Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Real Time Face Attendance System

using Deep Learning
Bala krishnan.V., Sinraj. S., Abinesh. R., Gowtham. R.
UG students
Department of Computer Science and Engineering
Paavai Engineering College,
Pachal, Namakkal

Abstract:- Face is an individual's unique representation, distractions faced by students during exam sessions have
and therefore, we propose an automated system for become a significant issue. In addition to verbally calling
student attendance using face recognition. Face out names, the traditional method of passing around an
recognition systems have significant applications, attendance sheet in the classroom is also disruptive. This
particularly in security control systems. For instance, the problem is particularly challenging in large lecture classes
airport protection system relies on face recognition to were passing around the attendance sheet becomes
identify potential suspects, while the Federal Bureau of impractical. To address these challenges, we propose the
Investigation (FBI) utilizes this technology for criminal implementation of a face recognition student attendance
investigations. Our proposed approach begins with video system. This automated system will eliminate the need for
framing, initiated through a user-friendly interface. By manual signing, which is not only burdensome but also
employing the Viola-Jones algorithm, we detect and leads to student distractions. Moreover, the face recognition
segment the region of interest (ROI) containing the face technology employed in this system will help mitigate
from the video frame. In the preprocessing stage, we fraudulent approaches, and lecturers will no longer have to
perform image scaling as necessary to preserve repeatedly count the number of students to ensure their
information integrity. Next, we apply median filtering to presence.
eliminate noise and convert color images to grayscale. To
enhance image contrast, we implement contrast-limited The challenges related to facial identification have
adaptive histogram equalization (CLAHE). In the face been highlighted in previous studies. One such challenge is
recognition stage, we utilize enhanced local binary distinguishing between known and unknown images [1].
pattern (LBP) and principal component analysis (PCA) Furthermore, [2] discovered that the training process for
to extract facial image features. Subsequently, we record face recognition student attendance systems can be slow and
the attendance of the recognized student, saving the data time-consuming. Additionally, [3] pointed out that
in an Excel file. Unregistered students have the variations in lighting conditions and head poses can
opportunity to register on the spot, and notifications are significantly impact the performance of face recognition-
triggered if a student signs in more than once. The based student attendance systems.
recognition accuracy is 100% for high-quality images,
94.12% for low-quality images, and 95.76% for the Yale Therefore, it is imperative to create a real-time student
face database when training with two images per person. attendance system that operates within specific time
constraints to ensure accurate identification without any
Keywords:- Face recognition system, Median filtering, omissions. The extracted facial image features, which serve
CLACHE, LBP, andPCA Image Extraction. as the students' identity representation, must remain
consistent despite changes in background, lighting
I. INTRODUCTION conditions, pose, and facial expressions. The performance of
the system will be evaluated based on its high accuracy and
The main objective of this project is to develop face fast computation time.
recognition based automated student attendance system. To
achieve better performance, the test images and training A. Objectives
images of this proposed approach are limited to frontal and The goal of this project is to create an automated student
upright facial images that consist of a single face only. The attendance system based on face recognition. The expected
test images and training images must be captured by using outcomes to achieve these objectives are as follows:
the same device to ensure no quality difference. In addition,  To detect the face segment from the video frame.
the students must register in the database to be recognized.  To extract the useful features from the face detected.
The enrolment can be done on the spot through the user-  To classify the features to recognize the face detected.
friendly interface.  To record the attendance of the identified student.
Traditional student attendance marking technique is
often facing a lot of trouble. The face recognition student
attendance system emphasizes its simplicity by eliminating
classical student attendance marking technique such as
calling student names or checking respective identification
cards. The disturbances during the teaching process and

IJISRT23MAY252 www.ijisrt.com 450

 Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. EXISTING SYSTEM Viola-Jones algorithm was recommended by [9] and [10] for
face detection in student attendance systems. These
Currently, face recognition systems have gained researchers concluded that among the different approaches
popularity owing to their ease of use and impressive such as face geometry-based methods, feature invariant
performance. For example, face recognition technology is methods, and machine learning-based methods, the Viola-
employed in airport protection systems and by the FBI for Jones algorithm stands out for its speed, robustness, high
criminal investigations, enabling them to track suspects, detection rate, and ability to perform well in different
locate missing children, and combat drug-related activities lighting conditions.
[4]. Similarly, popular social networking platform Facebook
utilizes face recognition to facilitate user-friendly features We can also propose that Viola-Jones algorithm is the
like tagging friends in photos for entertainment purposes [5]. most efficient among all algorithms, for instance the
Additionally, Intel Company provides users with the ability AdaBoost algorithm, the Float Boost algorithm, Neural
to access their online accounts through face recognition [6]. Networks, the S-AdaBoost algorithm, Support Vector
Apple also incorporates face recognition technology in their Machines (SVM) and the Bayes classifier.
iPhone X, allowing users to unlock their mobile phones [7].
C. PRE-PROCESSING
A. STUDENT ATTENDANCE SYSTEM Pre-processing plays a crucial role in enhancing the
The drawbacks of RFID (Radio Frequency performance and accuracy of the face recognition system.
Identification) card systems, fingerprint systems, and iris Scaling is an important step in pre-processing as it allows
recognition systems have been discussed. The RFID card for manipulation of the image size. Scaling down an image
system is chosen for its simplicity, but it is susceptible to can significantly increase the processing speed by reducing
misuse as users may assist their friends in checking in using the computational load, as it decreases the number of pixels
their ID cards. The fingerprint system, while effective, is not [4]. The size and pixel count of an image contains valuable
efficient as it requires individuals to line up for verification spatial information, which refers to the smallest
one by one, leading to time-consuming processes. On the distinguishable details within an image [4]. Therefore, it is
other hand, face recognition systems utilize the human face, important to handle spatial information carefully during
which is always visible but contains less information manipulation to avoid distortions and prevent artifacts like
compared to iris recognition Although iris recognition the checkerboard effect. For normalization and
provides more detailed information, it raises concerns about standardization, it is preferred to have images of the same
user privacy. Voice recognition is an alternative, but it is size [6]. In their work, [6] proposed using Principal
less accurate compared to other methods. Therefore, it is Component Analysis (PCA) to extract features from facial
suggested to implement a face recognition system for the images and preferred images to be scaled to 120x120 pixels
student attendance system to overcome these limitations [8]. in order to maintain consistent length and width.

Table 1: Advantages& Disadvantages of Different Existing In addition to image scaling, it is common practice to
System convert color images to grayscale during pre-processing.
System type Advantages Disadvantage Grayscale images are known to be less sensitive to
RFID card system Simple Fraudulent usage variations in illumination and require less computational
Fingerprint system Accurate Time-consuming time. Grayscale images are represented as 8-bit images, with
Voice recognition System _ Less accurate pixel values ranging from 0 to 255, while color images are
compared to others represented as 24-bit images, with pixels capable of having
Iris recognition System Accurate Privacy Invasion 16,777,216 different values. Consequently, color images
necessitate more storage space and computational power
B. FACE DETECTION compared to grayscale images.
There is often confusion between the concepts of face
detection and face recognition. Face detection specifically When color information is not essential for the
aims to locate and isolate the face segment or region within computation, it is considered noise and can be discarded.
an image. On the other hand, face recognition goes beyond Moreover, pre-processing is crucial for enhancing the
detection and focuses on identifying the owner or individual contrast of images. One commonly used method for pre-
depicted in the facial image. Various factors can contribute processing is histogram equalization, which aims to improve
to the challenges faced in both face detection and face image contrast. By providing a uniform distribution of
recognition processes. These factors include background intensities across the intensity level axis, histogram
variations, differences in illumination, changes in pose, equalization can reduce the impact of uneven illumination.
facial expressions, occlusions, rotations, scaling, and
D. FEATURE EXTRACTION
translations. Let us now define each of these factors in more
The feature represents the information contained within
detail.
an image and is crucial for face recognition. However,
Previous researchers have conducted various studies selecting appropriate features can be a challenging task. A
on face detection methods. However, many of these studies feature extraction algorithm needs to be consistent and
focused on utilizing frontal upright facial images containing stable across various changes to achieve high accuracy
a single face. These images typically depict fully exposed results. Principal Component Analysis (PCA) is a widely
face regions without any obstructions or glasses [9][10]. The recognized method known for its robustness and high-speed

IJISRT23MAY252 www.ijisrt.com 451

Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
computation. Essentially, PCA captures data variations and
removes unnecessary correlations among the original
features. It acts as a dimension reduction algorithm,
compressing each facial image, represented by a matrix, into
a single column vector.Additionally, PCA centralizes the
image data by subtracting the average value from each
image. The Principal Components of the distribution of
facial images are referred to as Eigenfaces. Each individual
facial image from the training set contributes to the creation
of Eigenfaces, encoding the most significant variations
among the known facial images. To perform recognition,
both the training and test images are projected onto the
Eigenface space, resulting in projected training images and
Fig. 2: Class Separation in LDA
projected test images, respectively. Recognition is then
carried out by computing the Euclidean distance between III. ALGORITHMS USED
the projected training images and the projected test image.
It’s important to note that PCA feature extraction involves Neural networks were initially used only in face
all trained facial images, and as a result, the extracted detection. It is then further studied to be implemented in
features retain correlations among the facial images in the face recognition.
training set.
A. ARTIFICAL NEURAL NETWORK
Artificial Neural Network (ANN) was studied for face
recognition. ANN consists of the network of artificial
neurons known as "nodes". The nodes act as human brain to
make recognition and classification. These nodes are
interconnected, and values are assigned to determine the
strength of their connections. High value indicates strong
connection. Neurons were categorized into three types of
nodes or layers which are input nodes, hidden nodes, and
output nodes. Input nodes are given weight based on its
impact. Hidden nodes consist of some mathematical
function and thresholding function to perform prediction or
probabilities that determine and block unnecessary inputs
and result is yield in output nodes. Hidden nodes can be
Fig. 1: PCA Dimension Reduction
more than one layer. Multiple inputs generate one output at
LDA (Linear discriminant analysis), also known as the output node.
Fisher face, is a widely used algorithm for face recognition.
B. CONVENTIONAL NEURAL NETWORK
In [10], LDA was introduced as a method for face
Convolutional Neural Network (CNN) is another neural
recognition. LDA extracts features by grouping images of
network algorithm for face recognition. Like ANN, CNN
the same class and differentiating them from images of
consists of the input layer, hidden layer and output layer.
different classes. One notable characteristic of LDA is its
Hidden layers of a CNN consist of multiple layers which are
ability to perform well even in the presence of variations in
convolutional layers, pooling layers, fully connected layers
facial expressions, illumination, and pose, thanks to its class
and normalization layers. However, a thousand or millions
separation capability.In LDA, images belonging to the same
of facial images have to be trained for CNN to work
class are defined as those of the same individual but with
accurately and it takes long time to train, for instance Deep
different facial expressions, lighting conditions, or pose. On
face which is introduced by Facebook.
the other hand, facial images of different individuals are
categorized as different classes. The within-class scatter IV. METHODOLOGY
matrix is computed based on images from the same class,
while the between-class scatter matrix is computed based on The proposed approach entails a face recognition-
images from different classes. LDA aims to maximize the based student attendance system. The methodology follows
ratio of the determinant of the between-class scatter matrix a sequence of steps, starting with the capture of images
to the determinant of the within-class scatter matrix. It is using a user-friendly interface. Subsequently, the captured
believed that LDA achieves lower error rates compared to facial images undergo pre-processing, feature extraction,
PCA when there are more samples per class in the training subjective selection, and classification for recognition. In
set and when the number of different classes is relatively this approach, both Local Binary Pattern (LBP) and
small. Principal Component Analysis (PCA) feature extraction
methods are extensively studied and applied to facilitate
comparisons. To mitigate the impact of illumination, LBP is
enhanced within this approach. Additionally, an algorithm is
devised to combine the enhanced LBP and PCA for

IJISRT23MAY252 www.ijisrt.com 452

Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
subjective selection, aimed at improving accuracy. The B. LIMITATION OF IMAGE
subsequent sections will delve into the specifics of each The input image for the proposed approach should meet
stage in detail. certain criteria: it should be frontal, upright, and contain
only a single face. While the system is designed to recognize
A. INPUT IMAGES students both with and without glasses, it is important for
Although our own database should be used to design real students to provide facial images with and without glasses
time face recognition student attendance system, the during training to improve accuracy in recognizing them
databases that are provided by the previous researchers are without glasses. To ensure consistency in image quality,
also used to design the system more effectively, efficiently both the training and testing images should be captured
and for evaluation purposes. using the same device. Additionally, students must register
in order to be recognized by the system. The registration
Yale face database is used as both training set and process can be conveniently completed through a user-
testing set to evaluate the performance. Yale face database friendly interface.
contains one hundred and sixty-five grayscale images of
fifteen individuals. There are eleven images per individual; C. PRE-PROCESSING
each image of the individual is in different condition. The Testing set and training set images are captured using a
conditions included center-light, with glasses, happy, left- camera. However, these images may contain unwanted noise
light, without glasses, normal, right-light, sad, sleepy, and uneven lighting, which can affect the accuracy of the
surprised and wink. These different variations provided by face recognition system. Therefore, several pre-processing
the database can ensure the system to be operated steps are required to enhance the quality of the images
consistently in a variety of situations and conditions. before proceeding to feature extraction. One of the pre-
processing steps is scaling of the images.

 Scaling of images
Scaling is a common task in image processing, and it
involves carefully adjusting the size of the images to prevent
any loss of spatial information. In the context of face
recognition, it is important to have images of the same size
for accurate feature extraction. In this proposed approach,
the test images and training images are standardized to a
Fig. 3: Sample Images in Yale Face Database size of 250 × 250 pixels to ensure consistency. Other pre-
processing steps include median filtering, conversion of
In our database, student images are captured using color images to grayscale images, and adaptive histogram
both the laptop's built-in camera and a mobile phone equalization. These steps will be discussed in detail in the
camera. Each student provides four images, two for the subsequent sections.
training set and two for the testing set. The images captured
with the laptop's built-in camera are considered low-quality  Median Filtering
images, while the images captured with the mobile phone Median filtering is an effective method for noise
camera are considered high-quality images. The high-quality reduction in images. It is known for its ability to remove
images include seventeen students, while the low-quality unwanted noise while preserving important details. In the
images include twenty-six students. The recognition rates of case of color images captured by a camera, median filtering
the low-quality and high-quality images will be compared to is applied to each of the three-color channels (red, green,
draw conclusions regarding the performance differences and blue). This ensures that noise is effectively removed
between the two sets of images. from all channels, leading to a cleaner image. Figure 4.4
provides a visual comparison of the image before and after
noise removal using median filtering in the three channels.
In the case of a grayscale image, median filtering can be
directly applied without the need for channel separation.

Fig. 4: Sample of High-Quality Images

Fig. 6: Median Filtering Done on Three Channels

Fig. 5: Sample of Low-Quality Images

IJISRT23MAY252 www.ijisrt.com 453

Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. PROJECT RESULT VI. OVERVIEW AND DISCUSSION

In this proposed approach, a user-friendly interface for The proposed approach has been trained and tested on
the face recognition student attendance system is developed various datasets. The Yale face database, containing 165
using MATLAB GUI (Graphic User Interface). The grayscale images of 15 individuals under different
interface incorporates several buttons, each serving a conditions, has been employed. However, to supplement this
specific function. The start button is used to initialize the database and account for color images, we have created our
camera and automatically perform face recognition based on own database. This database includes both high-quality and
the detected face. The register button facilitates the low-quality sets, as the images exhibit varying levels of
enrollment or registration of students, while the update clarity and blurriness. Detailed statistics for each dataset
button is designed to train the system with the latest have been previously discussed in the preceding chapter.
registered images stored in the database. Additionally, the
browse button allows the user to select facial images from a VII. CONCLUSION
designated database, and the recognize button is utilized to
test the system's functionality by recognizing the selected In this approach, a comprehensive description of a face
image.In this section, we have selected and utilized an recognition-based automated student attendance system is
enhanced Local Binary Pattern (LBP) algorithm with a provided. The proposed method aims to identify individuals
radius size of two. The reasoning behind this choice and the by comparing their input image, obtained from a recorded
analysis of the selected radius size will be elaborated on in video frame, with the training image. It includes the ability
the subsequent discussion. to detect and localize faces within an input facial image
from the recorded video frame. Furthermore, the approach
incorporates a preprocessing stage to enhance image
contrast and reduce the impact of illumination. Feature
extraction from the facial image is performed using both
Local Binary Patterns (LBP) and Principal Component
Analysis (PCA).

Regarding the analysis, facial feature extraction can be

particularly challenging, especially in varying lighting
conditions. To address this, Contrast Limited Adaptive
Histogram Equalization (CLAHE) is employed during the
preprocessing stage to effectively reduce the illumination
effects. CLAHE outperforms traditional histogram
equalization in terms of contrast improvement. Additionally,
enhanced LBP with a larger radius size, specifically a radius
size of two, demonstrates superior performance compared to
the original LBP operator. It is less affected by illumination
variations and exhibits greater consistency compared to
other radius sizes. In terms of image enhancement, previous
papers published in 2013 utilized histogram equalization to
improve image contrast, while another paper did not employ
any technique for enhancing image contrast. In this
proposed algorithm, CLAHE is used instead of histogram
equalization to enhance image contrast. While histogram
equalization may provide clearer visualization of bone
structures in X-ray applications, it tends to cause over-
enhancement in some regions while leaving other regions
inadequately enhanced. CLAHE is implemented to
overcome this limitation by preventing over-enhancement
and improving contrast more uniformly across the entire
image. A comparison between CLAHE and histogram
equalization is presented in the previous section.

A. Recommendation
In this proposed approach, there are several limitations
that need to be considered. Firstly, the input image must
meet specific requirements, including being frontal, upright,
and containing only a single face. Secondly, the accuracy of
the system may decrease significantly under extreme
illumination conditions. Thirdly, there is a risk of false
recognition if the captured image is blurred or of low
Fig. 7: Project Result quality. Additionally, it should be noted that LBP is a

IJISRT23MAY252 www.ijisrt.com 454

Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
textural-based descriptor that extracts local features, recognition-payment- technology/ [Accessed 25 Mar.
therefore, to achieve high accuracy, the test and training 2018].
images must have the same quality and be captured using [6.] Sidney Fussell. (2018). NEWS Facebook's New Face
the same device. Another limitation is that if an individual is Recognition Features: What We Do (and Don't)
wearing makeup in the facial image used for recognition, it Know. [online] Available at:
can cover important facial features and potentially lead to https://gizmodo.com/facebooks-new-face-
recognition errors. Furthermore, it would be beneficial to recognition-features-what-we-do-an- 1823359911
design a face recognition system that can handle multiple [Accessed 25 Mar. 2018].
faces in an image, as this would improve the efficiency of [7.] deAgonia, M. (2017). Apple's Face ID [The iPhone
the system. X's facial recognition tech explained]. [online]
Computerworld. Available at:
Lastly, it is important to acknowledge that the https://www.computerworld.com/article/3235140/app
performance of the test image and training image in this le-ios/apples-face-id-the- iphone-xs-facial-
approach is highly influenced by each other and is heavily recognition-tech-explained.html [Accessed 25 Mar.
dependent on the specific device used to capture the images. 2018].
[8.] Arun Katara, Mr. Sudesh V. Kolhe, Mr. Amar P.
B. Future Scope Zilpe, Mr. Nikhil D. Bhele, Mr. Chetan J. Bele.
The previous researcher who published the paper in 2015 (2017). “Attendance System Using Face Recognition
conducted their study using their own image databases. In and Class Monitoring System”, International Journal
contrast, the paper published in 2013 utilized an image on Recent and Innovation Trends in Computing and
database called NITW-database, which consisted of 80 CommunicationV5(2).
individuals with 20 images per person. However, the paper [9.] S.Aanjanadevi, Dr.V.Palanisamy,R.Anandha Jothi.
from 2015 did not specify the size of the image database (2017). A Study on Secure Online Voting System
used in their research. using Biometrics Face Detection and Recognition
In the proposed algorithm, multiple image databases Algorithms. International Journal for Modern Trends
are utilized for training and testing. One of the databases in Science and Technology, V3(8).
used is the Yale face database, which contains images [10.] P. Arun Mozhi Devan et al., (2017). Smart
captured under different lighting conditions and with various Attendance System Using Face Recognition.
expressions. This allows for the evaluation of the proposed Advances in Natural and Applied Sciences. 11(7),
algorithm's performance in scenarios with uneven lighting Pages: 139-144
and a variety of facial expressions. It should be noted that [11.] Wei-Lun Chao. (2007). Face Recognition, GICE,
the Yale face database solely consists of grayscale images National Taiwan University. Akshara Jadhav. (2017).
without any background. Therefore, in order to apply the Automated Attendance System Using Face
proposed algorithm to real-time applications and perform Recognition. International Research Journal of
face recognition, our own database of color images is also Engineering and Technology.V4 (1).
incorporated. [12.] Rahul V. Patil and S. B.Bangar. (2017). Video
Surveillance Based Attendance system. IJARCCE,
REFERENCES 6(3), pp.708-713.
[13.] Mrunmayee Shirodkar. (2015). Automated
[1.] Zhao, W., Chellappa, R., Phillips, P. and Rosenfeld, Attendance Management System using Face
A. (2003). Face recognition. ACM Computing Recognition. International Journal of Computer
Surveys, 35(4), pp.399-458. Applications and International Conference and
[2.] Pooja G.R, et al. (2010). An automated Attendance Workshop on Emerging Trends in Technology.
System Using Image Processing. International [14.] Naveed Khan Balcoh. (2012). Algorithm for Efficient
Journal of Advanced Networking & Applications. Attendance Management: Face Recognition based
[3.] Wagh, P., Thakare, R., Chaudhari, J. and Patil, S. approach.International Journal of Computer Science
(2015). Attendance system based on face recognition Issues, V9 (4), No 1.
using eigen face and PCA algorithms. International [15.] Varsha Gupta, Dipesh Sharma. (2014), “A Study of
Conference on Green Computing and Internet of Various Face Detection Methods”, International
Things. Journal of Advanced Research in Computer and
[4.] Robert Silk. (2017). Biometrics: Facial recognition Communication Engineering), vol.3, no. 5.
tech coming to an airport near you: Travel Weekly. [16.] P. Viola, M. J. Jones. (2004), “Robust Real-Time
[online] Available at: Face Detection”, International Journal of Computer
http://www.travelweekly.com/Travel-News/Airline- Vision 57(2), 137–154.
News/Biometrics-Facial- recognition-tech-coming- [17.] Mekha Joseph et al. (2016). Children's Transportation
airport-near-you [Accessed 25 Mar. 2018]. Safety System Using Real Time Face Recognition.
[5.] Reichert, C. (2017). Intel demos 5G facial- International Journal of Advanced Research in
recognition payment technology | ZDNet. [online] Computer and Communication Engineering V5 (3).
ZDNet. Available at: [18.] Srushti Girhe et al. (2015). Computer Vision Based
http://www.zdnet.com/article/intel-demos-5g-facial- Semi-Automatic Algorithm for face detection.

IJISRT23MAY252 www.ijisrt.com 455

Volume 8, Issue 5, May 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
International Journal on Recent and Innovation
Trends in Computing and Communication V3(2).
[19.] BurakOzen. (2017). Introduction to Boosting
Methodology &Adaboost algorithm. [online]
Available at:
https://www.linkedin.com/pulse/introduction-
boosting- methodology-adaboost-algorithm-burak-
ozen [Accessed 12 Apr. 2018].
[20.] Chris McCormick. (2013). Adaboost Tutorial.
[online] Available at:
http://mccormickml.com/2013/12/13/adaboost-
tutorial/ [Accessed 12 Apr. 2018].

IJISRT23MAY252 www.ijisrt.com 456