sizes for Face and Iris
String sizes for the face and iris biometric traits are displayed in the table 3.1 below:

Table 3.1: String Sizes for the Face and Iris biometrics

S/N TECHNIQUES FACE LEFT IRIS RIGHT IRIS

1 Original Size 720 x 960 640 x 480 640 x 480

2 Pre-processing Size 100 x 100 100 x 100 100 x 100

3 Feature Extraction 50 x 50 60 x 60 60 x 60

4 Concatenated features
170 x 170
5 Feature Selection 20 x 20 24 x 24 24 x 24

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25788

 Start

Input Feature Extraction for

Face and Iris Feature
Concatenation

Evaluate the objective function values of

firefly

Define firefly parameters and calculate light

intensity and absorption coefficient

Tuning of attractive parameter. Move Firefly i towards j. Vary attractiveness of each

firefly with distance r with roulette wheel selection

Update value of beta with decreasing iteration using chaotic

sinusoidal map function

Evaluate new solution and update light intensity

Iteration= max
generation?

Select the optimal features for the feature level fusion of

face and iris features

Stop

Figure 3.2: Flowchart for the Modified Firefly Algorithm (MFFA) (Researcher‟s Design)

2b. Development of a feature selection technique for multimodal system

3.3 Classification Using Support Vector Machine (SVM)
The fireflies (features) created in the preceding phase are presented to a classifier during the classification
phase. Support Vector Machine (SVM), one of many machine learning approaches that can be employed as
classifiers, was used in this study. The features selected by MFFA technique as derived in the objective 2a
above were classified using Support Vector Machine (SVM) which is a statistical learning theory-based
controlled classification technique. To differentiate between classes, SVM creates a hyper plane in
Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25789
multidimensional space. SVM iteratively develops the best hyper plane which is then utilized to reduce the
classification error. SVM's main goal is to create a maximum marginal hyper plane (MMH) that splits the
dataset into classes as evenly as possible. The following procedures are used by SVM to find the maximum
marginal hyper plane:
a. Create hyper planes that effectively separate the classes.
b. Choose the hyper plane with the greatest separation from the nearest data points.
TRAINING PHASE

Feature selection Using

MFFA and fusion With
Analysis technique For

Feature Normalization
Principal Component

Feature Extraction
Pre-processing and
Train Datasets

Using Min-max
Enhancement

Serial rule
1 Fused
Features
Gallery
Train Datasets
2

Feature selection Using

MFFA and fusion With
Feature Normalization
Principal Component
Analysis For Feature
TESTING PHASE

Pre-processing and

Enrolled

Using Min-max
Enhancement

Biometric trait 1
Extraction

Serial rule
Classification
Using Support
Vector Machine
Enrolled
Biometric trait 2

Identification
Figure 3.3: Multimodal Biometric Authentication System (Researcher‟s Design)

3.4 Ethical Consideration

This research was carried out under the supervision of the Babcock University Health Research Ethics
Committee (BUHREC) following standard rules and guidelines. Every effort to avoid any form of ethical
conflict was made.

4.0 Data Analysis, Results And Discussion Of Findings

This session presents the MFFA model for feature selection in multimodal biometric recognition system,
implementation of the model, and the evaluation of the developed MFFA. The model is represented in
Figure 4.1

4.1 The Meta-heuristic MFFA Model for Feature Selection Technique in Multimodal Biometric
Authentication System
The MFFA model shows the meta-heuristic optimization approach for feature level fusion in multimodal
biometric recognition system adopting Modified Firefly Algorithm (MFFA) as the feature selection
technique for the face and iris biometric traits.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25790

 Acquisition of Face
Image Database

Principal Component
Anaysis (PCA) Accept

Image Preprocessing
Min-Max Normalization

Feature Extraction
Fusion at Features Selection
using
(Modified Firefly Algorithm)
Image Preprocessing
Feature Concatenation

Classification (SVM Reject

Classifier)
Acquisition of Iris Segmentation Phase
Image Database

Principal Component
Anaysis (PCA)

Fig
ure 4.1: Model for Modified Firefly Algorithm for Feature Selection Techniques (Researcher‟s Model)

4.2. Image Pre-processing

1. FACE IMAGES
Samples of the original face images before preprocessing were shown below:

Figure 4.2: Original Captured Face Images

i. Image Cropping
Face images cropping were done by making use of an image cropping toolbox of MATLAB called “imcrop”
and the samples of the cropped face images were shown below:

Figure 4.3: Cropped Face Images

ii. Image Resizing
Face images were resized through a resizing tool in MATLAB called “imresize”.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25791

iii. Contrast Enhancement of Face Images
Conrast enhancement of face images were done by using Histogram Equalization. It is also one of the
preprocessed toolbox of MATLAB called “histeq”. The enhanced face images are shown below:

Figure 4.4: Contrast Enhanced Face Images

2. IRIS IMAGES
Preprocessing of both left and right irises images acquired was carried out separately.
i. LEFT AND RIGHT IRIS IMAGES
Samples of the original left iris images before preprocessing were shown below:

Figure 4.5: Original Captured Left Iris Images

Figure 4.6: Original Captured Right Iris Image

ii. Localization of Left Iris Images
The left and right iris images gone through the preprocessing called localization separately. We localize such
that the regions of the iris were rescaled. The extraction of the intensity values into the normalized polar
representation through interpolation techniques were done. Examples of the localized left and right irises
were shown in figures 4.7 and 4.8 below:

Figure 4.7: Localized Left Iris Image

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25792

Figure 4.8: Localized Right Iris Image
iii. Iris Segmentation
For the segmentation, Hough Transform technique was employed to capture the regions of the iris and the
pupils.

Figure 4.9: Segmented Left Iris Image

Figure 4.10: Segmented Right Iris Image

iv. Normalization
After the segmentation process, the segmented iris images were converted from the circular shape to
rectangular shape using Daugman‟s Rubber Sheet Method by un-wrapping the circular region into the
rectangular blocks of constant dimensions. For the purpose of appropriately extracting features, the
segmented images were reduced to a regularized size.

Figure 4.11: Rubber Sheet Left Iris Representation

Figure 4.12: Rubber Sheet Right Iris Representation

4.3 Results Based on Performance Metrics for Unimodal, Bimodal, Bi-instance and
Multimodal Biometrics
The results for unimodal, bimodal, bi-instance and multimodal biometrics are presented based on some
performance metrics: false positive rate, sensitivity, specificity, recognition accuracy and computational
time at the thresholds of 0.2, 0.34, 0.5 and 0.75.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25793

4.3.1 Results of Unimodal Biometric system
The outcome displayed in Tables 4.1 MFFA provides evidence of how effective the strategies based on
single biometric features were. Among the unimodal biometric traits are the face, the right and left irises.
Each dataset used for testing contains 756 images. The effectiveness of the algorithms was evaluated at the
threshold values of 0.2, 0.34, 0.5 and 0.75 as displayed in Tables 4.1.

Table 4.1: Results of the Modified Firefly Algorithm for Unimodal Biometric System

MFFA UNIMODAL (FACE, LEFT IRIS AND RIGHT IRIS)

Face Liris Riris Face LIri Riris Face LIris Riris Face LIri Riris
s s
TP 529 531 543 528 530 542 527 529 541 526 528 540
FN 38 36 24 39 37 25 40 38 26 41 39 27
FP 46 44 32 43 41 29 40 38 26 38 36 24
TN 143 145 157 146 148 160 149 151 163 151 153 165
FPR 24.3 23.2 16.93 22.7 21.6 15.34 21.1 20.11 13.76 20.11 19.0 12.70
(%) 4 8 5 9 6 5
SEN 93.3 93.6 95.77 93.1 93.4 95.59 92.9 93.30 95.41 92.77 93.1 95.24
(%) 0 5 2 7 5 2
SPE 75.6 76.7 83.07 77.2 78.3 84.66 78.8 79.89 86.24 79.89 80.9 87.30
C 6 2 5 1 4 5
(%)
PRE 92.0 92.3 94.43 92.4 92.8 94.92 92.9 93.30 95.41 93.26 93.6 95.74
C 0 5 7 2 5 2
(%)
ACC 88.8 89.4 92.59 89.1 89.6 92.86 89.4 89.95 93.12 89.55 90.0 93.25
(%) 9 2 5 8 2 8
TIM 67.6 83.7 115.45 65.3 81.4 114.77 64.5 82.54 113.88 66.54 85.0 111.51
E 3 9 6 7 1 3
THR 0.2 0.34 0.5 0.75
ES
(%)

4.3.2 Results of Bimodal Biometric System

The face and iris biometrics were used to build bimodal biometric schemes; the Face-Left Iris (FACE-
LIRIS) and the Face-Right Iris (FACE-RIRIS). The results displayed in Tables 4.2 and 4.3 demonstrate how
effective the bimodal biometric feature-based techniques are. The effectiveness of the approaches included
in this study was evaluated at the threshold values of 0.34.

Table 4.2: Result of MFFA using Face-Left Iris

MFFA (FACE AND LEFT IRIS)
TP 541 540 539 538

FN 26 27 28 29

FP 34 31 28 26

TN 155 158 161 163

FPR (%) 17.99 16.40 14.81 13.76

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25794

 SEN (%) 95.41 95.24 95.06 94.89

SPEC (%) 82.01 83.60 85.19 86.24

PREC (%) 94.08 94.57 95.06 95.39

ACC (%) 92.06 92.33 92.59 92.72

TIME (%) 91.94 90.82 91.64 91.51

THRES (%) 0.2 0.34 0.5 0.75

Table 4.3: Results of the MFFA using Face-Right Iris

MFFA (FACE AND RIGHT IRIS)
TP 542 541 540 539
FN 25 26 27 28
FP 33 30 27 25
TN 156 159 162 164
FPR (%) 17.46 15.87 14.29 13.23
SEN (%) 95.59 95.41 95.23 95.06
SPEC (%) 82.54 84.13 85.71 86.77
PREC (%) 94.26 94.75 95.24 95.57
ACC (%) 92.33 92.59 92.85 92.99
TIME (%) 97.69 98.50 98.94 97.98
THRES (%) 0.2 0.34 0.5 0.75
Results for sensitivity, precision, accuracy and computational time for the MFFA technique using the
bimodal Face-Left Iris biometric are shown in Table 4.2 and Table 4.3. At the threshold value of 0.2, 0.34,
0.5 and 0.75 for the face-left iris biometrics. The results showed that the MFFA bimodal performed better
than the MFFA unimodal in terms of timing, sensitivity, precision, and recognition accuracy. The Face-Left
Iris bimodal biometric has indicated that the MFFA bimodal is more accurate and computationally less
expensive. According to the results, the MFFA bimodal performed better than the MFFA unimodal in terms
of sensitivity, precision, recognition accuracy, and time. Concluding from the computational time for the
Face-Right Iris bimodal biometric, the results demonstrated that the MFFA bimodal approach is more
accurate and computationally less expensive. However, the addition of Iris biometrics appears to offer
greater performance with a significant increase in computational time. When MFFA bimodal was applied,
the computation time appeared to be shortened. It was shown that the MFFA bimodal beat the MFFA
unimodal modal biometrics, delivering greater performance with less sensitivity, precision, computing time,
and recognition accuracy. In this work, the bimodal biometric outperformed the unimodal biometric in terms
of sensitivity, precision, and recognition accuracy. However, because the training and testing sets included
more features, the bimodal biometric required more computing time.

4.3.3 Results of Bi-instance Biometrics System

The results of the unimodal biometric test showed that a certain person's left and right irises are different
from one another. In order to analyze the bi-instance biometrics based on the left and right iris, an approach
was adopted in this study. Table 4.4 showed the results for the left and right iris bi-instance biometrics
obtained using the MFFA approaches in terms of sensitivity, precision, accuracy and computational time.

Table 4.4: Results of the Evaluation performance of MFFA using Left Iris-Right Iris
MFFA (LIRIS_RIRIS)
TP 543 542 541 540
FN 24 25 26 27
FP 32 29 26 24

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25795

 TN 157 160 163 165
FPR (%) 16.93 15.34 13.76 12.70
SEN (%) 95.77 95.59 95.41 95.24
SPEC (%) 83.06 84.88 86.24 87.30
PREC (%) 94.43 94.92 95.41 95.74
ACC (%) 92.59 92.86 93.12 93.25
TIME (%) 115.44 114.77 113.88 111.41
THRES (%) 0.2 0.34 0.5 0.75

According to the results, the MFFA bi-instance performed better than the MFFA in unimodal and bimodal in
terms of sensitivity, precision, recognition accuracy and time. The fact that the left and right iris bi-instance
biometrics computational time was faster suggests that the MFFA bi-instance is accurate and
computationally less expensive. The results demonstrate that in terms of sensitivity, precision, recognition
accuracy and time, the MFFA bi-instance surpassed the MFFA unimodal. The left and right iris bi-instance
biometrics computational time which is faster demonstrated that it is more precise and computationally less
expensive.

4.3.4 Results of Multimodal Biometrics System

The results of the performance metrics for the multimodal biometrics (face, left iris and right iris) for the
developed MFFA based on the four thresholds values of 0.2, 0.34, 0.5 and 0.75 are displayed in table 4.5
below.
Table 4.5: Result Evaluation performance of MFFA using Face, Left Iris and Right Iris (Multimodal)
METRICS MFFA (FA_LIRIS_RIRIS)
TP 559 558 557 556
FN 8 9 10 11
FP 16 13 10 8
TN 173 176 179 181
FPR (%) 8.46 6.88 5.29 4.23
SEN (%) 98.59 98.41 98.24 98.06
SPEC (%) 91.53 93.12 94.71 95.77
PREC (%) 97.22 97.72 98.24 98.58
ACC (%) 96.83 97.09 97.35 97.49
TIME (%) 130.78 131.92 130.36 130.53
THRES (%) 0.2 0.34 0.5 0.75

The results showed that the multimodal biometric system (face, left iris and right iris) gave better and more
distinguishable results than the unimodal, bimodal, and bi-instance biometric systems utilizing the same
threshold values. The MFFA method is hence more precise and computationally faster.

5.0 Conclusion And Recommendations

5.1 Conclusion
For a multimodal biometric authentication system, a modified firefly algorithm (MFFA)-based meta-
heuristic approach for feature selection was created in order to reduce the dimensionality of feature vectors
and choose the most important and balanced features from black people's faces and both iris samples taken
locally. A chaotic sinusoidal map function and the roulette wheel approach were included into the current
FFA by the newly developed technique, called MFFA. In order to improve the classification performance of
the system and prevent being caught at the local optima, it selected the most pertinent characteristics and
reduced the high dimensional feature vectors space. These features make the MFFA a useful feature
selection method for biometr identification systems. Results from our experiments showed that the

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25796

developed technique was effective in fusing multimodal feature sets. Additionally, the MFFA technique is
highly efficient computationally and appropriate for real-time applications.
5.3 Recommendations
This study is highly advised for use in institutions like schools, businesses, governmental bodies, and
forensics that need multimodal biometric authentication systems that are dependable, effective, and quick to
process in order to authenticate a person. It can also be used due to its contactless capabilities, and to a
certain extent, the biometric qualities unaffected by old age or disease can be used by the INEC to lower the
false acceptance rate where many people were denied civic rights. It can also be used in JAMB for student
registration and identification, as the currently used fingerprint biometric technology is antiquated and
unsecure. Further research would look into the potential of combining iris and face with other heterogeneous
biometric traits such as palm prints ear and fingerprints. Additionally, by utilizing the developed MFFA
using hybridized meta-heuristic optimization approach for feature selection.

References
[1] O. Al-hamdani et al., “Multimodal Biometrics Based on Identification and Verification System,” J.
Biom. Biostat., vol. 4, no. 2, 2013, doi: 10.4172/2155-6180.1000163.
[2] M. Dutta, K. K. Psyche, and S. Yasmin, “ATM Transaction Security Using Fingerprint Recognition
American Journal of Engineering Research ( AJER ),” Am. J. Eng. Res., vol. 6, no. 8, pp. 41–45,
2017.
[3] N. Taleb and S. B. Abbes, “A Robust Multi-Biometric System with Compact Code for Iris and Face,”
Int. J. Ectrical Eng. Informatics, vol. 10, no. 1, pp. 1–13, 2018, doi: 10.15676/ijeei.2018.10.1.1.
[4] D. N. Bhavya and H. K. Chethan, “Feature and Decision Level Fusion in Children Multimodal
Biometrics,” Int. J. Recent Technol. Eng., vol. 8, no. 5, pp. 2522–2527, 2020, doi:
10.35940/ijrte.e6396.018520.
[5] V. Saini, N & Kang, “Comparative Analysis of Iris Recognition Techniques : A Review,” Int. J.
Comput. Appl., vol. 2, no. 7, pp. 23–27, 2016.
[6] L. A. Alaslani, M. G & Elrefaei, “Convolutional Neural Network Based Feature Extraction For Iris
Recognition,” Int. J. Comput. Inf. Technol., vol. 10, no. 2, pp. 65–78, 2018, doi:
10.5121/ijcsit.2018.10206.
[7] S. M. Prakash, P. Betty, and K. Sivanarulselvan, “Fusion of Multimodal Biometrics using Feature and
Score Level Fusion,” Iinternational J. Appl. Inf. Commun. Engineeering, vol. 2, no. 4, pp. 52–56,
2016.
[8] F. Yahya, H. Nasir, and K. Kadir, “Multimodal Biometric Algorithm : A Survey,” Biotechnology, vol.
15, no. 5, pp. 119–124, 2016, doi: 10.3923/biotech.2016.119.124.
[9] H. Jaafar and D. A. Ramli, “A Review of Multibiometric System with Fusion Strategies and
Weighting Factor,” Int. J. Comput. Sci. Eng., vol. 2, no. 04, pp. 158–165, 2013.
[10] R. O. Mahmoud, M. M. Selim, and O. A. Muhi, “Fusion time reduction of a feature level based
multimodal biometric authentication system,” Int. J. Sociotechnology Knowl. Dev., vol. 12, no. 1, pp.
67–83, 2020, doi: 10.4018/IJSKD.2020010104.
[11] E. S. Nil and A. Chilambuchelvan, “Multimodal biometric authentication algorithm at score level
fusion using hybrid optimization,” Wirel. Commun. Technol., vol. 2, no. 1, pp. 1–12, 2018, doi:
10.18063/wct.v2i1.415.
[12] S. Soviany, V. Sandulescu, S. Puşcoci, C. Soviany, and M. Jurian, “An optimized biometric system
with intraand inter-modal feature-level fusion,” Proc. 9th Int. Conf. Electron. Comput. Artif. Intell.
ECAI 2017, vol. 2017-Janua, pp. 1–8, 2017, doi: 10.1109/ECAI.2017.08166466.
Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25797
[13] A. Saha, “An Expert Multi-Modal Person Authentication System Based on Feature Level Fusion of
Iris and Retina Recognition,” 2019 Int. Conf. Electr. Comput. Commun. Eng., pp. 1–5, 2019.
[14] M. Haghighat, M. Abdel-Mottaleb, and W. Alhalabi, “Discriminant Correlation Analysis: Real-Time
Feature Level Fusion for Multimodal Biometric Recognition,” IEEE Trans. Inf. Forensics Secur., vol.
11, no. 9, pp. 1984–1996, 2016, doi: 10.1109/TIFS.2016.2569061.
[15] C. Kant, “Analysis of Different Fusion and Normalization Techniques,” Int. J. Adv. Res. Comput.
Commun. Eng. ISO, vol. 3297, pp. 363–368, 2007, doi: 10.17148/IJARCCE.2018.7369.
[16] P. Sai Shreyashi, R. Kalra, M. Gayathri, and C. Malathy, “A Study on Multimodal Approach of Face
and Iris Modalities in a Biometric System,” Lect. Notes Networks Syst., vol. 176 LNNS, no. May, pp.
581–594, 2021, doi: 10.1007/978-981-33-4355-9_43.
[17] S. N. Garg, R. Vig, and S. Gupta, “A Survey on Different Levels of Fusion in Multimodal
Biometrics,” vol. 10, no. November, pp. 1–11, 2017, doi: 10.17485/ijst/2017/v10i44/120575.
[18] S. Asim, A. Shah, H. M. Shabbir, S. U. Rehman, and M. Waqas, “A Comparative Study of Feature
Selection Approaches: 2016-2020,” Int. J. Sci. Eng. Res., vol. 11, no. 2, p. 2020, 2020.
[19] Z. Sadeghian, E. Akbar, H. Nematzadeh, and H. Motameni, “A Review of Feature Selection Method
Based on Optimization Algorithms,” vol. 16, no. 1, pp. 57–73, 2023.
[20] R. Mythily and W. A. Banu, “Feature selection for optimization algorithms: Literature survey,”
Journal of Engineering and Applied Sciences, vol. 12, no. Specialissue1. pp. 5735–5739, 2017, doi:
10.3923/jeasci.2017.5735.5739.
[21] L. Zhang, K. Mistry, C. P. Lim, and S. C. Neoh, “Feature selection using firefly optimization for
classification and regression models,” Decis. Support Syst., vol. 106, pp. 64–85, 2018, doi:
10.1016/[Link].2017.12.001.
[22] M. Sharma and P. Kaur, “A Comprehensive Analysis of Nature-Inspired Meta-Heuristic Techniques
for Feature Selection Problem,” Arch. Comput. Methods Eng., vol. 28, no. 3, pp. 1103–1127, 2021,
doi: 10.1007/s11831-020-09412-6.
[23] A. E. Ezugwu, M. B. Agbaje, N. Aljojo, R. Els, H. Chiroma, and M. A. Elaziz, “A comparative
performance study of hybrid firefly algorithms for automatic data clustering,” 2020, doi:
10.1109/ACCESS.2020.3006173.
[24] N. F. Johari, A. M. Zain, N. H. Mustaffa, and A. Udin, “Firefly algorithm for optimization problem,”
Appl. Mech. Mater., vol. 421, no. April 2014, pp. 512–517, 2013, doi:
10.4028/[Link]/AMM.421.512.
[25] S. L. Tilahun and J. M. T. Ngnotchouye, “Firefly algorithm for discrete optimization problems: A
survey,” KSCE J. Civ. Eng., vol. 21, no. 2, pp. 535–545, 2017, doi: 10.1007/s12205-017-1501-1.
[26] A. B. Channegowda and H. N. Prakash, “Image fusion by discrete wavelet transform for multimodal
biometric recognition,” IAES Int. J. Artif. Intell., vol. 11, no. 1, p. 229, 2022, doi:
10.11591/ijai.v11.i1.pp229-237.
[27] I. Ahmia and M. Aïder, “A novel metaheuristic optimization algorithm: The monarchy
metaheuristic,” Turkish J. Electr. Eng. Comput. Sci., vol. 27, no. 1, pp. 362–376, 2019, doi:
10.3906/elk-1804-56.
[28] A. S. Joshi, O. Kulkarni, G. M. Kakandikar, and V. M. Nandedkar, “Cuckoo Search Optimization- A
Review,” Mater. Today Proc., vol. 4, no. 8, pp. 7262–7269, 2017, doi: 10.1016/[Link].2017.07.055.
[29] A. Johari, N.F., Zain, A. Z., Mustaffa, N. H & Udin, “Firefly Algorithm for Optimization Problem

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25798

 Firefly Algorithm for Optimization Problem,” Appl. Mech. Mater., vol. 421, pp. 512–517, 2013, doi:
10.4028/[Link]/AMM.421.512.
[30] A. Ritthipakdee, A. Thammano, N. Premasathian, and D. Jitkongchuen, “Firefly Mating Algorithm
for Continuous Optimization Problems,” Comput. Intell. Neurosci., pp. 1–11, 2017.
[31] N. Ali, M. A. Othman, M. N. Husain, and M. H. Misran, “A review of firefly algorithm,” ARPN J.
Eng. Appl. Sci., vol. 9, no. 10, pp. 1732–1736, 2014.
[32] P. D. Umbarka, A. J., Balande, U. T & Seth, “Performance evaluation of firefly algorithm with
variation in sorting for non-linear benchmark problems,” 2017, doi: 10.1063/1.4981972.
[33] P. Cao, “Comparative Study of Several Improved Firefly,” no. 6117411, pp. 910–914, 2016.
[34] V. Kumar and D. Kumar, “A Systematic Review on Firefly Algorithm: Past, Present, and Future,”
Arch. Comput. Methods Eng., vol. 28, no. 4, pp. 3269–3291, 2021, doi: 10.1007/s11831-020-09498-y.
[35] N. N. H. Waqar A. Khan and S. L. T. and J. M. T. Ngnotchouye, “A Review and Comparative Study
of Firefly Algorithm and Its Modified Versions,” Intech, no. tourism, p. 13, 2016, [Online].
Available: [Link]
in-biometrics.
[36] M. H. Mojarrad and P. Ayubi, “Particle swarm optimization with chaotic velocity clamping (CVC-
PSO),” 2015 7th Conf. Inf. Knowl. Technol. IKT 2015, 2015, doi: 10.1109/IKT.2015.7288811.
[37] R. Sujatha, B. Gomathi, and T. Padma, “An efficient method for high performance prediction
mechanism for diabetes using enhanced Firefly algorithm and Map-Reduce,” J. Phys. Conf. Ser., vol.
1964, no. 4, 2021, doi: 10.1088/1742-6596/1964/4/042056.
[38] Y. Zhang, X. fang Song, and D. wei Gong, “A return-cost-based binary firefly algorithm for feature
selection,” Inf. Sci. (Ny)., vol. 418–419, pp. 561–574, 2017, doi: 10.1016/[Link].2017.08.047.
[39] J. M. Rubio, C. L. Vidal-Silva, R. Soto, E. Madariaga, F. Johnson, and L. Carter, “Applying FireFly
Algorithm to solve the problem of balancing curricula,” Int. J. Adv. Comput. Sci. Appl., vol. 10, no. 1,
pp. 68–75, 2019, doi: 10.14569/IJACSA.2019.0100108.
[40] H. J. Mohammed, F. Abdulsalam, A. S. Abdulla, and R. S. Ali, “Evaluation of Genetic Algorithms ,
Particle Swarm Optimisation , and Firefly Algorithms in Antenna Design,” no. 1, pp. 3–6, 2016.
[41] H. C. Jiufang Zhou, Jianhui Wu, “Immune Multi-population Firefly Algorithm and Its Application in
Multimodal Function Optimization,” no. 1, pp. 4–7, 2018.
[42] A. Altherwi, “Application of the Firefly Algorithm for Optimal Production and Demand Forecasting
at Selected Industrial Plant,” pp. 2451–2459, 2020, doi: 10.4236/ojbm.2020.86151.
[43] G. Linhares, G. Costa, L. Jacobson, L. Rodrigues, P. S. Rodrigues, and G. A. Wachs-lopes, “Parallel
Approach to the Firefly Algorithm,” no. Session III, 2018, doi: 10.1109/CLEI.2018.00094.
[44] & N. G. Elahi, Atif, Mehreen Atif, Ijaz Mansoor Qureshi, “Interference Reduction in Cognitive Radio
Networks Using Genetic and Firefly Algorithms,” no. 1, pp. 96–100, 2017.
[45] Y. Huang, M. Huang, and C. Ye, “A Fusion Firefly Algorithm with Simplified Propagation for
Photovoltaic MPPT under Partial Shading Conditions,” vol. 3029, no. c, 2020, doi:
10.1109/TSTE.2020.2968752.
[46] S. Kitaronka, “Optimization effects of Firefly algorithm in Smart Grids,” no. February, 2022, doi:
10.6084/[Link].19115237.
[47] S. K. Sarangi, “A New Modified Firefly Algorithm for Function,” pp. 2944–2949, 2016.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25799

[48] Z. Orman, “A MODIFIED FIREFLY ALGORITHM-BASED FEATURE SELECTION METHOD
AND ARTIFICIAL IMMUNE SYSTEM,” vol. 25, no. 1, pp. 269–288, 2020, doi:
10.17482/uumfd.649003.
[49] A. Liu, P. Li, X. Deng, and L. Ren, “A sigmoid attractiveness based improved firefly algorithm and
its applications in IIR filter design,” Conn. Sci., vol. 33, no. 1, pp. 1–25, 2021, doi:
10.1080/09540091.2020.1742660.
[50] E. M. Mashhour, E. M. F. El Houby, K. T. Wassif, and A. I. Salah, “Feature selection approach based
on firefly algorithm and chi-square,” Int. J. Electr. Comput. Eng., vol. 8, no. 4, pp. 2338–2350, 2018,
doi: 10.11591/ijece.v8i4.pp2338-2350.
[51] S. K. Sarangi, “Crazy Firefly Algorithm for Function Optimization,” vol. 2, no. 1, 2017.

Agbaje M.O., IJECS Volume 12 Issue 07 July, 2023 Page 25800