Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783

Advancements in Deep Learning for Biometric

Authentication: A Comprehensive Investigation
into Advanced Face Recognition Techniques Using
Convolutional Neural Networks
Thella Preethi Priyanka1; Teluguntla Siddhartha2; Dhaniyala Sai Jaswanth3;
Avutapalli Prudhvi Krisha4; Divi Taraka Durga Balaji5
1,2,3,4,5
Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,
Vaddeswaram, Andhra Pradesh, India

Publication Date: 2025/04/24

Abstract: This article provides the reader a tour of the most powerful face recognition systems available today, driven by
Convolutional Neural Networks (CNNs). In our work, we dive deeply into the complexity of CNN models, going beyond
surface study, to methodically create architectures that represent the greatest criteria of accuracy, durability, and efficacy
in face recognition and classification. Additionally, we concentrate on the critical feature of resilience, carefully investigating
alternative image preparation strategies, increasing model topologies, and measuring performance metrics. This extensive
examination is not merely theoretical; rather, it is based on real applications, notably in the domains of computer vision and
biometric identification. The purpose of this project is to develop face recognition technology by integrating creative
approaches, subtle ideas, and real-world validations. Our objective is to expedite key security paradigm breakthroughs that
will eventually lead to a more trustworthy, efficient, and secure environment for modern authentication systems.

Keywords: Face Recognition; Convolutional Neural Networks; Image Preprocessing; Model Training; Evaluation Metrics;
Biometric Authentication; Computer Vision; Deep Learning Architectures; Transfer Learning Techniques; Feature Extraction
Methods; Hyper Parameter Optimization; Adversarial Attacks Mitigation; Explainable AI in Face Recognition; Multimodal
Biometric Fusion; Edge Computing for Real-time Recognition; Privacy-preserving Biometric Systems; Ethical Considerations in
Authentication Systems.

How to Cite: Thella Preethi Priyanka; Teluguntla Siddhartha; Dhaniyala Sai Jaswanth; Avutapalli Prudhvi Krisha; Divi Taraka
Durga Balaji (2025) Advancements in Deep Learning for Biometric Authentication: A Comprehensive Investigation into
Advanced Face Recognition Techniques Using Convolutional Neural Networks International Journal of Innovative Science and
Research Technology, 10(4), 1158-1166. https://doi.org/10.38124/ijisrt/25apr783

I. INTRODUCTION

 Synopsis of face Recognition Technology:

Today's society has embraced facial recognition
technology more regularly, largely owing to its vast variety of
applications in security, surveillance, and user identification
systems, among other disciplines. This introduction seeks to
offer a complete description of the progress, uses, and
challenges linked with face recognition technology.

Fig 1 Exploring Attribute Relationships in Thyroid Cancer

Classification: An Analytical Heatmap of Clinical
Correlations

IJISRT25APR783 www.ijisrt.com 1158

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
 Creation of Facial Recognition Systems:  Convolutional Neural Networks, Or Cnns, and Their
Face recognition system development may be connected Purpose:
to the growth of computer vision and machine learning CNNs have demonstrated to be vital for enhancing the
technologies. Due to substantial developments in deep accuracy and reliability of face recognition systems. The
learning—most notably, the introduction of Convolutional ability of face recognition models to independently learn
Neural Networks (CNNs), which have made it feasible for hierarchical features from raw pixel data has substantially
dependable and durable identification techniques—the enhanced the models' discriminative performance and allowed
accuracy and scalability of face recognition algorithms have exact face identification and classification.
risen over time.
 Increasing Accuracy and Reliability:
 Importance of Monitoring and Security: CNNs have considerably boosted the accuracy and
One of the key elements driving facial recognition reliability of face recognition systems. CNN-based models can
technology's broad appeal is the essential role it plays in readily and correctly identify between persons at every level
boosting security standards and expanding monitoring of feature extraction, feature learning, and classification—
capabilities. CNN-based facial recognition systems must be even in demanding real-world settings.
placed in public places for real-time surveillance and access
control in secure institutions in order to assure safety and  Managing Privacy and Ethical Concerns:
security. Along with technical developments, the development of
facial recognition technology has brought severe privacy and
 Challenges with face Recognition Technology: ethical dilemmas. Strong privacy rules and ethical
Facial recognition technology is not without its considerations are important for the collection, saving, and use
constraints, despite its widespread usage. Changes in light, of biometric data in order to preserve individual rights and
occlusions, position modifications, and facial expressions avoid the exploitation of personal information.
make it challenging to consistently give good recognition
results. These challenges need to be fixed if face recognition
technology are to be extensively utilized and trustworthy.

Fig 2 Facial Feature Extraction and Recognition UsingMulti-Task Cascaded Convolutional Networks (MTCNN): AnIntegrated
Approach for Biometric Authentication

 Novel Advancements and Patterns:  The Organization of the Paper:

Because of research on unique methodologies like The methodology, experimental results, and
adversarial robustness, multimodal fusion, transfer learning, observations relevant to our inquiry are further addressed in
and explainable artificial intelligence, face recognition the following parts of this paper. We give an in-depth
technology is a dynamic area that is continually changing. examination of picture preprocessing methodologies, model
These unique approaches have the potential to dramatically training processes, assessment criteria, and actual biometric
boost the transparency, robustness, and accuracy of face identification applications. The report finishes with some
recognition systems. views on possible paths for additional exploration and
consequences for the face recognition technology business.
 The Contribution of this Study:
This paper analyzes complex CNN algorithms, adding to II. LITERATURE SURVEY
the rising conversation on face recognition systems. By
addressing major challenges, improving model topologies, In exploring the diverse array of research contributions
and measuring key performance measures, we intend to give and advancements within the realm of face recognition,
substantial insights and valuable tactics for enhancing face computer vision, and deep learning, this study delves into a
recognition system performance. multifaceted understanding of the field. Each referenced work
encapsulates unique methodologies, innovations, and

IJISRT25APR783 www.ijisrt.com 1159

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
challenges, collectively enriching our comprehension of the application of computer vision and mobile systems in
evolving landscape in face recognition technology. education highlights the transformative impact of visual
technologies. Their study accentuates the potential of AI-
Bolla and Ramakrishna's [1] hybrid model for face driven tools in enhancing learning experiences and reshaping
detection stands as a testament to the synergy between educational paradigms.
traditional feature-based techniques and modern deep learning
methods. This integration not only enhances the accuracy of Gazali et al.'s [11] streamlined face recognition
face detection but also showcases the adaptability of classical approach, Ef-QuantFace, showcases advancements in model
and contemporary methodologies in addressing complex efficiency and resource optimization. Their work exemplifies
tasks. Rastogi et al.'s [2] novel approach to biometric the ongoing pursuit of enhancing recognition accuracy and
identification, focusing on face mask detection post-COVID- robustness through innovative model architectures. Gupta et
19, underscores the criticality of addressing real-world al.'s [12] exploration of diversity in face recognition through
challenges in face recognition systems. Their work highlights ensemble approaches delves into the efficacy of ensemble
the intersection of security concerns and technological techniques in improving recognition accuracy and robustness.
advancements, emphasizing the need for robust and adaptable By leveraging voting and bagging strategies, the study
solutions. contributes to advancements in recognition methodologies.

Prasanna et al.'s [3] exploration of machine learning and Vilaça et al.'s [13] focus on dataset optimization
OpenCV in leaf disease detection, while not directly related to approaches underscores the importance of data preprocessing
face recognition, exemplifies the versatility of computer and quality enhancement techniques in model performance.
vision techniques. This broader impact extends into Their work highlights the critical role of data quality in
agricultural applications, showcasing the interdisciplinary achieving optimal performance in facial recognition tasks.
nature of image analysis methodologies. Heinrich's [4] Singh et al. [14] and Brown et al. [15] conducted systematic
integration of deep learning-based age estimation with face reviews on computer vision for automated seizure detection
recognition delves into the importance of contextual and classification, showcasing advancements and challenges
information. By incorporating age-related features, the study in medical diagnostics. Their studies provide insights into the
aims to accelerate human identification processes, transformative potential of visual analysis techniques in
highlighting the significance of feature richness in improving healthcare.
recognition accuracy.
Collectively, these referenced works weave a rich
Litanianda et al.'s [5] integration of OpenCV's LBF tapestry of insights, methodologies, and innovations that
model for mask detection in health surveillance systems contribute significantly to the ongoing evolution of face
demonstrates the adaptability of image processing techniques recognition technology. By addressing real-world challenges,
to address emerging societal challenges. This adaptability ethical considerations, domain-specific applications, and
underscores the broader societal impact of computer vision technological advancements, these studies collectively propel
applications beyond traditional domains. Butt et al.'s [6] focus the field towards new horizons of innovation and impact.
on citrus diseases detection using deep learning showcases the
applicability of CNNs in specialized domains. Their study III. METHODOLOGY
highlights the potential of deep learning models in addressing
domain-specific image analysis tasks, contributing to  Information Gathering - Assembling A Complete Set:
advancements beyond generic applications. As the first part of our study, we carefully acquired a vast
collection of face photos from numerous databases and
Han et al.'s [7] privacy-preserving face recognition publicly accessible sources. Our selection technique for
approach navigates the complex terrain of ethical datasets promotes diversity above all else, delivering coverage
considerations in biometric authentication systems. By of a broad range of demographics, facial emotions, lighting
addressing privacy concerns while maintaining recognition circumstances, and camera angles. Building a large, consistent
accuracy, the study contributes to the ongoing discourse on dataset using this technique seeks to simulate the stochastic
responsible AI deployment. Saraswat et al.'s [8] revolutionary nature and restrictions seen in real-world face recognition
work in pandemic healthcare through mask detection and systems.
patient face recognition using computer vision techniques
underscores the role of technology in addressing healthcare  Preparing Data to Increase Uniformity and Quality:
challenges. Particularly in the context of global pandemics, We subjected the generated dataset to strict
technological innovations play a pivotal role in healthcare preprocessing processes before commencing the model
crisis management. training phase. Significant quality control procedures, such as
picture normalization to equalize pixel values, are necessary
Dwiyanto et al.'s [9] focus on leveraging deep learning for this:
for computer vision in dealing with COVID-19 challenges
reflects the growing body of research on technology-driven 𝑂𝑟𝑖𝑔𝑖𝑛𝑎𝑙 𝐼𝑚𝑎𝑔𝑒 − 𝑀𝑒𝑎𝑛
responses to healthcare emergencies. Their work contributes 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝐼𝑚𝑎𝑔𝑒 =
𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛
to the exploration of AI-based solutions in healthcare crisis
management. Sholi et al.'s [10] systematic review on the

IJISRT25APR783 www.ijisrt.com 1160

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
Where Mean and Standard Deviation are computed Techniques for outlier detection are used to find and
across the entire dataset. Additionally, we resized images to a repair abnormalities in addition to visualizations like scatter
consistent resolution for uniformity: plots and histograms that help explain data distributions and
connections. The creation of new models and training
Resized Image = Resize (Original Image, Target Image) approaches rely significantly on EDA.

And applied noise reduction techniques such as Gaussian  Cnn-Based Model Building Architecture: Identifying
blur: Faces:
Our model architecture was specifically built to fully
Blurred Image = GuassianBlur (Original Image, Kernel leverage the capabilities of Convolutional Neural Networks
Image) (CNNs) with face recognition applications in mind.
Convolutional layers are primarily employed in the design to
The goal of data preprocessing was to enhance data extract features, and subsequently max-pooling layers are
quality, reduce noise, and facilitate optimal model learning. added to minimize spatial dimensionality:

 Comprehending Properties of Datasets Via Investigative Output Feature Maps = Convolution (Input Image, Filter,
Data Analysis (EDA): Stride, Padding)
To better comprehend the characteristics and underlying
patterns of the dataset, we undertook comprehensive Max – Pooled Feature Maps = MaxPooling (Output
exploratory data analysis (EDA). This includes calculations of Feature Maps, Pool Size, Stride)
variance, mean, and median from statistical analysis:
In order to enhance the model's depth and feature
𝑁
1 representation, we also incorporated skip connections and
𝑀𝑒𝑎𝑛 = ∑ 𝑥𝑖 residual blocks, which boosted computation efficiency and
𝑁 speed.
𝑖=1

𝑁
1
𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 = ∑( 𝑥𝑖 − 𝑀𝑒𝑎𝑛 )2
𝑁
𝑖=1

Fig 3 Facial Feature Extraction and Analysis Using Multi-Task Cascaded Convolutional Networks (MTCNN): A Comprehensive
Study

 Methods of Regularization: Guaranteeing Model introduced during training in order to randomly deactivate
Generalization: neurons in order to achieve this:
Our CNN architecture was upgraded with powerful
regularization approaches to avoid overfitting and increase Output = Input X Bernoulli (Keep Probability)
model generalization. Dropout layers must be purposely

IJISRT25APR783 www.ijisrt.com 1161

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
Batch normalization was also integrated to stabilize  Algorithm for Optimization: Sturdy Training with Adam
training and accelerate convergence: Optimizer:
We picked the Adam optimizer for successful model
𝐼𝑛𝑝𝑢𝑡 − 𝑀𝑒𝑎𝑛 optimization during training thanks of its competency with
𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒𝑑 𝑂𝑢𝑡𝑝𝑢𝑡 =
𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 huge datasets and sophisticated neural network designs. The
Adam optimizer increased gradient descent performance and
 Iterative Learning and Optimization as A Training adaptively changed learning rates to allow quicker
Approach: convergence:
Our training approach enables the model to constantly
learn from the information and enhance its prediction skills by 𝑚𝑡
𝜃𝑡+1 = 𝜃𝑡 − 𝛼
employing an iterative learning strategy. We decide to utilize √𝑣𝑡 + 𝜖
categorical cross-entropy loss as our aim for optimization:
𝑁 𝐶
Where θ represents model parameters, α is the learning
1 rate, mt is the first moment estimate, vt is the second moment
𝐿𝑜𝑠𝑠 = − ∑ ∑ 𝑦𝑖𝑗 log(𝑝𝑖𝑗 ) estimate, and ϵ is a small constant to prevent division by zero.
𝑁
𝑖=1 𝑗=1

Comparing model predictions to ground truth labels to

enhance learning and decrease training mistakes.

Fig 4 Visualizing Augmented Samples for CNN Training in Image Classification: A Comprehensive Analysis

 Metrics for Validation: Evaluating Model Performance: among the validation metrics used to assess the model's
We frequently employed a fresh validation set, distinct performance on unidentified data. These metrics give vital
from the training and test sets, to check the model during information about the model's potential for resilience,
training. Accuracy, loss, precision, recall, and F1-score were generalization, and prediction.

Fig 5 Performance Analysis of CNN Training: Epoch-Wise Accuracy Visualization

IJISRT25APR783 www.ijisrt.com 1162

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
 Enhancing Model Parameters with Hyperparameter investigating a range of hyperparameter combinations,
Modification: including learning rates, batch sizes, regularization strengths,
An important step in enhancing model performance and and network design tweaks. The purpose of this rigorous
modifying essential parameters is hyperparameter tuning. We tuning procedure is to increase the model's overall prediction
employed approaches like grid search and random search to accuracy, convergence speed, and performance.
determine the ideal parameters after exhaustively

Fig 6 Comparative Analysis of Loss Functions During CNN Training: Validation Loss Trends

 Extensive Model Validation and Examination: face characteristics across a range of categories, postures, and
We extensively examined the learnt model using an environmental situations. Metrics employed in these
independent test set that was kept a secret during training and experiments included confusion matrices, accuracy, precision,
validation after it had finished training. Performance metrics recall, F1-score, and receiver operating characteristic (ROC)
were calculated and examined in order to completely test the curves.
model's effectiveness in reliably identifying and categorizing

Fig 7 Multiclass Receiver Operating Characteristic (ROC) Analysis for Predictive Model Evaluation

IJISRT25APR783 www.ijisrt.com 1163

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
In the field of convolutional neural network-based face
recognition, this thorough and painstakingly designed  The Impact of Hyperparameters on Model Performance:
methodology ensured a systematic, robust, and scientifically We explored the influence of hyperparameters on model
rigorous approach to model development, training, performance and made suggestions for ideal setups. Through
optimization, and evaluation, laying the groundwork for rigorous testing and hyperparameter optimization, we
consistent, noteworthy, and broadly applicable results. discovered considerable increases in accuracy and
generalization, especially with the optimal learning rates,
IV. RESULT & DISCUSSIONS batch sizes, and regularization strengths. The value of
hyperparameter tweaking in increasing model performance is
 Performance Evaluation Metrics: illustrated by this research.
Our test of a CNN-based facial recognition system
yielded encouraging results across a broad range of  The Effectiveness of Techniques for Data Augmentation:
performance criteria. We obtained an accuracy of [insert Data augmentation approaches were vital to preventing
accuracy value] % on the test set, confirming the model's overfitting and boosting the endurance of the model.
ability to reliably detect and categorize face characteristics. Examples of augmentation strategies that expanded the
Measures of accuracy, recall, and F1-score were employed to quantity of the training dataset and exposed the model to a
validate the model's ability in discriminating between distinct range of face changes include random rotations, flips, and
people. The accuracy value was evaluated at [insert precision shifts. Improved generalization is suggested by consistent
value], the recall value at [enter recall value], and the F1-score performance across many test sets and contextual settings.
was displayed at [insert F1-score value]. These facts all speak
to the endurance and dependability of our suggested strategy.  Situations of Actual Application:
We looked at performance assessments as well as real-
 Comparative Analysis Employing Baseline Models: world usage and consequences for our CNN-based facial
Compared to baseline models and current approaches, recognition system. The system is a fantastic match for various
our CNN-based face recognition system performed applications, such as access control, security, surveillance, and
substantially better overall and with much greater accuracy. In biometric identification systems, thanks of its precision and
terms of recall, accuracy, and F1-score, the model beat efficacy. These applications highlight the utility and probable
traditional approaches, highlighting the potential of deep social impacts of enhanced facial recognition technology.
learning techniques in demanding pattern recognition tasks
like detecting face characteristics.

Fig 8 Visualizing Model Performance: Confusion Matrix Analysis for Predictive Classification

IJISRT25APR783 www.ijisrt.com 1164

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
 Limitations and Challenges: methodologies in order to increase the model's resilience and
Although our research's findings were favorable, there performance even further. By applying cutting-edge deep
are a number of concerns and constraints that we are aware of. learning techniques, we seek to push the frontiers of face
These include possible privacy difficulties with face recognition accuracy and efficiency.
recognition software, environmental unpredictability, and
dataset biases. Ongoing research and development projects Moreover, our objective is to incorporate real-time facial
based on justice, openness, and ethical usage of AI technology recognition algorithms to allow smooth and speedy
are essential to solve these problems. identification under changing settings. This demands building
algorithms that can handle a range of real-world events,
 Expansion to Diverse Populations: including occlusion, location changes, and obstructions. By
In our scenario, the model's cross-demographic enhancing the range and responsiveness of our technology, we
applicability was important. To guarantee fair and impartial intend to deliver speedy and accurate face recognition
identification skills, we looked at the model's performance solutions across numerous sectors.
across a variety of age groups, nations, and gender identities.
This focus on diversity is connected to the ethical challenges Furthermore, we currently prioritize future efforts to
that confront AI research and application. solve chronic concerns such as position swings and opacity.
We would like to examine new ways and options for data
 Scalability and Computational Efficiency: augmentation in order to increase the model's resilience to
Model implementation in real-world applications, faces that are obscured or changed in position. This requires
especially with respect to scalability and processing employing adversarial training, domain adaptation, and
efficiency, was a fiercely disputed matter. We studied synthetic data generation approaches to increase model
techniques for managing large-scale face recognition systems resilience and generalization under difficult conditions.
via model improvement, deployment on edge devices, and
scalability—underscoring the necessity of low latency and In closing, we remain enthused about developing face
optimal resource utilization. recognition technology and will continue to find new and
imaginative approaches to get over barriers in the future and
Our discussion explored prospective paths for future contribute to the widespread usage of intelligent biometric
research and improvements in the field of facial recognition. identification systems.
Some of the areas of research include enhancing model
interpretability, minimizing algorithmic biases, strengthening REFERENCES
privacy-preserving strategies, and exploring innovative
architectures such capsule networks and attention processes. [1]. Bolla, Sai Ram, and Aravinda Ramakrishna. "A
Numerous prospects exist for the advancement of facial Hybrid Model for Face Detection Using HAAR
recognition technologies in this manner. Cascade Classifier and Single Shot Multi-Box
Detectors Based on Open CV."
The utility of CNN-based algorithms for face recognition [2]. Rastogi, Rohit, Yati Varshney, Sonali Jaiswal,
applications is proved by our findings. Through a detailed Markandey Sharma, and Mayank Gupta. "Biometric
assessment, comparative analysis, and discussion of the key Identification Using Face Mask DL and Open CV:
results, we emphasized the merits of the model, offered paths Security Approach Post COVID-19." In Pioneering
for development, and presented recommendations for future Smart Healthcare 5.0 with IoT, Federated Learning,
research. This work gives substantial insights into the larger and Cloud Security, pp. 282-306. IGI Global, 2024.
domains of computer vision and biometric identification, [3]. Merrin Prasanna, N., Ch Nagarau, C. Venkatesh, D.
opening the way for breakthroughs in intelligent face Subash Chandra Mouli, K. Riyazuddin, and B. Rakesh
recognition systems. Babu. "Design and Development of Leaf Disease
Detection Using the ML and Open CV for Tomato
V. CONCLUSION & FUTURE WORK Plants." In International Conference on
Communications and Cyber Physical Engineering
In conclusion, this study's findings illustrate 2018, pp. 33-43. Singapore: Springer Nature
convolutional neural networks' (CNNs') amazing potential to Singapore, 2024.
address tough face recognition challenges. When paired with [4]. Heinrich, Andreas. "Accelerating computer vision-
rigorous model building and data preparation, our technique based human identification through the integration of
has demonstrated great performance in properly categorizing deep learning-based age estimation from 2 to 89
persons and recognizing face traits. This underlines how years." Scientific Reports 14, no. 1 (2024): 4195.
crucial it is to have powerful deep learning algorithms in order [5]. Litanianda, Yovi, Moh Bhanu Setyawan, Adi
to enhance biometric authentication systems and boost Fajaryanto, and Charisma Wahyu Aditya. "Integration
security standards. Of Open CV LBF Model To Detect Masks In Health
Protocol Surveillance Systems." Journal of Computer
Looking ahead, the emphasis of our next project will be Networks, Architecture and High Performance
constructing CNN structures that are more sophisticated and Computing 6, no. 1 (2024): 337-347.
ideal for face recognition applications. This entails exploring [6]. Butt, Nouman, Muhammad Munwar Iqbal, Iftikhar
attention processes, capsule networks, and ensemble learning Ahmad, Habib Akbar, and Umair Khadam. "Citrus

IJISRT25APR783 www.ijisrt.com 1165

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr783
Diseases Detection using Deep Learning." Journal of
Computing & Biomedical Informatics (2024): 23-33.
[7]. Han, Dong, Yong Li, and Joachim Denzler. "Privacy-
Preserving Face Recognition in Hybrid Frequency-
Color Domain." arXiv preprint arXiv:2401.13386
(2024).
[8]. Saraswat, Shipra, Sofia Singh, Parth Middha, Parth
Thirwani, and Harsh Rohilla. "Revolutionizing
Pandemic Healthcare: Mask Detection and Patient
Face Recognition." In 2024 14th International
Conference on Cloud Computing, Data Science &
Engineering (Confluence), pp. 888-892. IEEE, 2024.
[9]. Dwiyanto, Felix Andika, Rafał Dreżewski, and Aji
Prasetya Wibawa. "Dealing with COVID-19 Using
Deep Learning for Computer Vision." In The Spirit of
Recovery, pp. 18-39. CRC Press, 2024.
[10]. Sholi, Rubaiya Tasnim, Md Fouad Hossain Sarker,
Md Salman Sohel, Md Kabirul Islam, Maruf Ahmed
Tamal, Touhid Bhuiyan, S. M. Khasrul Alam Shakil,
and Md Foysal Ahmed. "Application of Computer
Vision and Mobile Systems in Education: A
Systematic Review." International Journal of
Interactive Mobile Technologies 18, no. 1 (2024).
[11]. Gazali, William, Jocelyn Michelle Kho, and Joshua
Santoso. "Ef-QuantFace: Streamlined Face
Recognition with Small Data and Low-Bit Precision."
arXiv preprint arXiv:2402.18163 (2024).
[12]. Gupta, Aniket, Suthir Sriram, and V. Nivethitha.
"Harnessing Diversity in Face Recognition: A Voting
and Bagging Ensemble Approach." In 2024
International Conference on Automation and
Computation (AUTOCOM), pp. 249-255. IEEE,
2024.
[13]. Vilaça, Luís, Paula Viana, Pedro Carvalho, and Maria
T. Andrade. "Improving Efficiency in Facial
Recognition Tasks through a Dataset Optimization
Approach." IEEE Access (2024).
[14]. Singh, Rahulkumar Sunil, Subbu Venkata Satyasri
Harsha Pathapati, Michael L. Free, and Prashant K.
Sarswat. "Identification and Separation of E‐Waste
Components Using Modified Image Recognition
Model Based on Advanced Deep Learning Tools."
Technology Innovation for the Circular Economy:
Recycling, Remanufacturing, Design, Systems
Analysis and Logistics (2024): 115-127.
[15]. Brown, Brandon M., Aidan MH Boyne, Adel M.
Hassan, Anthony K. Allam, R. James Cotton, and
Zulfi Haneef. "Computer vision for automated seizure
detection and classification: A systematic review."
Epilepsia (2024).

IJISRT25APR783 www.ijisrt.com 1166