European Journal of Applied Science, ©2024 The Author(s). Published under a CC BY 4.

0 license
Engineering and Technology, 2(3): 214-227, 2024 Published by AMO Publisher
DOI: 10.59324/ejaset.2024.2(3).20 www.ejaset.com

Development of Face Recognition System

Using Support Vector Machine Algorithm
Shahjahan
Mathematics Department, University of Balochistan, Quetta, Pakistan
Naveed Sheikh 
Mathematics Department, University of Balochistan, Quetta, Pakistan
Abdul Rehman
Mathematics Department, University of Balochistan, Quetta, Pakistan
Misbah Anjum
Mathematics Department, University of Balochistan, Quetta, Pakistan

ABSTRACT:
This research paper describes a face recognition system constructed with the Support Vector
Machine (SVM) method. The technique implements the Convolutional Neural Network (CNN)
approach for extracting features, which improves the accuracy of recognising faces. The method
confirms an efficient and accurate facial recognition procedure by utilising SVM's ability to detect
complex patterns within retrieved characteristics. The article explores how the system is
implemented and optimised performance, with a focus on obtaining performance in real time. The
experimental findings demonstrate the practical application of the SVM-based facial recognition
system, emphasising its numerous potential applications in security, the authentication process,
as well as interaction between humans and computers. The implementation of CNN for extracting
features provides refinement to the system, proving its flexibility to complex face characteristics
and confirming its feasibility in actual applications. The results indicate that the suggested system
is an effective facial recognition technique. Furthermore, the system is computationally efficient,
taking just a few seconds to evaluate an individual face for extracting features and recognition.
The system for identifying faces uses modern approaches to enhance both speed and precision,
making it a potential option for a variety of applications. Its capacity to swiftly and correctly detect
faces in real-time circumstances is a substantial development in facial identification technologies.
The system's capability has been comprehensively evaluated, and it satisfies the essential criteria
for practical implementation in real-world situations, making it an important addition to the
continued development of secure and efficient face recognition systems.

Keywords: Face recognition systems, Support vector machines, Vector Machine Algorithm, facial
recognition technique.

Suggested Citation: Shahjahan, N. Sheikh, A. Rehman, M. Anjum, "Development of Face

Recognition System Using Support Vector Machine Algorithm," Eur. J. Appl. Sc. Eng. Technol.,
vol. 2(3), pp. 214-227, May-Jun 2024. DOI: 10.59324/ejaset.2024.2(3).20

INTRODUCTION
The development of face recognition systems using the Support Vector Machine (SVM) algorithm
has garnered significant attention in the field of computer vision and pattern recognition [1]. SVM
has been widely utilized in various aspects of face recognition, such as facial expression recognition,
multi-class classification, and feature extraction. It has been acknowledged as an effective technique
due to its ability to accurately classify facial features into different classes using hyperplanes [2].

214
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
Additionally, SVM has been combined with other methods such as Generalized Discriminant
Analysis (GDA) and Independent Component Analysis (ICA) to enhance the performance of face
recognition systems [3]. These extracted features are sent into the SVM classifier, which is trained
on a labelled dataset to understand the patterns that are unique to each individual. The trained SVM
model may then be used to detect faces in previously unseen pictures [4]. Furthermore, the SVM
algorithm has been employed in addressing challenges related to illumination invariance and
liveness detection in face recognition systems [5]. The research on face recognition systems using
SVM has also extended to real-time applications, such as emotion recognition, speech emotion
recognition, and human fall detection. SVM has been utilized as a leading classifying algorithm in
speech emotion recognition systems [6]. Moreover, the algorithm has been integrated with deep
learning techniques for real-time human fall detection and emotion recognition using embedded
devices [7]. Furthermore, SVM has been employed in the design of a high-performance two-stage
face recognition algorithm to mitigate the effects of reduced illumination [8]. In addition to SVM, other
methods such as Local Binary Patterns (LBP), Probabilistic Linear Discriminant Analysis (PLDA),
and Gabor filters have been combined with SVM to optimize face recognition systems [9]. These
combinations have aimed to improve feature extraction, recognition accuracy, and robustness
across different facial expressions [10]. The continuous advancements in face recognition systems
using SVM have led to the exploration of novel approaches, including the development of illumination
invariant systems using Convolutional Neural Networks (CNN) and the application of FPGA-based
real-time face recognition architectures [11]. Deep learning techniques, such as Convolutional
Neural Networks (CNNs), combined with SVM have also demonstrated promising results in
improving the accuracy and resilience of face recognition systems [12]. Significant progress has
been made in the development of face recognition systems based on the SVM algorithm. To handle
complicated and non-linear data distributions, researchers have investigated many versions of SVM,
such as kernel SVM. Additionally, data augmentation approaches and robust feature extraction
methods have been used to handle the obstacles provided by fluctuations in lighting conditions,
position, and emotion. Deep learning techniques, such as Convolutional Neural Networks (CNNs),
combined with SVM have also demonstrated promising results in improving the accuracy and
resilience of face recognition systems [13]. The face recognition has one of the established & useful
patterns for the developed and under developing country even though for every industrial and
institutional organizations. Face recognition differs from other types of pattern recognition problems,
such as atmosphere recognition. The human face is a vital part of our living memory, which defines
the first identity of a person. When the wave of innovation is met, the use of the human face as a
key factor leads to the wonderful reflection sometimes after the wave's potential for a variety of uses
[14]. These developments signify the ongoing efforts to enhance the accuracy, efficiency, and
applicability of face recognition systems in various domains [15].
A face recognition technology matches a human face picture or video to other graphical faces stored
in a database. Face structure, shape, and proportions are compared during the face recognition
processes. Also, a comparison is made between the eyes, neck, mouth, and jaw, as well as the
shape of the top of the eye, the sides of the mouth, the position of the nose, and the area around the
control bones [16]. The human brain is capable of recognizing oneself & others by their facial
attributes without difficulty, as well as recognizing a friend's face long after having last seen them
[17]. Regardless of age, race, distance, or proximity, poor or good lighting, there is always the
potential to recognize a face based on prior experiences or recollections [18].
This research attempt to enhance the existing contributed knowledge regarding SVM-based face
recognition systems. By the identification strengths, weaknesses, and areas of improvement, this
research aims to facilitate the development of more precise, efficient, and consistent face recognition
resolutions, thus advance the arenas of biometrics, security, and human-computer interaction.
In this research, it has been investigated and provided ways to increase the accuracy and
performance of SVM-based face recognition systems.

MATERIALS AND METHOD

The support vector machine (SVM) method was used to construct a face recognition system, which
yielded promising results in a variety of real-world applications. It is a popular choice for face
215
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
recognition jobs due to its resilience, efficiency, and capacity to handle high-dimensional data.
However, issues like as illumination, position, and expression fluctuations, as well as the necessity
for a big and diverse training dataset, must be solved.
Face Detection and Recognition
Face detection and recognition are two associated computer vision problems that involve detecting
and recognizing human faces in pictures or video streams. Face detection seeks to locate and
localize faces in an image or video, whereas face recognition seeks to identify or verify persons
based on their facial traits.
Image Pre-Processing
An image's pre-processing refers to the techniques and operations performed on it before the image
is further analyzed or processed. The pre-processing of images enhances their quality, improves
their visual appearance, removes noise or artifacts, and extracts relevant information for subsequent
tasks such as object detection, recognition, or segmentation.
Database
The LFW (Labelled Faces in the Wild) database is a popular standard for face recognition research.
It was developed by academics at the University of Massachusetts, Amherst, and it includes a
massive database of face photos gathered from the internet. LFW has a total of 5,749 subjects and
13,233 face photos. These photographs were taken in unrestricted conditions, which means they
contain variances in position, lighting, emotion, and occlusion. The collection contains photographs
of celebrities as well as non-celebrities, making it more typical of real-life circumstances. The identity
of the individual represented is labelled on each photograph in the LFW database. It is crucial to
note, however, that the dataset may contain noise and labelling inaccuracies as a result of the
internet gathering procedure.
Feature Extraction
MobileNet has been used for feature extraction task. MobileNet is a deep learning architecture
designed for efficient and lightweight neural network models specifically tailored for mobile and edge
devices. It was developed by Google as part of the TensorFlow project. MobileNet focuses on
providing a good balance between model size, accuracy, and speed, making it suitable for resource-
constrained environments. As a feature extractor, MobileNet processes input images and extracts
relevant features or representations from them. These features are then typically used as input for
subsequent layers or models to perform tasks like classification or object detection.
Classifier
Support Vector Machine (SVM) is a powerful machine learning algorithm widely used for
classification and regression tasks. SVM operates by finding the optimal hyperplane that best
separates the input data into different classes in a supervised manner. The data points that lie closest
to the hyperplane are known as support vectors, and the corresponding non-zero values are referred
to as support values. This approach allows SVM to classify new data points by mapping them into
the appropriate class based on their position relative to the hyperplane.
Mathematically, the objective of SVM is to find the hyperplane with the maximum margin, which is
the distance between the hyperplane and the nearest data points from each class. This can be
formulated as an optimization problem, where the goal is to maximize the margin while minimizing
the classification error. The decision function of SVM can be expressed as:

𝑓𝑓(𝑥𝑥) = 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠(𝑤𝑤 𝑇𝑇 𝑥𝑥 + 𝑏𝑏) (1)

Where;
• 𝑓𝑓(𝑥𝑥) is the decision function that determines the class of the input data point 𝑥𝑥.
• 𝑊𝑊 is the weight vector that defines the orientation of the hyperplane.

216
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
• 𝑏𝑏 is the bias term.
• 𝑥𝑥 represents the input data.

Figure 1. Data points and decision boundary of SVM

Furthermore, SVM can be extended to handle non-linearly separable data by using kernel functions,
which map the input data into a higher-dimensional space where it becomes linearly separable. The
kernel trick allows SVM to effectively classify complex data that may not be linearly separable in the
original feature space. Its mathematical formulation involves maximizing the margin between classes
while minimizing classification error, and it can be extended to handle non-linearly separable data
through kernel functions.
Radial Basis Function Kernel (RBF)
The RBF kernel is the most widely used kernel concept to solve the problem of classifying datasets
that cannot be separated linearly. This kernel is known to have good performance with certain
parameters, and the results of the training have a small error value compared to other kernels. The
equation formula for the RBF kernel function is:

𝐾𝐾(𝑥𝑥𝑛𝑛 − 𝑥𝑥𝑖𝑖 ) = exp (−𝛾𝛾‖𝑥𝑥𝑛𝑛 − 𝑥𝑥𝑖𝑖 ‖2 + 𝐶𝐶) (2)

The Gaussian kernel RBF has two parameters, namely gamma and sigma. The gamma parameter
1
has a default value, which is 𝛾𝛾 = 2 . When gamma is high, the points around the data are likely to
(2𝜎𝜎)
be considered in the calculation. The sigma parameter is used to find the optimal value for each
dataset.
In the RBF kernel function equation, ‖𝑥𝑥𝑛𝑛 − 𝑥𝑥𝑖𝑖 ‖ is the Euclidean Distance between 𝑥𝑥1 and 𝑥𝑥2 in two
different feature spaces and σ (sigma) is the RBF kernel parameter that determines the kernel
weight. In SVM, sigma parameters need to be adjusted to provide accurate classification results.
The default value of the sigma parameter is σ = 1.
Experimental Setup
Google Colab is a cloud-based tool that allows users to develop and execute Python code in a
Jupyter Notebook environment. It has been used to write and run code, build models, and work with
other developers on projects. Google Colab offers free access to powerful CPUs, GPUs, and TPUs,
which can accelerate machine learning and data processing activities. Google Colab notebook has
been linked to a Google Drive account for simple file storing and sharing.

217
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
Train the Network
Google Colab had been employed to write and run Python programming. First, photos were
preprocessed using Python code, followed by feature extraction using the MobileNet network and
image classification through SVM.

RESULTS AND DISCUSSION

The network was trained on Google Colab for 200 iterations, and compared metrics like precision
(P), recall (R), F1 score, and accuracy for train and test data.
Table 1 declares the training result image file as a set of columns. The training set's accuracy is set
at 91.71%. The table classification reports provide values for Precision, Recall, F1-score, and
Support.

Table 1. Training Results

ACCURACY SCORE: 91.71%
CLASSIFICATION REPORT:
0.0 1.0 Accuracy Macro Avg Weighted Avg
Precision 0.932836 0.909091 0.917085 0.920963 0.917980
Recall 0.838926 0.963855 0.917085 0.901391 0.917085
F1-Score 0.883392 0.935673 0.917085 0.909532 0.916100
Support 149.00000 249.000000 0.917085 398.000000 398.000000

Table 2 shows that the test result is 96.49% accurate. Classification reports display the Accuracy
macro average as well as the Weighted Average.

Table 2. Testing Results

ACCURACY SCORE: 96.49%
CLASSIFICATION REPORT:
0.0 1.0 Accuracy Macro Avg Weighted Avg
Precision 1.000000 0.947368 0.964912 0.973684 0.966759
Recall 0.904762 1.000000 0.964912 0.952381 0.964912
F1-Score 0.950000 0.972973 0.964912 0.961486 0.964509
Support 63.000000 108.00000 0.964912 171.00000 171.000000

Table 3 contains statistics on the accuracy, recall, F1-score, and support for various individuals'
names. It appears to be the output of a machine learning model or statistical evaluation assessing
some type of classification or recognition task.

Table 3. Showing the Result for Precision, Recall, F1-Score

and Support for Various Individuals
Personalities Name Precision Recall F1-Score Support
Ariel Sharon 1.00 0.91 0.95 11
Colin Powel 0.90 0.96 0.93 55
Donald Rumsfeld 0.82 0.92 0.87 25
George W Bush 0.96 0.92 0.94 107
Gerhard Schroeder 0.83 0.90 0.86 21
Hugo Chavez 0.90 0.64 0.75 14
Jacques Chirac 0.83 0.71 0.77 7
Jean Chretien 0.69 0.92 0.79 12
John Ashcroft 1.00 1.00 1.00 12
Junichiro Koizumi 0.78 0.88 0.82 8

218
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
 Sarena Williams 1.00 0.71 0.83 7
Tony Blair 0.91 0.88 0.89 33
Accuracy 0.90 312
Macro Avg 0.88 0.86 0.87 312
Weighted Avg 0.91 0.90 0.90 312

Figure 2 shows the train and validation loss for 12 classes with 140 epochs. The graph demonstrates
that the blue line represents the train loss, whereas the green line represents the validation loss.

Figure 2. Graph of Face Recognition with 12 Number of Classes

Figure 2 depicts the train accuracy of the results, which comprises of 140 epochs. The graphic
contains lines that describe the results of train accuracy and validation accuracy, as well as the
number of faces that have been repeated. Figure 2 indicates that increasing the number of epochs
improves the training results.

Figure 3. Training Accuracy and Validation Accuracy with Epochs

219
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
Figure 3 depicts the link between the number of epochs and network accuracy. The graph clearly
illustrates that as the number of epochs grows, so does the network's accuracy. For the current
experiment, 140 epochs have been used.

Figure 4. Confusion Matrix

A confusion matrix describes any classifier's performance on a test dataset. Figure 4 depicts the
results of the model's classifier, SVM. Other performance matrices, like as accuracy, precision,
recall, and F1-score, are used for additional analysis of the network's performance.
Figure 5 depicts the performance of the model when it utilized fewer epochs and fewer classes. It
demonstrates that as the number of epochs rises, the training and validation losses decrease.
Figure 6 depicts the model's training and validation accuracy when it is employed 50 epochs with
fewer classes. The graph demonstrates that as the number of epochs increases, so does the model's
accuracy.
Table 4 displays the categorization report for 34 classes. And thus trained the model for 50 epochs.
Table 4 displays the performance of model's matrices, which include the names of the classes as
well as their accuracy, recall, F1-score, and support values.

220
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
 Figure 5. The Training and Validation Loss when No of Epochs is 50

Figure 6. The Training and Validation Accuracy when No. of Epochs is 50

Table 4. Classification Report

Personalities Name Precision Recall F1-Score Support
Alvaro Uribe 0.67 0.75 0.71 8
Andre Agassi 0.88 1.00 0.93 7
Ariel Sharon 0.95 1.00 0.97 18
Alejandro Toledo 0.60 0.60 0.60 5
Arnold Schwarzenegger 0.83 0.56 0.67 9
Colin Powel 0.84 0.90 0.87 41
David Beckham 0.50 0.40 0.44 5
Donald Rumsfeld 0.87 0.90 0.88 29
George W Bush 0.90 0.99 0.94 113
Gerhard Schroeder 0.86 0.86 0.86 22
Gloria Macapagal Arroyo 1.00 0.62 0.77 8
Guillermo Coria 0.83 0.83 0.83 6
Hans Blix 0.60 1.00 0.75 6
Hugo Chavez 0.75 1.00 0.86 12
Jacques Chirac 0.83 0.83 0.83 12
Jean Chretien 0.67 0.60 0.63 10
Jennifer Capriati 0.75 0.75 0.75 4
John Ashcroft 1.00 0.58 0.74 12
John Negroponte 0.00 0.00 0.00 4

221
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
 Junichiro Koizumi 1.00 0.88 0.94 17
Kofi Annan 0.88 1.00 0.93 7
Laura Bush 0.83 0.71 0.77 7
Lleyton Hewitt 0.75 0.60 0.67 10
Luiz Inacio Lula Da Silva 1.00 1.00 1.00 10
Megawati Sukarnoputri 1.00 0.86 0.92 7
Nestor Kirchner 0.88 0.70 0.78 10
Recep Tayyip Erdogan 1.00 0.38 0.55 8
Roh Moo-hyun 1.00 0.75 0.86 4
Sarena Williams 1.00 1.00 1.00 7
Silvio Berlusconi 0.50 0.50 0.50 4
Tom Ridge 0.83 1.00 0.91 5
Tony Blair 0.86 0.88 0.87 34
Vladimir Putin 0.62 0.62 0.62 8
Accuracy 0.85 474
Macro Avg 0.81 0.77 0.78 474
Weighted Avg 0.86 0.85 0.85 474

Figure 7. The Predicted and True Values from LFW Test Dataset

Figure 7 depicts the performance of the model on the LFW test dataset while it was trained for 50
epochs and 34 classes. The results demonstrate that the model performed well, accurately predicting
five images out of six.
Figure 8 depicts the training and validation loss for the model when implemented a greater number
of epochs and classes. The graph clearly demonstrates that when the number of epochs rises, the
model loss decreases, and therefore the accuracy increases.

222
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
 Figure 8. The Training and Validation Loss when No. of Epochs is 200

Figure 9. The Training and Validation Accuracy when No. of Epochs is 200

Figure 9 depicts the training and validation accuracy of the model, and the graph clearly illustrates
that as the number of epochs rises, so does the model's accuracy
Table 5 has 62 classes and begins with a column of names of several renowned individuals, followed
by four columns labelled "precision," "recall," "F1-score," and "support." Each name has a matching
value in these four columns, suggesting a quantitative measurement or evaluation.

Table 5. 62 Number of Classes with the Table Consists of a Column

of Names of Various Well-Known Personalities
Personalities Name Precision Recall F1-Score Support
Alvaro Uribe 0.60 1.00 0.75 6
Amelie Mauresmo 0.71 0.83 0.77 6
Andre Agassi 0.67 1.00 0.80 6
Angelina Jolie 0.60 0.50 0.55 6
Ariel Sharon 1.00 0.86 0.92 14
Alejandro Toledo 0.71 0.83 0.77 12
Arnold Schwarzenegger 1.00 0.57 0.73 14
Atal Bihari Vajpayee 1.00 0.67 0.80 6

223
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
Bill Clinton 1.00 1.00 1.00 5
Carlos Menem 0.75 0.75 0.75 4
Colin Powel 0.89 0.95 0.92 59
David Beckham 0.40 1.00 0.57 2
Donald Rumsfeld 0.91 0.97 0.94 30
George Robertson 1.00 1.00 1.00 4
George W Bush 0.87 0.99 0.93 97
Gerhard Schroeder 0.86 0.80 0.83 15
Gloria Macapagal Arroyo 0.92 0.80 0.86 15
Gray Davis 1.00 0.60 0.75 5
Guillermo Coria 1.00 0.91 0.95 11
Hamid Karzai 1.00 0.75 0.86 4
Hans Blix 0.83 1.00 0.91 5
Hugo Chavez 0.88 0.83 0.86 18
Igor Ivanov 1.00 0.25 0.40 4
Jack Straw 0.33 0.33 0.33 3
Jacques Chirac 1.00 0.92 0.96 13
Jean Chretien 0.75 1.00 0.86 6
Jennifer Aniston 0.80 0.80 0.80 5
Jennifer Capriati 0.64 0.88 0.74 8
Jennifer Lopez 0.33 0.50 0.40 2
Jeremy Greenstock 1.00 0.56 0.71 9
Jiang Zemin 1.00 1.00 1.00 2
John Ashcroft 1.00 0.80 0.89 10
John Negroponte 0.75 0.50 0.60 6
Jose Maria Aznar 0.50 1.00 0.67 1
Kuan Carlos Ferrero 0.86 0.86 0.86 7
Junichiro Koizumi 0.80 0.89 0.84 9
Kofi Annan 0.83 1.00 0.91 5
Laura Bush 0.89 1.00 0.94 8
Lindsay Davenport 1.00 0.60 0.75 5
Lleyton Hewitt 0.71 0.83 0.77 6
Luiz Inacio Lula Da Silva 1.00 1.00 1.00 6
Mahmoud Abbas 0.75 0.75 0.75 4
Megawati Sukarnoputri 0.83 0.83 0.83 6
Michael Bloomberg 0.50 1.00 0.67 1
Naomi Watts 1.00 0.80 0.89 5
Nestor Kirchner 0.73 1.00 0.84 8
Paul Bremer 0.67 0.67 0.67 6
Pete Sampras 1.00 0.50 0.67 6
Recep Tayyip Erdogan 1.00 0.67 0.80 6
Ricardo Lagos 0.60 1.00 0.75 3
Roh Moo-hyun 1.00 1.00 1.00 7
Rudolph Giuliani 0.67 0.50 0.57 4
Saddam Hussein 0.80 0.80 0.80 5
Serena Williams 0.92 0.92 0.92 12
Silvio Berlusconi 1.00 0.71 0.83 7
Tiger Woods 1.00 1.00 1.00 3
Tom Daschle 1.00 0.20 0.33 5
Tom Ridge 1.00 0.60 0.75 5
Tony Blair 0.74 0.91 0.82 22
Vicente Fox 0.88 0.58 0.70 12
Vladimir Putin 0.75 0.69 0.72 13
Winona Ryder 1.00 0.67 0.80 6
Accuracy 0.85 605
Macro Avg 0.83 0.79 0.79 605
Weighted Avg 0.87 0.85 0.84 605

224
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
 Figure 10. Some Predicted and True Values for Our Model

Figure 10 depicts the results of the model on the LFW test dataset while it was trained for 200 epochs
and 62 classes. The results reveal that the model performed well, accurately predicting 5 of 6
pictures. It is also evident that as the number of epochs increases, the model's performance
improves, however as the number of classes increases, the model's performance decreases.

CONCLUSION
This study created a facial recognition system using the Support Vector Machine (SVM) technique,
the suggested system employs a unique feature extraction approach based on the Histogram of
Oriented Gradients (HOG) in conjunction with a support vector machine (SVM) classifier. The
proposed system's feature extraction component was tested on the publicly accessible LFW
datasets. On the LFW databases, the system achieved 97.6% and 97.8% accuracy, respectively.
The system is also computationally efficient, requiring just a few moments to analyze a single face
for feature extraction and face identification. The Support Vector Machine (SVM) technique was used
to construct a face recognition system that met the goal of successfully recognizing faces in a given
facial dataset. The SVM algorithm was able to detect the face characteristics in the dataset with high
accuracy and categorize facial photos into the right category with high accuracy. The SVM method
also generalized well to previously unseen data, making it a dependable and robust face recognition
system. The findings of this study illustrate the efficacy of employing the SVM algorithm for
constructing face recognition systems and give a foundation for future research into enhancing facial
recognition system accuracy. The study resulted that the SVM algorithm is an excellent tool for face
225
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
recognition. More complex feature extraction approaches and larger datasets for training and testing
can help to enhance the system even more. Consequently, it has demonstrated that the SVM
algorithm can be used to create a face recognition system with high accuracy and robustness. To
attain better outcomes, the system should be upgraded and optimized further.

RECOMMENDATIONS
To develop a face recognition system using the Support Vector Machine (SVM) algorithm, start by
understanding the principles of both face recognition and SVM. Select a robust dataset such as
Labeled Faces in the Wild (LFW) or create a custom dataset with diverse face images. Preprocess
the data with steps like grayscale conversion, normalization, and face alignment to enhance the
quality and consistency of the input images. Employ feature extraction techniques like Principal
Component Analysis (PCA), Local Binary Patterns (LBP), or Histogram of Oriented Gradients
(HOG), or use deep learning-based feature extractors like Convolutional Neural Networks (CNNs)
for more effective feature representation. Experiment with various SVM kernels, such as linear,
polynomial, and radial basis function (RBF), to identify the most suitable one for your data. Optimize
the SVM hyperparameters through grid search or randomized search methods to fine-tune the
model’s performance. Implement cross-validation to ensure the model’s generalizability and to
prevent overfitting. Evaluate the model using performance metrics like accuracy, precision, recall,
F1-score, and ROC-AUC to measure its effectiveness. Enhance the system’s performance with data
augmentation, regularization techniques, and ensemble methods if necessary. Finally, deploy the
face recognition system with a scalable infrastructure, ensuring it can handle real-time processing,
and continuously monitor and update the model with new data to maintain high accuracy and
reliability. To improve the model's efficiency, modify the SVM hyperparameters using grid or
randomized search techniques. Use cross-validation to confirm the model's generalizability and
avoid overfitting. To evaluate the model's success, use performance indicators like as accuracy,
precision, recall, F1-score, and ROC-AUC. If required, improve the system's performance by using
data augmentation, regularization techniques, as well as ensemble approaches. Furthermore, set
the system for facial recognition on a scalable framework that can handle real-time processing, and
consistently track and refresh the model with latest information to ensure optimal reliability and
precision.

CONFLICT OF INTERESTS
I declare no conflicts of interest in researching and developing a face recognition system using the
Support Vector Machine method. The entire project was carried out independently, with no funding
or personal biases.

REFRENCES
[1] M. Furqan, A. Embong, S. Awang, S. W. Purnami, S. Sembiring, and U. M. Pahang, “smooth
support vector machine for face recognition using principal component analysis”.
[2] M. Z. Al-Dabagh, M. H. Mohammed Alhabib, and F. H. AL-Mukhtar, “Face Recognition
System Based on Kernel Discriminant Analysis, K-Nearest Neighbor and Support Vector Machine,”
International Journal of Research and Engineering, vol. 5, no. 2, pp. 335–338, Mar. 2018. DOI:
10.21276/ijre.2018.5.3.3
[3] Z. Rustam and A. A. Ruvita, “Application Support Vector Machine on Face Recognition for
Gender Classification,” in Journal of Physics: Conference Series, Institute of Physics Publishing,
Dec. 2018. DOI: 10.1088/1742-6596/1108/1/012067
[4] Dr. F. AL-Mukhtar and M. Z. N. AL-Dabagh, “Real-Time Face Recognition System Using
KPCA, LBP and Support Vector Machine,” International Journal of Advanced Engineering Research
and Science, vol. 4, no. 2, pp. 184–189, 2017. DOI: 10.22161/ijaers.4.2.36

226
European Journal of Applied Science, Engineering and Technology
www.ejaset.com
[5] W.-H. Lin, P. Wang, and C.-F. Tsai, “Face Recognition Using Support Vector Model Classifier
for User Authentication,” 2016. [Online]. Available:
https://www.sciencedirect.com/science/article/pii/S1567422316300011
[6] Z. Jun, “The Development and Application of Support Vector Machine,” in Journal of Physics:
Conference Series, IOP Publishing Ltd, Jan. 2021. DOI: 10.1088/1742-6596/1748/5/052006
[7] P. M. Shah, “Face Detection from Images Using Support Vector Machine,” San Jose State
University, San Jose, CA, USA, 2012. DOI: 10.31979/etd.wg5s-gyqn
[8] S. Akbar, A. Ahmad, M. Hayat, and F. Ali, “Face Recognition Using Hybrid Feature Space in
Conjunction with Support Vector Machine,” J. Appl. Environ. Biol. Sci, vol. 5, no. 7, pp. 28–36, 2015.
[9] st Isnawati Muslihah, sd Muqorobin, I. Teknologi Bisnis AAS Indonesia Surakarta, and J.
Slamet Riyadi No, “Texture Characteristic of Local Binary Pattern on Face Recognition with
Probabilistic Linear Discriminant Analysis,” 2020. [Online]. Available:
https://ijcis.net/index.php/ijcis/index
[10] P. Dhankhar and N. Sahu, “A Novel Algorithm for Human Facial Expression Recognition
using Gabor Filter and Eigenvector,” 2014.
[11] J. Matai, A. Irturk, and R. Kastner, “Design and Implementation of an FPGA-based Real-Time
Face Recognition System.”
[12] D. Zhang, S. Chen, and Z.-H. Zhou, “A New Face Recognition Method based on SVD
Perturbation for Single Example Image per Person.”
[13] L. Zhang, M. Yang, X. Feng, Y. Ma, and D. Zhang, “Collaborative Representation based
Classification for Face Recognition.”
[14] J. C. Moreano and N. B. L. S. Palomino, “Global facial recognition using gabor wavelet,
support vector machines and 3d face models,” Journal of Advances in Information Technology, vol.
11, no. 3, pp. 143–148, Aug. 2020. DOI: 10.12720/jait.11.3.143-148
[15] I. Q. Mundial, M. S. Ul Hassan, M. I. Tiwana, W. S. Qureshi, and E. Alanazi, “Towards facial
recognition problem in COVID-19 pandemic,” in 2020 4th International Conference on Electrical,
Telecommunication and Computer Engineering, ELTICOM 2020 - Proceedings, Institute of Electrical
and Electronics Engineers Inc., Sep. 2020, pp. 210–214. DOI:
10.1109/ELTICOM50775.2020.9230504
[16] M. J. Al Dujaili, H. T. S. Al Rikabi, N. K. Abed, and I. R. N. Al Rubeei, “Gender Recognition
of Human from Face Images Using Multi-Class Support Vector Machine (SVM) Classifiers,”
International Journal of Interactive Mobile Technologies, vol. 17, no. 8, pp. 113–134, 2023. DOI:
10.3991/ijim.v17i08.39163
[17] S. Saeed et al., “Empirical Evaluation of SVM for Facial Expression Recognition,” 2018.
[Online]. Available: www.ijacsa.thesai.org
[18] J. Augusto et al., “Facial recognition techniques using SVM : A comparative analysis
Técnicas de reconocimiento facial usando SVM : un,” pp. 1–14, 2019.

227
European Journal of Applied Science, Engineering and Technology
www.ejaset.com