Face Recognition with Convolutional Neural Networks for Enhanced

Security

S. Nagendrudu, A. Sai Pavan, B. Sreekanth Rao, P. Narendra, S. Khaja Moinuddin

and O. Surya Arjun

Computer science and engineering, Santhiram engineering College, NH-40, Nandyal - 518501, Andhra Pradesh, India

Keywords: Network of Convolutional Neurons, Vector Machine Support, Rate of Recognition and Training Duration.

Abstract: Convolutional neural networks (CNNs) have become an effective tool for face recognition in the domain of

access control, surveillance and identity verification. They are well suited for extracting and learning large

amounts of hierarchical facial features and performing accurate and robust recognition under various

condition which include variation in lighting conditions, pose and occlusion. This work presents the

implementation of a CNN-based face recognition system implementing deep learning architectures such as

FaceNet or VGG-Face for feature extraction and classification. Incorporating the technology into biometric

security frameworks, the proposed system improves authentication reliability, reduces risks of unauthorized

access and provides real-time monitoring capabilities. This work represents an overview of the applications

of CNNs in biometric security as well as a scalable and efficient approach to modern facial recognition

systems.

1 INTRODUCTION

Recognition of faces is a biometric authentication

method that uses facial features to identify or

authenticate a person. The method is widely used in

security, surveillance, user authentication and social

media applications. The problem in face recognition

is related to variations in lighting, pose, expression,

occlusion and aging.

The best-known face recognition machine

learning approach is convolutional neural networks

(CNNs), which automatically extract hierarchical

features from image data. CNNs beat other face

recognition methods in learning distinguishing

characteristics from unprocessed pixel data.

A crucial area of human-computer interaction,

face recognition has had a significant impact on the

creation of artificial intelligence. Since the concept

was proposed in 1960s, there are currently about five

methods to implement face recognition, such as

geometrical characteristic method, methods such as

neural networks, hidden Markov models, elastic

graph matching, and subspace analysis. Neural

network-based methods are categorized as deep

learning because they can extract more complex

features like corner point and plane features, while the

first four methods are generally categorized as

shallow learning because they can only exploit certain

basic features of images; the majority of them are

based on artificial experience extraction of sample

features.

As usual, CNN, Deep Belief Network (DBN), and

Stacked Denoising Autoencoder (SDAE) are the

neural networks that are mentioned while discussing

facial recognition techniques. In other words, CNN

may be used to directly input images and can

withstand image deformations such as rotation,

translation, and scaling. More significantly, CNN is

capable of automatically extracting useful facial

traits. When it comes to face recognition, CNN is by

far the best option. LeCun used multilayer CNN to

successfully handle 2D images in 1988. As computer

hardware advanced, Hinton and Krazhevsky

processed the ImageNet database in 2012 using deep

CNN, and the results were better than before.

In addition to face recognition, CNN is frequently

used for face verification, which has shown very good

results. Face recognition Sun has conducted research

and created the CNN-based DeepId method for face

verification. Up till 2015, they had developed three

DeepId versions. They have demonstrated that their

Nagendrudu, S., Pavan, A. S., Rao, B. S., Narendra, P., Moinuddin, S. K. and Arjun, O. S.

Face Recognition with Convolutional Neural Networks for Enhanced Security.

DOI: 10.5220/0013925500004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

219-223

ISBN: 978-989-758-777-1

219

DeepId techniques produce face verification results

that surpass those of human eyes. CNN can be used

as a feature extractor to extract useful features in

addition to being a classifier to address two- or multi-

class classification issues. In this work, we extract

facial features using CNN.

2 LITERATURE REVIEW

A summary of current research and outcomes relevant

to a particular subject. For facial recognition utilizing

Convolutional Neural Networks and Support Vector

Machines, this section will encompass important

studies, methodologies, and advancements. Below is

a fixed structure for a literature review on this topic:

• An Introduction to Facial Recognition

Definition and Significance: A summary of

facial recognition technology and how it is

utilized in fields like security, surveillance,

and personal identification. Challenges: A

discussion on challenges that facial

recognition technology faces, including

lighting, pose, expression, and occlusions.

• Traditional Methods of Facial Recognition

Eigenfaces & Fisher faces: Early methods

using LDA (linear discriminant analysis) or

PCA (principal component analysis). Local

Binary Patterns: Texture-based methods for

face recognition. Limitations: The limitations

of traditional facial recognition methods for

handling complex variations and the

fluctuating impacts of lighting on these

technologies.

• Deep Learning for Facial Recognition

Introduction to CNNs: A discussion on CNNs

and how they "extract" hierarchical features

from images. Key Studies: An examination of

landmark studies (e.g. Deep Face-Facebook,

FaceNet-Google, DeepID x2), which

contributed to modern facial recognition

accuracy, or the field of facial recognition in

general. Architectures: A discussion on

common CNN architectures in facial

recognition (i.e. VGG, Inception, ResNet).

Deep Learning: Generative Adversarial

Networks (GANs) and autoencoders have

shown promise in identifying complex

spatiotemporal correlations.

3 SYSTEM MODEL DESIGN

An arrangement of system model design wherein pre-

training and target training stages are included and

convolutional neural networks (CNNs) and support

vector machines (SVMs) are used. Here is the system

model design procedure illustrated from the Figure 1:

• Pre-training Phase: Pre-training Dataset:

This is the initial dataset used to train the

CNN. It helps the model learn general features

that can be useful for a variety of tasks.

• Pre-training CNN: The CNN is pre-trained

on the pre-training dataset to learn feature

representations; the problem here is to make

the weights of the network minimize the loss

function.

• Weight Transfer: After pre-training, the

learned weights of the CNN are transferred

into the next stage (these weights comprise the

feature representations learned in the pre-

training phase).

• Target Training Phase: Target Training

Dataset is the training dataset for the task in

question. It may be smaller or more

specialized than the pre - training dataset.

• Train CNN: After initializing the CNN with

pre-trained weights, the CNN is further trained

on the training data set corresponding to the

task to adjust the feature representations

according to the task.

Figure 1: The whole training system framework.

• Feature extraction: After the CNN has been

trained on the target data set its features are

taken out of the CNN and these are

representations of the data in the input.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

220

• Train SVM: The features learned from the

extracted images are then used to train a

Support Vector Machine (SVM), a classifier

that can be used to make predictions based on

the features learned from the C.

A face recognition process using convolutional

neural networks (CNNs) and support vector machines

(SVMs).

• Target Using Dataset: Dataset gathering:

gather a dataset of faces. The dataset should

include different kinds of faces in different

conditions (lighting, angle, expressions) so the

model can generalize well.

• Preprocessing (preparing the images for

normalization) usually just resizing and/or

grayscale conversion or normalization of pixel

values.

• CNN (Convolutional Neural Network):

Feature extraction CNN is used to extract

features from the face images. As you can see

in the diagram below there are multiple layers,

the convolutional, pooling and fully connected

layers, learning hierarchical representations of

the input images.

• The CNN training The CNN training on the

face dataset. During the training, the CNN

learns to identify features that are most

important for distinguishing different faces.

• Feature: Feature vector The CNN after

training outputs a feature vector for each input

face image: the feature vector is a high

dimensional representation of the face that

contains all the necessary characteristics.

• SVM (Support Vector Machine): Trained

SVM: The feature vectors that the CNN

extracts are used to train an SVM classifier

and the SVM learns to classify each of the

feature vectors into different classes (with

different people representing different

classes). Results can be then compared against

classification models to see which kind of

image has better feature vectors.

• Recognition: Face Recognition Whenever a

new face image is uploaded to the system the

CNN extracts its feature vector and the SVM

Classifies the image to recognize the person.

• Lack: How to cope with lack of data: If there

is not enough training data, data augmentation

(rotation, flipping, scaling) or transfer learning

(using a pre-trained CNN on an already trained

dataset such as ImageNet) can be used to

increase the performance of the model.

Figure 2: The testing framework.

3 METHODOLOGY

The above-mentioned techniques of face recognitions

are explained in the reference. Geometrical

characteristic method is one that extract geometric

features like distance between key features or points

of the face. Techniques such as PCA (Principal

Component Analysis), LDA, etc. are used for face

image dimension reduction and feature extraction

through subspace analysis method. This technique

uses a graph-based approach for matching facial

features and allows some deformation in the face

structure. The Hidden Markov Model (HMM)

method is used for modelling the spatial and temporal

variation in the face images. In the past, face

recognition was done using simple neural networks

like radial basis function (RBF) networks. Although

these methods could extract more complex features

but were limited due to the computational resources

at that time. techniques, such as Fourier transforms

and wavelet analysis, are used to identify anomalies

in temporal data.

4 PROPOSED METHODOLOGY

4.1 CNN Feature Extraction

• Pre-training: To learn general facial traits,

the CNN is pre-trained on a sizable auxiliary

dataset called Casia-Webfaces.

• Fine-tuning: To extract more precise facial

features, the pre-trained CNN is

subsequently adjusted on the target dataset

(FERET).

Face Recognition with Convolutional Neural Networks for Enhanced Security

221

• SVM classification: An SVM uses the

CNN's retrieved features as input for

classification. To categorize facial features

into distinct classes (individuals), the SVM

is trained.

4.2 Methods of Optimization

Rectified Linear Units (ReLU): A CNN's activation

function that boosts efficiency.

Weight Penalty: To avoid overfitting, L2

regularization is used.

Dropout: To enhance generalization, units are

randomly dropped during training.

Enhancement of Data: To improve the diversity of

training data, photos are flipped and added to the

collection.

5 RESULTS AND DISCUSSIONS

the performance of the proposed face recognition

system, which combines Convolutional Neural

Networks (CNNs) for feature extraction and Support

Vector Machines (SVMs) for classification. The

experiments were conducted on the FERET and ORL

datasets, and the results demonstrate the effectiveness

of the proposed approach.

5.1 Recognition Performance of CNN

• The CNN was tested on six different

configurations of the FERET dataset, with

recognition rates ranging from 87.29% to

99.66%. The highest recognition rate was

achieved with the "test samples135"

configuration, which included images under

varying lighting conditions and expressions.

• The recognition error rate was significantly

low, with most configurations converging to

an error rate of 0.03% after 50 epochs. This

indicates that the CNN is highly effective in

extracting discriminative features from face

images.

5.2 CNN and SVM Recognition

Performance:

• When the features extracted by the CNN

were used to train an SVM, the recognition

rates improved further. For example, the

recognition rate for "test samples123"

increased from 98.25% (CNN alone) to

98.63% (CNN + SVM).

• The combination of CNN and SVM

consistently outperformed the CNN alone

across all dataset configurations,

demonstrating the SVM's ability to better

classify the extracted features.

5.3 Training Time

• The proposed system significantly reduced

training time compared to traditional methods.

For instance, the CNN + SVM approach

achieved a recognition rate of 97.5% in just 28

seconds on the ORL dataset, whereas a

traditional method (Global + Local Expansion

ACNN) took 343 seconds to achieve a lower

recognition rate of 93.30%.

• The use of pre-training with the Casia-

Webfaces dataset contributed to faster

convergence and reduced training time,

highlighting the efficiency of the proposed

method.

5.4 Data Augmentation

Data augmentation techniques, such as flipping

images, were employed to increase the diversity of

the training dataset. This approach helped improve

the generalization ability of the model, leading to

better performance on the testing dataset

6 CONCLUSIONS

In conclusion, the suggested face recognition system

shows notable gains in recognition accuracy and

training efficiency by combining CNNs for feature

extraction and SVMs for classification. Compared to

conventional techniques, the system drastically cuts

down on training time while achieving high

recognition rates some configurations can reach up to

99.83% accuracy. The system is suitable for real-

world applications due to its resilience to changes in

lighting, pose, and expression. To improve

performance, future research might concentrate on

investigating bigger datasets and further refining the

CNN architecture. Overall, a potent method for face

recognition that strikes a balance between high

accuracy and computational efficiency is presented

by the combination of CNNs and SVMs.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

222

REFERENCES

Muhammad, Farooq Sumar, et al. "Key distribution scheme

for preventing key reinstallation attack in wireless

networks." AIP Conference Proceedings. Vol. 3028.

No. 1. AIP Publishing, 2024.

Suman, Jani Venkata, et al. "Leveraging natural language

processing in conversational AI agents to improve

healthcare security." Conversational Artificial

Intelligence (2024): 699-711.

Sumar, Mohammad Farooq, and V. Madhu Viswanathan.

"A fast approach to encrypt and decrypt of video

streams for secure channel transmission." World

Review of Science, Technology and Sustainable

Development 14.1 (2018): 11–28.

Muhammad, Farooq Sumar, Karthik Balasubramanian, and

T. Sudhakar Babu. "A comprehensive research on video

imaging techniques." All Open Access, Bronze (2019).

Muhammad, Farooq Sumar, and V. Madhu Viswanathan.

"Performance analysis of data compression algorithms

for heterogeneous architecture through parallel

approach." The Journal of Supercomputing 76.4

(2020): 2275–2288.

Devi, M. Sharmila, et al. "Extracting and Analyzing

Features in Natural Language Processing for Deep

Learning with English Language." Journal of Research

Publication and Reviews 4.4 (2023): 497–502.

Devi, M. Sharmila, et al. "Machine Learning Based

Classification and Clustering Analysis of Efficiency of

Exercise Against Covid-19 Infection." Journal of

Algebraic Statistics 13.3 (2022): 112–117.

Mandalapu, Sharmila Devi, et al. "Rainfall prediction using

machine learning." AIP Conference Proceedings. Vol.

3028. No. 1. AIP Publishing, 2024.

Chaitanya, V. Lakshmi, et al. "Identification of traffic sign

boards and voice assistance system for driving." AIP

Conference Proceedings. Vol. 3028. No. 1. AIP

Publishing, 2024.

Chaitanya, V. Lakshmi. "Machine Learning Based

Predictive Model for Data Fusion Based Intruder Alert

System." Journal of Algebraic Statistics 13.2 (2022):

2477–2483.

Chaitanya, V. Lakshmi, and G. Vijaya Bhaskar. "Apriori vs

Genetic algorithms for Identifying Frequent Item Sets."

International Journal of Innovative Research &

Development 3.6 (2014): 249–254.

Parumanacha Bhaskar, et al. "Incorporating Deep Learning

Techniques to Estimate the Damage of Cars During the

Accidents." AIP Conference Proceedings. Vol. 3028.

No. 1. AIP Publishing, 2024.

Parumanacha Bhaskar, et al. "Cloud Computing Network in

Remote Sensing-Based Climate Detection Using

Machine Learning Algorithms." Remote Sensing in

Earth Systems Sciences (Springer).

Parumanacha Bhaskar, et al. "Machine Learning Based

Predictive Model for Closed Loop Air Filtering

System." Journal of Algebraic Statistics 13.3 (2022):

416–423.

Paradiesi Subba Rao, "Detecting malicious Twitter bots

using machine learning." AIP Conf. Proc. 3028, 020073

(2024), https://doi.org/10.1063/5.0212693

Paradiesi Subba Rao, "Morphed Image Detection using

Structural Similarity Index Measure." M6 Volume 48

Issue 4 (December 2024) https://powertechjournal.com

Mr. M.Amaraeswar Kumar, "Effective Feature Engineering

Technique For Heart Disease Prediction With Machine

Learning" in International Journal of Engineering &

Science Research, Volume 14, Issue 2, April–2024 with

ISSN 2277-2685.

Face Recognition with Convolutional Neural Networks for Enhanced Security

223