2022 8th International Conference on Smart Structures and Systems (ICSSS)

A FACE RECOGNITION SECURITY MODEL USING

TRANSFER LEARNING TECHNIQUE

G NaliniPriya R.Meenakshi
Professor , Professor , Mythili R
Department of IT , Department of Computer Science , Department of IT,
Saveetha Engineering College, Chennai Institute of Technology, Saveetha Engineering College,
Chennai, Tamil Nadu Chennai, Tamil Nadu, India Chennai, Tamil Nadu
[email protected] [email protected] [email protected]

Harini E Harini S B
Mathiyazhagi B
Department of IT, Department of IT,
Department of IT,.
Saveetha Engineering College, Saveetha Engineering College,
Saveetha Engineering College,
2022 8th International Conference on Smart Structures and Systems (ICSSS) | 978-1-6654-9761-9/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICSSS54381.2022.9782238

Chennai, Tamil Nadu Chennai, Tamil Nadu

Chennai, Tamil Nadu
[email protected] [email protected]
[email protected]

ABSTRACT: Nowadays security plays a vital role in day-to-day Deep learning is a machine learning technique that teaches
life. Without it, so many problems will arise and it also pave way computers to learn by doing what people do naturally. Deep
for the occurrences of illegal entries and information theft, learning is a fundamental component of driverless cars,
security attacks. Due to this, confidential information gets lost allowing them to detect a stop sign or differentiate between a
and this led to many issues. To avoid this, we have created a
pedestrian and a streetlight. Deep learning is a type of
system of facial recognition which completely detects and
recognizes the face by capturing images of user, crop them and machine learning in which a computer model learns to execute
saves them in a database. If any unauthorized user enters the categorization tasks directly from images, text, or sound that
security system, then it will detect them, and prevent does not have been educated by people. Deep learning models can
allow them from entering into the security system. In this project, produce accurate findings as a result of training, and their
we have applied a transfer learning model based on Alexnet Deep functionality can occasionally outperform human
learning application to develop a real time facial recognition performance. Models are trained utilizing a huge amount of
system which has good robustness to cope with pose and lighting, labelled data and multilayer neural network topologies.
reduce dimensionality, complexity and better recognition
accuracy. This system accurately detects and compares the image One of the most widely utilized applications in deep
that are stored in the database. And accurately provides the results learning is transfer learning. You can utilize a pre-trained
at faster rates. Also, this system would be able to create and read network as a jumping off point for learning a new skill.
a database of different users. Transfer learning is typically faster and easier to set up than
training a network with randomly initialized weights from
Key words: Facial recognition, MATLAB, Transfer
zero. AlexNet has been trained on over a million photos and
learning, Alexnet, Security model.
can categories them into over many different object categories
for a wide spectrum of images, the network learned feature-
I.INTRODUCTION rich representations. The network accepts an image as input
and generates a label for each object in the image, along with
The process of identifying a person in their probabilities as result.
photos or videos by analyzing and comparing patterns.
Recognition of people's faces is known as Face Recognition.
The facial features of a person are frequently extracted using II. RELATED WORKS
algorithms compares them to the facial traits in the movie
database and finds the most appropriate match. It has evolved
into a vital component of bio-metric, security, and We had gone through many papers that use face
surveillance systems. Systems for indexing images and recognition system for various applications in real time
videos. A face recognition system is a complex image and we have found certain paper related to our paper.
processing system. In real-world applications with K. Sharma et al [1], the proposed the paper on the
complicated data, there is a difficulty light, occlusion, and study and Implementation of Face Detection Algorithm
imaging condition influences on the photos in real time. The which is done with the help of using MATLAB. C.
components of a facial recognition system are as follows: In Sanchez et al [2], the authors developed a system of
image analysis, face detection and identification methods are MATLAB Simulation of Algorithms used for Face
used. To find the location of the object, a detection process is Detection process involved in Video Surveillance. A. A.
used faces in a photograph in the meantime. Mohamed et al [3], this paper explains about Face
Recognition technique by applying Artificial Neural

978-1-6654-9761-9/22/$31.00 ©2022 IEEE

1

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.
 2022 8th International Conference on Smart Structures and Systems (ICSSS)

Network Using MATLAB technologies and simulation used to store images of the users, and the model will be
techniques. Lazar et al [4], they proposed a model of trained to identify the face of the user. We use Alexnet
practical implementation that is used for face detection by model that checks for accuracy and can rate the people
using MATLAB technologies in cascade object detector after training the model with the images stored in the
stacks. J. Daudpoto et al [5], the authors have developed a database.
Real time face detection/monitor to detect accurate faces
for enhancing better performance using raspberry pi and
MATLAB simulations.
M. Ayi et al [6], proposed a new model based on
Interfacing of MATLAB technologies with Arduino
functionalities for face detection, recognition and
developed tracking algorithm using serial communication
in the model. N. R. Borkar et al [7], developed a Real-time
implementation of face recognition systems. L. Saad et al
[8], the authors have developed a system based on
surveillance authorization for Monitoring and
Complaining Web-based Face Recognition system using
Haar-MATLAB.N. Singh et al [9], this paper is a study of
Template matching schema that is used for detection and
recognition of frontal view of human face through
MATLAB simulation techniques. Kumar et al [10], the
authors have worked on the field of study on Comparison
of HOG methodologies and Fisherman’s faces Based Face
Recognition System for security purposes Using
MATLAB simulation. Fig. 1. Proposed System Architecture
M. Iskandar et al [11], has developed a system
In a live stream, the proposed system can detect multiple
model of Face Detection with Expression Recognition
faces and identify the exact face with face of the user that
using Artificial Neural Networks applications of deep
is stored in the database. The proposed system improves
learning concepts in brief. P. Chawla et al [12], proposed
the accuracy of the Authorized User and Non- Authorized
a system based on Authentication of Face security system
user faces.
which deals with secure and safer constraint with help of
PROPOSED ALGORITHM -FRTL ALGORITHM:
using MATLAB. M. Jridi et al [13], the authors have
worked on enhancing a new approach of Fast face
recognition system approach using a graphical processing Input: Dataset
unit GPU as a base element for the entire system module. Output: Accuracy is detected
Lazar et al [14], this paper exhibits an idea and A practical Step 1: Start
implementation of face detection methodologies by using Step 2: Datasets are taken as input (Image Capture).
MATLAB cascade object detector system. F. Sultana et al Step 3: Datasets are images of Authorized Users.
[15], the authors developed a systematic model of Face Step 4: Train the model with the images that are stored in
recognition system using PCA algorithm in CNN-based the database as resource.
method. Step 5: Images in the dataset are pre-processed and the
model id pre-trained to recognize the faces.
Step 6: Faces of the persons will be detected exactly by
III. SYSTEM ARCHITECTURE the model after training by matching with the faces in
the dataset.
Step 7: The person whose face is present in the dataset
The well-constructed system architecture for this project, will be recognized as Authorized user and the other will
a real-time face recognition security system employing be un-authorized user.
Alexnet-deep learning application of transfer learning, as Step 8: Multiple faces of the users can be stored as
well as its modules and requirements, is outlined below. different layer in the model at a time.
As shown in Figure 1, the method entails obtaining facial Step 9: The output is detected with accuracy.
images from individuals, preprocessing the images to a Step 10: Stop
standard normalization, and then directing them into pairs
of convolutions using the pretrained AlexNet Fig. 2. Proposed Algorithm for Face Recognition using
convolutional neural network transfer learning approach Transfer Learning.
for feature extraction and classification.
To detect a face of a human, the proposed system The proposed system gives an edge over the
employs Deep Learning application transfer learning and existing system by detecting multiple faces at a time and
using MATLAB. We will use the exact database that is can detect in live stream.

978-1-6654-9761-9/22/$31.00 ©2022 IEEE

2

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.
 2022 8th International Conference on Smart Structures and Systems (ICSSS)

Datasets are loaded to detect the faces of Let us consider two images as an input (layer) and that
authorized users and non-authorized users. Figure2 needed to be analysed with the help of face recognizer
explains proposed Face Recognition using Transfer model.
Learning Algorithm. The algorithm shows that we can
detect the faces by using the datasets as inputs. After the
A. MATLAB:
datasets are loaded, the process begins to detect the faces
that are authorized users or un-authorized users. The output
will be implemented by the Face accuracy by the model MATLAB is a high-level language used for better
with respect to the dataset information. performance of technical computing. It integrates
mathematical calculation, data visualization and high-
level programming in a user-friendly environment where
problems and solutions are expressed in familiar
PROPOSED PROCESS FLOW: mathematical notation. It is commonly used in the fields
of Math and computation, for Algorithm development, in
Modeling, for simulating, and also in prototyping. And
other areas of specializations are Data analysis,
exploration, and visualization, in Scientific and
engineering graphics, Application development and also
includes Graphical User Interface building.

B. ALEXNET:

AlexNet is the fundamental architecture that serves as a

basis for any type of object identification, and it comes
with a wide range of applications in the field of computer
vision for the purpose of addressing artificial intelligence
challenges. AlexNet was trained on over a million photos
Fig. 3. Proposed Process Flow Diagram. and is capable of classifying images into 1000 different
item categories. For a wide variety of images, the network
The proposed system explains who the model is trained learned feature-rich representations. The network accepts
inside the MATLAB by applying the transfer learning an image as input and outputs a label for each object in the
techniques commonly used in AlexNet as follows, image, along with the probabilities for each object
category.
1. Load the pre-trained network model: Train the 1. Face Detection and Capture
System model with images of the users.
2. Replace final layer: You can adjust the layers by
adding or removing images of users.
3. Train Network: Train the model with the images
stored in the dataset.
4. Predict and access network accuracy: Trained
model is tested by running and the accuracy of the
network is found.
5. Results: Model exhibits the results after testing
along the accurate faces of the authorized users or
unauthorized users.

IV. IMPLEMENTATION

The concept is implemented using MATLAB and Fig.4 Input capture User image
ALEXNET. They are trained by following ways:

1. Load datasets from the user capture images. Alexnet was a deep innovative neural network introduced in
2. Training the model with the dataset information. the Visual ImageNet Visual Recognition challenge,
3. Recognize faces in the image by normalizing. where it defined a previous learning range.
4. Use a face analyser to distinguish between faces The complete process of a real-time face-lift is
of authorized users and faces of unauthorized demonstrated as following steps.
users.

978-1-6654-9761-9/22/$31.00 ©2022 IEEE

3

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.
 2022 8th International Conference on Smart Structures and Systems (ICSSS)

OF THE USER” as shown in Fig 3 above. It displays

the result at 100% accuracy at time rate of 0.10 seconds.
2. Training Model
V .COMPARISON BETWEEN PROPOSED MODEL AND
EXISTING MODEL

While comparing our proposed project model with

other existing project model, we get that the proposed project
system gives an accuracy of 100% higher than other existing
system and we have compared the Accuracy recognition,
Mean, Variance, Model, time taken and by inferring this we
could able to see that number of models that we collected 5
existing system models and the time taken in proposed system
is also less when compared to 5 existing system which will be
helpful in providing quick result. This comparison is useful to
detect the accuracy of the existing system and proposed FRTL
Fig.5 Training Model with images in the database. system.

3. Testing Model Table 1. Comparison of Face Recognition Models

Recognition Varian
S.NO Model Mean Time
Accuracy ce
Global expansion 4.58
1 91.67% 91.65 0.02
ACNN min
Global + local
2 93.30% 93.27 0.03 5.7 min
Expansion ACNN
0.46
3 CNN + SVM 97.50% 97.47 0.03
min
AlexNet +
SVM,RasNet-50 +
4 95.63% 95.6 0.03 0.087 s
SVM,Transfer
learning (AlexNet)
Transfer Learning
5 100.00% 100 0 0.10s
using Alexnet

Fig .6 Face recognition output (Authorized User)

5.1 COMPARISON OF FACE RECOGNITION
The system model represents the face recognition TECHNIQUES.
scheme in the order of pre-processing the dataset with
captured images of the users, training the network with The below bar chart describes the relative survey of
the stored dataset images and finally testing the trained the proposed system with other face recognition system
model. models. On comparing this with other literature papers, we
were able to infer that our paper is much better in terms of
attributes such as accuracy and time taken.
COMPARATIVE RESULTS: PROPOSED SYSTEM VS EXISTING SYSTEM

10

0
TIME TAKEN

RECOGNITION

RELIABILITY
ACCURACY

Fig.7. Face recognition output (Un-authorized user)

In the testing model, the trained model recognizes the

faces accurately along with user names as stored and Proposed system Existing System
saved in the dataset folders and if the user is unauthorized
it shows “NO FACE DETECTED” as shown in Fig 4
above and If it is authorized user it displays “NAMES
Fig 8. Comparison of Face Recognition Techniques.
978-1-6654-9761-9/22/$31.00 ©2022 IEEE
4

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.
 2022 8th International Conference on Smart Structures and Systems (ICSSS)

 The primary goal of our model is to help the world

The proposed system's reliability is higher than the reliability to build a secured security system for authorization.
of the existing system.Overall, we can see a clear difference Here, first we need to capture the images of the users
between the existing project and the proposed project. It also and store it in the datasets and then we need to pre-process the
shows that it is more reliable, durable, and highly scalable. captured images of the user’s s’ faces in the dataset. For each
user s’ image, the model is trained to recognize, we need to
VI. RESULTS AND DISCUSSION apply a face classifier to determine the faces of authorized
users and unauthorized users. Here, we need to load the user
The proposed system is an outcome of one of the applied s’ face dataset to detect faces in images and find accurately
applications of Deep Learning based on face recognition the result.
techniques by applying transfer learning using Alexnet. The
model is trained with the capability of recognizing faces and
also it allows the user to add the faces of many persons at a VII. CONCLUSION
time in the dataset which is considered as layers for analysing
and comparing the faces foe better results. Those who or not
an authorized user they will not be permitted, only the The goal of our project is basically a Safety System
authorized users are permitted to access the resources. that captures and recognizes face and prevents illegal entry into
any system and unauthorized persons entering the system. This
Face Recognition system can be used in any system for its
confidential information Security, and for users who wants to
prevent their information from unauthorized access. This Face
Recognition system is very effective than other systems as this
is built using Transfer Learning using Alexnet. This
methodology is very simple and it collects sample pictures of
the multiple users and stores them in the Database whom the
user wants to give access. Other than the person, whose image
is stored in the Database, the system does not recognize their
face and prevent access.

This system is very secured and safe, which is the

most important needed application in today's world. We
enhance this system by designing them with advanced
features like image capturing with fingerprint recognition,
allowing access when the user encloses proper information
into the system. Also, in this system we enhance the features
so that it will prevent Cyber-attacks and other forms of
security attacks. This will really help in building Security
System safer.

REFERENCES

[1] K. Sharma, V. Gupta, S. Verma and S. Avikal, "Study and

Implementation of Face Detection Algorithm Using Matlab,"
2018 International Conference on Recent Innovations in
Electrical, Electronics & Communication Engineering
(ICRIEECE), 2018, pp. 2745-2748, doi:
Fig. 9. Activity Diagram for Face Recognition Using 10.1109/ICRIEECE44171.2018.9008990.
Transfer Learning https://ieeexplore.ieee.org/document/9008990
[2] C. Sanchez, D. Martinez-Mosquera and R. Navarrete, "Matlab
Simulation of Algorithms for Face Detection in Video
The Activity Diagram in the Fig 9 represents the entire
Surveillance," 2019 International Conference on Information
activities that are involved starting from the image capturing, Systems and Software Technologies (ICI2ST), 2019, pp. 40-47,
image pre-processing, training the model, image classification doi: 10.1109/ICI2ST.2019.00013.
and testing model. https://ieeexplore.ieee.org/document/8940420
[3] A. A. Mohamed, K. H. Bilal and I. E. Elmutasim, "Face
The details of our model can be classified as: Recognition by Artificial Neural Network Using MATLAB,"
2021 IEEE 6th International Conference on Computing,
 Our model lays as an interface between the user and Communication and Automation (ICCCA), 2021, pp. 686-690,
Camera module. doi: 10.1109/ICCCA52192.2021.9666434.
https://ieeexplore.ieee.org/document/9666434

978-1-6654-9761-9/22/$31.00 ©2022 IEEE

5

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.
 2022 8th International Conference on Smart Structures and Systems (ICSSS)

[4] E. Alionte and C. Lazar, "A practical implementation of face 0.1109/ICGCIoT.2015.7380598.https://ieeexplore.ieee.org/docu

detection by using Matlab cascade object detector," 2015 19th ment/7380598.
International Conference on System Theory, Control and [14] P. Tamrakar and H. Amhia, "ANN and AdaBoost
Computing (ICSTCC), 2015, pp. 785-790, doi: Based Effective Face Recognition and Improvement
10.1109/ICSTCC.2015.7321390.https://ieeexplore.ieee.or of Classification Training," (ICSCAN), 2021, pp. 1-
g/document/7321390 5, doi: 10.1109/ICSCAN53069.2021.9526499.
[5] M. Ayi, A. K. Ganti, M. Adimulam and B. Karthik, "Interfacing https://ieeexplore.ieee.org/document/9526499
of MATLAB with Arduino for face detection and tracking
[15] S. Rao and M. V. B. Rao, "A novel triangular DCT
algorithm using serial communication," 2017 International
Conference on Inventive Computing and Informatics (ICICI), feature extraction for enhanced face recognition,"
2017, pp. 944-948, doi: 2016 10th International Conference on Intelligent
10.1109/ICICI.2017.8365276.https://ieeexplore.ieee.org/d System (ISCO), 2016, pp. 1-6, doi:
ocument/8365276 10.1109/ISCO.2016.7726977.
[6] S. L. Saad, M. M. Kamal and N. A. Zamri, "Monitoring and https://ieeexplore.ieee.org/document/7726977
Complaining Web-based Face Recognition using Haar-
MATLAB," 2018 8th IEEE International Conference on Control
System, Computing and Engineering (ICCSCE), 2018, pp. 206-
211, doi:
10.1109/ICCSCE.2018.8685017.https://ieeexplore.ieee.or
g/document/8685017
[7] N. Singh, A. K. Daniel and P. Chaturvedi, "Template matching
for detection & recognition of frontal view of human face through
Matlab," 2017 International Conference on Information
Communication and Embedded Systems (ICICES), 2017, pp. 1-
7, doi:
10.1109/ICICES.2017.8070792.https://ieeexplore.ieee.org
/document/8070792
[8] N. V. Megha Chandra Reddy and K. Kumar, "Comparison of
HOG and Fisherfaces Based Face Recognition System Using
MATLAB," 2021 2nd International Conference for Emerging
Technology (INCET), 2021, pp. 1-5, doi:
10.1109/INCET51464.2021.9456366.https://ieeexplore.iee
e.org/document/9456366
[9] M. Owayjan, R. Achkar and M. Iskandar, "Face Detection with
Expression Recognition using Artificial Neural Networks," 2016
3rd Middle East Conference on Biomedical Engineering
(MECBME), 2016, pp. 115-119, doi:
10.1109/MECBME.2016.7745421.https://ieeexplore.ieee.o
rg/document/8279123
[10] S.Tanwar, P. Chawla, R. Maadan and P. Bhadana,
"Authentication of Face using Matlab," 2020 5th International
Conference on Communication and Electronics Systems
(ICCES), 2020, pp. 1229-1234, doi:
10.1109/ICCES48766.2020.9137979.https://ieeexplore.ieee.org/
document/9137979
[11] Y. Ouerhani, M. Jridi and A. AlFalou, "Fast face recognition
approach using a graphical processing unit “GPU”," 2010 IEEE
International Conference on Imaging Systems and Techniques,
2010, pp. 80-84, doi:
10.1109/IST.2010.5548545.https://ieeexplore.ieee.org/document
/5548545
[12] E. Alionte and C. Lazar, "A practical implementation of face
detection by using Matlab cascade object detector," 2015 19th
International Conference on System Theory, Control and
Computing (ICSTCC), 2015, pp. 785-790, doi:
10.1109/ICSTCC.2015.7321390.https://ieeexplore.ieee.org/docu
ment/7321390
[13] R. Subban, "Human face recognition using facial
feature detection techniques," 2015 International Conference on
Green Computing and Internet of Things (ICGCIoT), 2015, pp.
940-947,doi:

978-1-6654-9761-9/22/$31.00 ©2022 IEEE

6

Authorized licensed use limited to: Nirma University Institute of Technology. Downloaded on June 13,2024 at 13:39:12 UTC from IEEE Xplore. Restrictions apply.