Enhanced Face Recognition Algorithm for Real-Time Applications

with Improved Accuracy

Raji Pandurangan, Lavanya C and Swetha C

Dept. of Electronics and Communication Engineering, Saveetha Engineering College, Chennai, India

Keywords: Face Recognition, Real-Time Detection, Feature Extraction, Accuracy, Precision, Robustness.

Abstract: Encouraging the development of safe and effective face recognition systems in applications such as

surveillance, access control, and identity verification, developing accurate and reliable algorithms has become

a significant focus in computer vision. Traditional methods like Dense U-Nets, Retina Face, and Hierarchical

Networks provide satisfactory results but often fall short in balancing accuracy, precision, sensitivity, and

specificity, especially in real-time scenarios. To address these limitations, we propose an enhanced face

recognition algorithm that leverages improved feature extraction and optimized classification techniques. Our

proposed method was evaluated alongside existing algorithms, achieving superior performance achieving an

overall accuracy of 89.50% while also achieving superior values for specificity, sensitivity, and precision.

These results demonstrate the method's effectiveness and suitability for high-stakes applications where

reliability is paramount. Future work will focus on adapting the model for diverse conditions such as low-

light and multi-angle environments to further improve robustness.

1 INTRODUCTION

Recognition of Faces is a Biometric Approach for

Detecting Faces. The challenge of verifying or

identifying a face from the multimedia images is

accomplished by employing a facial recognition

method. Recognising people by face has become

increasingly crucial as our civilisation has

progressed. Across the globe, face detection and

identification systems have been expanding (Shetty,

Rebeiro, et al. , 2021). Although the classroom

security camera is not fixed, the angle at which faces

are captured in the footage varies as well. This face

verification technique, which is based on deep

learning (DL), has demonstrated encouraging

outcomes in controlled settings, its performance in

uncontrolled environments might use a lot of

improvement. Both the network's output and the

dataset's diversity and quality affect the model's

performance (Li, Shen, et al. , 2023). Human people,

at birth, possess a distinctive feature that enables

them to recognise other persons, namely the "face."

In addition to speech and fingerprints, facial

recognition is a crucial component of biometrics, as

no two individuals possess identical facial features.

Myths that an individual may share facial

characteristics with seven different people, and face

detection and identification systems assist in

distinguishing among them. Facial expression

detection, face detection, and recognition have

significantly influenced the domain of image

processing (Wattamwar, Mate, et al. , 2021).

The development of algorithms for facial

recognition has been rapid and influenced by

numerous variables. Many approaches to veiled facial

recognition in light of the unexpected elements

encountered in actual life. Nevertheless, masked

facial recognition diverged from occluded face

recognition of the SarS-CoV2 pandemic (Neto, Pinto,

et al. , 2022). One thing that differentiates face

recognition algorithms is the fact that it uses a

confirmation or identification approach that assesses

personality. Two processes: facial recognition and

facial identification, form the backbone of the

method. Facial recognition is applicable to personal

computers with a few different ways that photos of

people's faces can be recognised. The "right" data

from frontal facial photos is typically used to

accomplish face recognition. Unblemished faces seen

on CCTV are a great example, even if there are many

contexts in which entire faces are not visible

(Shamrat, Majumder, et al. , 2022). Several research

214

Pandurangan, R., C, L. and C, S.

Enhanced Face Recognition Algorithm for Real-Time Applications with Improved Accuracy.

DOI: 10.5220/0013612400004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 214-219

ISBN: 978-989-758-763-4

solutions have been proposed for the automatic

recognition of human emotional states. The World

Health Organisation reported that tragically, many

were unable to administer first aid in a timely manner

after falling on the road, leading to fatalities

(Bharathiraja, Sakthivel, et al. , 2023) By the age-

invariant face recognition, is a system that uses a

correlation between traits related to identity and age.

On the other hand, combining faces of varying ages,

face age synthesis (FAS) gets rid of age variance into

one group. On the other hand, FAS's artefacts hinder

downstream recognition, while face recognition

doesn't provide any visual findings to aid with model

interpretation (Huang, Zhang, et al. , 2022).

Nowadays, face recognition is much sought after

because of its practical use individual verification.

With just a little raw image or a glimpse, the

technology may aid in retrieving the person's

information from a database. Though not widely used

in India, biometric identification systems should use

facial recognition to render safer and without physical

contact, the system; this technology is urgently

needed (Saleem, Shiney, et al. , 2023).

Furthermore, the human brain has a special region

dedicated to face recognition called the fusiform face

area, which is crucial for our survival (Alankar,

Ammar, Kumar, 2020). Face recognition technology

relies on face detection, which entails identifying a

face in an initial image. Pandemic proportions have

been reached as the COVID-19 virus has spread to

over 200 countries. However, the virus is evolving at

a rapid pace, with new strains spreading via both

direct and indirect contact in areas where vaccines are

not yet available. Face masks and other social

distancing measures are necessary because many

infectious diseases spread through droplets, and

micro-droplets in particular (Deepa, Hariprasad, et al.

, 2022).

2 LITERATURE SURVEY

A person's face is the most distinctive feature that

allows others to recognise them. No two people are

exactly same, and that includes identical twins.

Therefore, distinguishing between them necessitates

facial recognition and identification. An example of a

biometric verification system is a facial recognition

system. A fresh collection of test cases is constructed

with extensive property information, and the face

recognition time constraint protocol. Also, for the

purpose of biometrics evaluation under COVID-19,

we have compiled a large-scale masked face subset.

There are three different types of recognition tasks:

conventional, masked, and impartial to ensure a

thorough evaluation of face matchers. An effective

method for training facial recognition models that

does not compromise performance is established

using a distributed architecture (Zhu, Huang, Kumar,

2022). The term "face" refers to the front portion of

an animal's or human's head that extends from the

jawline to the chin. Because it contains so many

crucial facts about a person or thing, face is the most

fundamental aspect of being human. It is believed that

humans can identify one another only by looking at

their faces. Class control for instructors at the

Technical Informatics College of Akre using facial

recognition technology to monitor student attendance

in a classroom setting (Mohammed, Zeebaree,

Kumar, 2021). Two methods, "Template Matching"

and "Local Binary Pattern Histogram (LBPH)," are

being compared. Python, the Raspberry Pi 3 Model

B+, OpenCV, and the LBPH algorithm were used to

build the prototype of the face recognition and

identity security system. This idea presents a method

for recognising random faces using the Haar

classifier. Instead than using databases, this method

compares individuals to a static collection and then

provides matches based on first, second, and third

results. Unlike biometric devices, it does not seek for

specific matches. It functions similarly to a CCTV in

that it can identify people, but it only stores a short

amount of footage (Chowdhury, Sakib, Kumar,

2022). The model can function in a wide range of

environments, including those with different lighting

and backgrounds, to Face mesh. Additionally, this

model can handle processing non-frontal images

containing both sexes, regardless of age or race.

images from the train the deep neural network of the

model, real-time images and the wild face dataset are

used. When testing, the model will report the person's

name if their facial landmarks match those in any of

the training images; else, it will output "unknown."

(Hangaragi, Neelima, et al. , 2023).

The process of creating altered or encrypted

versions of original biometric templates is known as

cancelable biometrics. Modern hacking tools can

retrieve the original biometric data stored in

databases, which led to the development of

cancellable biometrics. One workaround for this issue

is to replace the original biometric templates stored in

the database with cancellable ones. An approach to

cancellable face identification using a Fractional-

Order (FO) Lorenz chaotic system to encrypt facial

images is presented in this study. An individual's red,

green, and blue face image components can be

XORed with randomly generated keys that are

exclusive to that user. The Lorenz chaotic system

Enhanced Face Recognition Algorithm for Real-Time Applications with Improved Accuracy

215

with fractional orders generates these keys. The face

photos' encrypted colour components also undergo

some post-processing, including matrix rotation and

transposition. The last step is to combine the

decrypted and processed parts of the face image using

a wavelet fusion algorithm(Badr, Radwan, et al. ,

2021). The usage of face recognition systems for user

identification is widespread. It is a program that 3D

models utilise to accomplish a wide variety of

machine-based visual tasks, such as detecting edges

from various perspectives to minimise collisions. by

the incorporation of iris recognition into the face

recognition system. Python and OpenCV are utilised

to alter photographs of the face and iris (Srivastava,

Katiyar, et al. , 2022). Local binary pattern (LBP)

classifiers are invariant under different lighting

situations, making them ideal for face detection, and

Haar classifiers are highly accurate in this regard. To

enhance the identified faces, image processing

methods like quantisation, histogram equalisation,

bilateral filtering, and contrast correction are utilised.

To further assess the possibility of the approach for

successfully identifying faces in low-quality images,

have quantised raw face photographs at different

levels (Padmashree, Karunakar, et al. , 2022). A face

recognition program is a piece of technology that can

identify a person in a high-tech photo or video clip. A

safer and more user-friendly world can be discovered

with its help. Some instances include locating missing

people, detecting shoplifting, identifying security

personnel, locating social media profiles, and

recognising vehicle drivers. Detection, extraction,

and identification are the three main components of

face recognition. When it came time for face

detection, the algorithms used were Haar-Cascade,

Eigenfaces, Fisherfaces, and local binary pattern

histograms (Pandey, Yadav, et al. , 2022).

One viable

alternative for the day-to-day management of student

attendance systems is smart attendance with

instantaneous face detection. A facial recognition-

system that tracks attendance uses a person's likeness

to verify their identity. Nowadays, schools are facing

a big issue with the consistency of student attendance.

To take attendance, teachers employ face biometrics

that are based on a high-definition monitor to identify

each pupil. Both of them were more laborious and

demanding of time. media, in addition to several

kinds of digital technologies (Yadav, Sharma, et al. ,

2022). Intelligent technology that can recognise and

identify faces in order to collect attendance. the

attendance system more user-friendly, efficient, and

secure from proxy attacks. There was a possibility of

proxy in the previous manual-based attendance

system, but we will be able to fix it. This method can

clearly convey the idea to the computer as to whether

or not it is a valid proxy or legal attendance. This

system is less complicated and more secure(Bairagi,

Ahmed, et al. , 2021).

Smart city apps are becoming more popular in

many countries because they improve people's

quality of life, make better use of people's time and

resources, and decrease pollution. Because they

provide improved access management and space

allocation, multi-location parking garages are

ubiquitous in smart cities. This helps to alleviate

traffic and delays in densely populated commercial

areas. The IoT has the ability to connect billions of

devices and services throughout the globe, in real-

time, for many purposes. One of the hot topics in

internet of things research right now is smart parking.

Modern metropolitan cities have over one million

vehicles on the road, but there is a severe lack of

parking spaces to accommodate all of them.

3 METHODOLOGY

Modern parking garages are notoriously inefficient.

This means that drivers could waste a lot of time on

busy days just driving around a parking lot looking

for a spot. Better public service, less emissions and

pollutants, increased parking utilisation, improved

city tourist experience, and prevention of unneeded

capital investments are all possible outcomes of

implementing this system, which will also aid in

resolving the increasing problem of traffic

congestion, wasted time, and lost money. Facial

recognition technology finds extensive applications

in diverse security systems, ranging from physical

access control to computer user databases. All the

necessary software and hardware components of the

proposed system have been painstakingly designed.

Its primary objective is to introduce an automated

parking system that enhances the efficiency and

simplicity of parking for both drivers and

administrators.

Figure 1 illustrates a face recognition pipeline

utilizing the SSD (Single Shot Detector) algorithm to

achieve real-time detection and classification of

known individuals. The process is divided into three

primary stages such as Data Set Collection, Feature

Set Extraction and SSD Algorithm for Detection and

Classification. The SSD algorithm enables high-

speed, efficient face recognition, suitable for

applications that require rapid and accurate

identification of individuals. This setup is particularly

useful for security systems, attendance monitoring,

and other real-time identity verification systems.

INCOFT 2025 - International Conference on Futuristic Technology

216

Figure 1: Proposed System Architecture

The system architecture integrates a camera for

real-time monitoring. The SSD algorithm is

employed for face detection. When a face is detected

at a parking slot, the toll gate will open. Multiple

parking slots are visualized using a graphical

interface. This system ensures efficient management

of parking spaces while providing a seamless

experience for users. It combines real-time image

processing with automated access control for an

optimized parking solution. Figure 2 describes the

proposed system block diagram. The system

architecture integrates a camera for real-time

surveillance and an SSD-based face detection

algorithm for efficient monitoring of car parking

areas. As vehicles approach the toll, the camera

captures live video feed. The SSD algorithm

processes the frames to detect faces in real-time,

ensuring precise identification. Simultaneously, the

system visualizes multiple parking slots across

different locations, indicating availability. Detected

faces are cross-referenced with a database to

authorize access to the parking area. If a recognized

face is associated with a valid parking permit, the toll

gate opens, allowing entry. This intelligent system

optimizes parking operations, enhancing user

experience and security at various parking locations.

Figure 2: Proposed System Block Diagram

3.1 Image Acquisition

Image acquisition refers to the process of

capturing a digital image from the physical world,

typically using a camera or some form of optical

sensor. This process involves converting analog

information from the real world into a digital format

that can be stored, processed, and manipulated by

computers.

3.2 Face Detection Module

The SSD is a DL algorithm for object detection,

including faces. Forecasts the classes of objects and

their bounding boxes at the same time, saving time.

SSD uses a multi-scale feature map to detect objects

of different sizes, enhancing its accuracy and speed.

In the SSD algorithm, face alignment typically

involves a preprocessing step before face detection.

This step aims to ensure that facial features within the

detected bounding boxes are consistently positioned.

Common techniques for face alignment include using

facial landmarks or pose estimation to adjust the

orientation and alignment of the detected faces. This

helps improve the accuracy of subsequent face

recognition or analysis tasks.

4 RESULTS AND DISCUSSION

Parking systems are only one area where the IoT has

revolutionised contemporary living by capitalising on these

enormous technological advances. These potential

prompted the development of a smart parking system,

which uses an integrated multiple-slot approach to

automatically notify cars of available parking spots, saving

both time and money by reducing the need to staff parking

lots.

Figure 3: Car Parking Booking System Interface for Multi-

Location Slot Selection and Availability

Figure 3 depicts the user interface of a car parking

booking system, designed to allow users to select a

parking location and view the availability of parking

slots. The interface includes two main sections, each

representing a different location (Location L1 on the

left and Location L3 on the right). Each section

Enhanced Face Recognition Algorithm for Real-Time Applications with Improved Accuracy

217

displays a form for selecting a location and a parking

slot, along with a "Park" or "Unbook" button. Red

Slots: represents occupied parking spaces. An icon of

a car is displayed within each red slot to indicate that

it is currently in use. Green Slots represents available

parking spaces, which users can select for booking.

Figure 4 shows the precision, Sensitivity, and

Specificity Comparison by Method, with grouped

bars representing each metric for the algorithms. In

this chart, the Proposed Method consistently

outperforms other methods across all three metrics:

precision, sensitivity, and specificity, indicating its

balanced and robust performance in face recognition

tasks. Table I presents the performance indicators—

specificity, sensitivity, accuracy, and precision —for

four face recognition algorithms: Dense U-Nets,

Retina Face, Hierarchical Network, and the Proposed

Method. The Proposed Method demonstrates the

highest values across all metrics, indicating superior

performance in face recognition tasks.

Figure 4: Performance Comparison of Face Recognition

Algorithms

Table 1: Performance Metrics of Face Recognition

Algorithms

Method Accura

Precisi

Sensitivi

Specifici

Dense U-

Nets

81.43% 0.82 0.79 0.83

Retina

Face

83.87% 0.85 0.84 0.86

Hierarchi

cal

Networ

87.36% 0.88 0.87 0.89

Proposed

Metho

89.50% 0.90 0.89 0.91

Figure 5: Accuracy Comparison of Face Recognition

Algorithms

Figure 5 shows the Accuracy Comparison by

Method, where each algorithm's accuracy is

displayed as a percentage. The Proposed Method

exhibits the highest accuracy at 89.50%, followed by

the Hierarchical Network at 87.36%, Retina Face at

83.87%, and Dense U-Nets at 81.43%. This indicates

the overall improved performance of the Proposed

Method over existing approaches.

5 CONCLUSION

The comparative analysis of face recognition

algorithms demonstrates that the Proposed Method

outperforms the Dense U-Nets, Retina Face, and

Hierarchical Network approaches across all key

metrics: accuracy, precision, sensitivity, and

specificity. With an accuracy of 89.50% and

consistently higher precision, sensitivity, and

specificity values, the Proposed Method proves to be

more reliable and effective for face recognition tasks.

These results suggest that the enhancements

introduced in the Proposed Method lead to more

accurate detection and classification, making it a

suitable choice for uses necessitating exact

measurements and reliability, such as security

systems and identity verification solutions. Future

work could focus on further optimizing the Proposed

Method for faster real-time processing and testing it

across diverse datasets to enhance its robustness and

adaptability in various real-world environments.

REFERENCES

Shetty AB, Rebeiro J. Facial recognition using Haar

cascade and LBP classifiers. Global Transitions

Proceedings. 2021 Nov 1;2(2):330-5.

INCOFT 2025 - International Conference on Futuristic Technology

218

Li N, Shen X, Sun L, Xiao Z, Ding T, Li T, Li X. Chinese

face dataset for face recognition in an uncontrolled

classroom environment. IEEE Access. 2023 Aug 7.

Wattamwar S, Mate R, Rainchwar P, Mantri S, Sorate G.

Optimal Face Recognition System using Haar

Classifier. In2021 International Conference on Smart

Generation Computing, Communication and

Networking (SMART GENCON) 2021 Oct 29 (pp. 1-

7). IEEE.

Neto PC, Pinto JR, Boutros F, Damer N, Sequeira AF,

Cardoso JS. Beyond masks: On the generalization of

masked face recognition models to occluded face

recognition. IEEE Access. 2022 Aug 16;10:86222-33.

Javed Mehedi Shamrat FM, Majumder A, Antu PR,

Barmon SK, Nowrin I, Ranjan R. Human face

recognition applying haar cascade classifier.

InPervasive Computing and Social Networking:

Proceedings of ICPCSN 2021 2022 (pp. 143-157).

Springer Singapore.

Bharathiraja N, Sakthivel M, Deepa T, Hariprasad S,

Ragasudha N. Design and implementation of selection

algorithm based human emotion recognition system.

In2023 7th International Conference on Trends in

Electronics and Informatics (ICOEI) 2023 Apr 11 (pp.

1348-1353). IEEE.

Huang Z, Zhang J, Shan H. When age-invariant face

recognition meets face age synthesis: a multi-task

learning framework and a new benchmark. IEEE

Transactions on Pattern Analysis and Machine

Intelligence. 2022 Oct 28;45(6):7917-32.

Saleem S, Shiney J, Shan BP, Mishra VK. Face recognition

using facial features. Materials Today: Proceedings.

2023 Jan 1;80:3857-62.

Alankar B, Ammar MS, Kaur H. Facial emotion detection

using deep learning and Haar Cascade Face

Identification algorithm. InAdvances in Intelligent

Computing and Communication: Proceedings of ICAC

2020 2021 (pp. 163-180). Springer Singapore.

Deepa T, Hariprasad S, Bharathiraja N, Chokkalingam A.

A Real Time Face Mask Detection and Health Status

Monitoring using Deep Learning after Pandemic.

In2022 International Conference on Augmented

Intelligence and Sustainable Systems (ICAISS) 2022

Nov 24 (pp. 430-435). IEEE.

Zhu Z, Huang G, Deng J, Ye Y, Huang J, Chen X, Zhu J,

Yang T, Du D, Lu J, Zhou J. Webface260M: A

benchmark for million-scale deep face recognition.

IEEE Transactions on Pattern Analysis and Machine

Intelligence. 2022 Apr 26;45(2):2627-44.

Mohammed MA, Zeebaree DQ, Abdulazeez AM, Zebari

DA, Fadhil ZD, Ahmed FY, Rashed EM. Machine

learning algorithm for developing classroom

attendance management system based on haar cascade

frontal face. In2021 IEEE Symposium on Industrial

Electronics & Applications (ISIEA) 2021 Jul 10 (pp. 1-

6). IEEE.

Chowdhury MI, Sakib NM, Masum Ahmed SM, Zeyad M,

Walid MA, Kawcher G. Human face detection and

recognition protection system based on machine

learning algorithms with proposed AR technology.

Advances in Augmented Reality and Virtual Reality.

2022:177-92.

Hangaragi S, Singh T, Neelima N. Face detection and

Recognition using Face Mesh and deep neural network.

Procedia Computer Science. 2023 Jan 1;218:741-9.

Badr IS, Radwan AG, El-Rabaie ES, Said LA, El Banby

GM, El-Shafai W, Abd El-Samie FE. Cancellable face

recognition based on fractional-order Lorenz chaotic

system and Haar wavelet fusion. Digital Signal

Processing. 2021 Sep 1;116:103103.

Srivastava SK, Katiyar S, Kumar S. Pattern matching using

face recognition system. InSoft Computing: Theories

and Applications: Proceedings of SoCTA 2020,

Volume 1 2022 (pp. 161-171). Springer Singapore.

Padmashree G, Karunakar AK. Improved LBP face

recognition using image processing techniques.

InInformation and Communication Technology for

Competitive Strategies (ICTCS 2021) ICT:

Applications and Social Interfaces 2022 Jun 23 (pp.

535-546). Singapore: Springer Nature Singapore.

Pandey N, Yadav PK, Arjun KP. Face recognition system

using computational algorithms. In2022 2nd

International Conference on Advance Computing and

Innovative Technologies in Engineering (ICACITE)

2022 Apr 28 (pp. 1739-1744). IEEE.

Yadav A, Sharma A, Yadav SS. Attendance Management

System Based on Face Recognition Using Haar-

Cascade. In2022 2nd International Conference on

Advance Computing and Innovative Technologies in

Engineering (ICACITE) 2022 Apr 28 (pp. 1972-1976).

IEEE.

Bairagi R, Ahmed R, Tisha SA, Sarder MS, Islam MS,

Islam MA. A real-time face recognition smart

attendance system with haar cascade classifiers. In2021

third international conference on inventive research in

computing applications (ICIRCA) 2021 Sep 2 (pp.

1417-1425). IEEE.

Enhanced Face Recognition Algorithm for Real-Time Applications with Improved Accuracy

219