Third Generation Security System for Face

Detection in ATM Machine Using Computer

Vision
1th 2th 3th
Sridevi S K.M. Monica Senthil G. A
Assistant Professor, Dept. of Computer Assistant Professor, Dept. of Computer Associate Professor,
Science and Engineering, Science and Engineering, VIT Chennai Dept. of Information Technology,
Sri Ramachandra Engineering and Campus, Chennai, India. Agni College of Technology,
2022 International Conference on Computer, Power and Communications (ICCPC) | 979-8-3503-9784-0/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICCPC55978.2022.10072096

Technology, Chennai, India [email protected] Chennai, India.

[email protected] [email protected]

4th 5th
R. Prabha A. Sathya 6th
Ramya J
Associate Professor, Dept. of Assistant Professor, PG Student,
Electronics and Communication Dept. of Artificial Intelligence and Data Dept. of Computer Science and
Engineering, Science, Engineering,
Sri Sai Ram Institute of Technology, Sri Sai Ram Institute of Technology, Vels Institute of Science, Technology &
Chennai, India. Chennai, India. Advanced Studies (VISTAS),
[email protected] [email protected] Chennai, India.
[email protected]

Abstract– A workflow is proposed for the detection of commands on an ATM screen which are red by using a
unauthorized persons in an ATM. Certain assumptions, consumer to offer for the cooperative process of the ATM.
including the context of identification are made by training their
features and storing the datasets. However, not every individual By reading the commands in the display, a customer
will be a suspect. Suspicious activity on ATMs that are located in
remote areas and to reduce the risk of fraudulent transactions.
is capable of using the ATM through records and records
such as using another person's card to withdraw money etc., entered on a keypad. But in lots of instances, customer
Various video survey and image processing have been discussed overlook it. So most effectively the authors designed a device
in relation to surveillance methods. It discusses the various that facilitates users to apply the ATM system with no need of
processes, preprocessing, classification, feature extraction and any ATM card, the surroundings, pace of interaction, and co-
the corresponding video processing methods. located user displays are all factors that affect how secure the
technique is. Finding fraudulent card transactions is one of the
main concerns from all of these factors. Here, we advocate for
Keywords: Image Processing, Workflow, Automatic detection,
the facial recognition technology to lessen ATM transaction
Feature detection, ATM, Image Processing, Haar Cascade
Algorithm, Machine Learning. fraud.

II. RELATED WORK

I. INTRODUCTION

The project aims to reduce the risk associated with In the paper, the existing system, an RFID card serves as
ATMs that are installed in remote areas, as well as the an ATM card, while an infrared sensor senses the presence of
possibility of fraudulent transactions like using someone else's cardholders and turns on a fan, and the lights GSM network is
card to withdraw money. To overcome these challenges, the used to send SMS to two main stations in case of ATM
authors have developed a solution that uses machine learning tampering. There are a number of ways to spoof biometrics
& artificial intelligence to ensure that card access will only be such as fingerprints and eyeballs. fingerprint authentication
granted to authorized users who have been identified by a face additionally has a few negative aspects. As it may easily be
recognition algorithm. ATM presents PIN to all its customers ripped off via making use of a fake finger and may be
with the assistance of which they are able to get admission to unrecognized because of the presence of injuries or specks of
their process to hold clients out, economic transactions, and/or dirt in the finger [5-8].
banking for the reason, that transaction is frequently reliant on
PIN verification. To preserve the privacy for user In the paper, in iris biometric is used for network security
authentication at ATMs, a number of aesthetic elements were at customer aspect and cryptography and steganography at
examined. present ATMs come handy as well as smooth to server side the usage of the MD5 set of rules. However, iris
apply for every client. Present ATMs commonly offer biometrics is greater highly Iris recognition is based on the

979-8-3503-9784-0/22/$31.00 2022
c IEEE 143

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.
 depth of the light, its recognition rate is much less accurate. while the other plausibility can be, on the off chance that the
The light situation, expression of the face, digicam position, shopper photo do not get coordinated with the photos from the
and many others. Makes trouble inside the face identification account holder profile interior the bank database inside a division
method. of a moment an alert will be sent as a connect to the account
holder’s enlisted Mail Id

In the paper, the ATM that sound reputation machine is

hooked up is characterized in that by means of the card reader,
loudspeaker, voice inducting device, deposit groove, button
display screen are formed with bar printer and external speech
cipher transcript. Voice recognition approach for verification
in ATM machines. Change of voice tone and similar voices
might be an issue in this process. voice recognition pattern
which has a few problems like spoofing assault with the aid of
recorded voice. Henceforth, a more comfortable biometric
authentication framework is needed [8,9].

In the paper, IoT primarily based systems that has been

proposed has many ranges of gadget towards bodily in addition
to electronic assaults. The system that has been evolved is at
the Arduino for securing the inputs with the aid of using many
sensors like modem, vibration sensors, temperature sensors,
sound sensors. This gadget works on a simple formula which
is that if the sensor crosses its threshold degree, then the alarm
sound comes into play and this indicates the better officers
indicating that there's something wrong is happening in the
ATM [10].

In the paper, the implementation of ATM protection

device he usage of fingerprint and GSM version offers the
benefits of balance and reliability. worldwide device for cell
phones offers the same old within the global. disadvantage is
because of inadequate variety of mobile couldn’t get the right
of entry to the safety code of ATM PIN [11-12].

III. PROPOSED METHOD

The pre-processing is the preliminary step for the

technique of video frames, it includes the conversion of image
frames right into a grayscale photo. The pre-processing of the
input video is carried out to cast off the noise and outliers,
that's completed by way of Gaussian filter. It’s entered and
output are depth snap shots. The aim of pre-processing is the
improvement of the photo information that suppresses useless Fig. 1. System design of ATM Processing
distortions or enhances a few image functions essential for
similarly processing. The legal person could be allowed for the 3.1. Feature Detection
further transaction Then the unauthorized individual’s picture
will be dispatched to the respective mail identity. To offer a Feature detection of someone is detected through
protected exchange the pinhole advanced camera which is using the principal component analysis together with the Haar-
settled inside the ATM gadget will capture an image of the cascade algorithm, detection is based on the threshold, the
individual who is progressing to withdraw the cash from the primary purpose of this technique is to come across the
ATM. unauthorized individual primarily based on the threshold.
There is a possibility of fraud when ATMs are installed in
remote areas and people can use others' cards to withdraw
money.
At that point the photo of the user will be in comparison
with the account holder’s photo inside the individual bank
dataset. In case the person’s picture receives matched with the
anybody of the photo interior the dataset strategy at that point it's
reaching to allow the individual to carry out an exchange inside

144 2022 International Conference on Computer, Power and Communications (ICCPC)

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.
 for binary class that changed into truly successful. It is the best
place to start if you are new to boosting.

3.4 Local Binary Pattern

By thresholding each pixel's physical area and

viewing the result as a binary value, the Local Binary Pattern
(LBP) structuring algorithm can identify each pixel in a
picture. The discriminative strength and computational
simplicity of the LBP texture operator have made it a popular
method in many applications. It can be seen as a unifying
technique for the statistical and structural models of texture
Fig. 2. Haar – Cascade Algorithm analysis, which are typically at odds with one another. The
LBP operator's ability to adapt to periodic gray-scale
variations caused on, for example, by changes in illumination,
could be its most important feature in practical applications.
3.2 Haar Cascade Another key aspect of it is its algorithmic adaptability, which
permits photo analysis in challenging real-time circumstances.
An image or video can be identified as having objects The basic premise of the LBP operator is that of two
by using the Haar cascade, which is a machine learning complementary measurements—localized spatial arrangement
algorithm. Their 2001 paper on "rapid object Detection the and grey level contrasting—can be utilised to describe two-
usage of a Boosted Cascade of easy functions" the authors dimensional characteristic properties. The classical LBP
proposed the Haar Cascade as a system learning object operator (Ojala et al. 1996) generates labels for the picture
detection algorithm to perceive items in images or motion pixels by quantizing the 3 x 3 region of every pixel only with
pictures. there may be absolute confidence that the cascade centre value and handling the output as a binary integer. The
classifier is an exceedingly a hit technique for face detection. distribution of such 28 = 256 unique labels can then be used to
In fact, this method can be used for the detection of a great describe the texture. With these operators and a simple local
number of objects with rigid structures, not just faces. Human contrast metric, unsupervised textural segmentation achieved
faces have certain things that are similar. Using Haar features, positive outcomes (Ojala and Pietikäinen 1999). Other related
these regularities can be matched. methods for dividing up patterns and colours have since been
developed.

Fig. 3 Training Haar Cascades

3.3 AdaBoost
The Adaptive Boosting approach combines more
than one classifier to growth the accuracy of classifiers. It's
miles an ensemble technique that mixes numerous poorly
appearing classifiers after which builds a robust classifier with
the resulting excessive accuracy sturdy classifier.
There are two basic requirements for Adaboost: It
needs to be trained on weighted sets of training examples. In Fig. 4. Local Binary Pattern
each iteration set, it should try to minimize training errors
while providing an excellent fit for the examples. A boost is a
well-known ensemble technique that mixes some of vulnerable 3.4 Principal Component Analysis
classifiers to create a robust one. It really works by In conclusion, Principal Component Analysis (PCA)
constructing a version from the education records, then aims to reduce the dimensionality of a data set made up of
constructing some other model that corrects the mistakes from several variables that may be associated among one another.
the primary. The model is introduced till the training set is
anticipated flawlessly or a maximum quantity of fashions is
brought. Adaboost became the primary boosting set of rules

2022 International Conference on Computer, Power and Communications (ICCPC) 145

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.
 Since PCA requires scaling of the dataset, it is crucial that the
dataset is scaled as well.

Fig. 5. Flow Diagram Principal Component Analysis

IV. EXPERIMENTAL RESULTS

The implementations of detection of the authorized person
from the trained dataset will be allowed to process the
transaction and the unauthorized person’s image will be sent
to the respective mail.
A. Creating and Training the Datasets

Fig. 7. Object Detection

Fig. 6. Face Reorganization and Detection of Trained Dataset

146 2022 International Conference on Computer, Power and Communications (ICCPC)

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.
 B. Identifying the person from the dataset and identifying the V. CONCLUSION
unknown
Face Recognition is established as a protection
function for third-generation ATMs. The primary purpose is to
reduce the danger worried inside the ATM system. To keep
away from fraudulent transactions with the assistance of
artificial intelligence technology. The machine can be
recognized by way of face recognition strategies. On this, to
get Haar cascading classifier (algorithm) with shape predictor
and to get more accuracy.

VI. FUTURE WORK

The Future work of Face Recognition is enhanced as

a protection in real-time security applications include
industrial 5.0 ATMs surveillance systems, military
surveillance, and slightly elevated firms. This can also be used
to get access to bank lockers. The lighting offered to the system
is an important consideration. The use of fast machines can
boost efficiency and to enhanced with Artificial Intelligence -
Machine Learning.

REFERENCES
[1] Banerjee, S. Mookherjee, S. Saha, S. Ganguli, S. Kundu,
and D. Chakravarti, "Advanced ATM System Using Iris
Scanner," 2019 International Conference on Opto-
Electronics and Applied Optics (Optronix), Kolkata, India,
2019, pp. 1-3.doi: 10.1109/OPTRONIX.2019.8862388.
[2] S. Sankhwar and D. Pandey, "A Safeguard against ATM
Fraud," 2016 IEEE 6th International Conference on
Advanced Computing (IACC), Bhimavaram, India, 2016,
C. Unauthorized Person Image will be sent to Mail pp. 701-705. DOI: 10.1109/IACC.2016.135.
[3] Y. Cheng, W. Shang, L. Zhu and D. Zhang, "Design and
implementation of ATM alarm data analysis
system," 2016 IEEE/ACIS 15th International Conference
on Computer and Information Science (ICIS), Okayama,
Japan, 2016, pp. 1-3. doi: 10.1109/ICIS.2016.7550948.
[4] C. Porretti, R. Lahaije and D. Kolev, "A New Vision for
ATM Security Management: The Security Management
Platform," 2016 11th International Conference on
Availability, Reliability and Security (ARES), Salzburg,
Austria, 2016, pp. 493-498. doi: 10.1109/ARES.2016.50.
[5] H. R. Babaei, O. Molalapata, and A. A. Pandor, Face
Recognition Application for Automatic Teller Machines
(ATM), in ICIKM, 3rd ed. vol.45, pp.211-216, 2012.
[6] Aru, O. Eze and I. Gozie, “Facial Verification Technology
for Use in ATM Transactions”, American Journal of
Engineering Research (AJER), [Online] 2013, pp. 188-
193, Available:http://www.ajer.org/papers/v2
(5)/Y02501880193.pdf ← K.
[7] Bone, Mike, Wayman, Dr. James L., and Blackburn,
Duane. “Evaluating Facial Recognition Technology for
Drug Control Applications.” ONDCP International
Counterdrug Technology Symposium: Facial Recognition
Vendor Test., June 2001.
[8] R. Prabha, M. Razmah, G. Saritha, R. Asha, S. G. A and R.
Fig. 8. Unauthorized Person Detection
Gayathiri, "Vivoice - Reading Assistant for the Blind using
OCR and TTS," 2022 International Conference on
Computer Communication and Informatics (ICCCI), 2022,

2022 International Conference on Computer, Power and Communications (ICCPC) 147

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.
 pp. 01-07, doi: 10.1109/ICCCI54379.2022.9740877.
[9] R. Prabha, S. G. A, N. N. Saranya, A. M, K.
Somasundaram and K. C, "Design of Adaptive Priority
Based IoT Communication in Wireless Network," 2022
International Conference on Advanced Computing
Technologies and Applications (ICACTA), 2022, pp. 1-5,
doi: 10.1109/ICACTA54488.2022.9753550.
[10] Senthil, G. A., Prabha, R., Roopa, D., Babu, D. V., &
Suganthi, S. (2021, March). Improved Cluster Head
Selection for Data Aggregation in Sensor Networks.
In 2021 7th International Conference on Advanced
Computing and Communication Systems (ICACCS) (Vol.
1, pp. 1356-1362). IEEE.
https://doi.org/10.1109/ICACCS51430.2021.9442048.
[11] S. G. A, R. Prabha, M. Razmah, S. Senthilpandi, S.
Suganthi and S. Sridevi, "Design of a Novel Group
Communication Framework to Improve Security in
Internet of Things," 2022 8th International Conference on
Advanced Computing and Communication Systems
(ICACCS), 2022, pp. 967-970, doi:
10.1109/ICACCS54159.2022.9785080.
[12] S. Sridevi, R. Prabha, K. N. Reddy, K. M. Monica, G. A.
Senthil and M. Razmah, "Network Intrusion Detection
System using Supervised Learning based Voting
Classifier," 2022 International Conference on
Communication, Computing and Internet of Things
(IC3IoT), 2022, pp. 01-06, doi:
10.1109/IC3IOT53935.2022.9767903.

148 2022 International Conference on Computer, Power and Communications (ICCPC)

Authorized licensed use limited to: K K Wagh Inst of Engg Education and Research. Downloaded on August 26,2024 at 05:41:20 UTC from IEEE Xplore. Restrictions apply.