Automated Attendance System Based on Facial Recognition using Viola-Jones Algorithm

Asian Journal of Applied Science and Technology (AJAST)
Volume 6, Issue 2, Pages 82-91, April-June 2022
ISSN: 2456-883X www.ajast.net
82
Automated Attendance System Based on Facial Recognition using Viola-Jones Algorithm
N. Dhana Lakshmi1
, Pranavi Nagubandi2
, Muralidhar Yeleti3
& K.Vishnu Vardan4
1
Assistant Professor, Department of ECE, CBIT, Hyderabad, Telangana, India.
2-4
UG Students, Department of ECE, CBIT, Hyderabad, Telangana, India.
DOI: http://doi.org/10.38177/ajast.2022.6210
Copyright: © 2022 N.Dhana Lakshmi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Article Received: 22 February 2022 Article Accepted: 26 April 2022 Article Published: 30 May 2022
1. Introduction
Automated Time and Attendance marking system can help schools and higher education in many ways. There is no
doubt that an attendance management system will surely save lot of time and money by eliminating a great deal of
manual processes involved in attendance and leave entry and calculating hours attended. Since the attendance is
automated, the data is accurate and error-free. Once the attendance is marked, the captured data gets stored in the
student attendance system. This feature is especially useful while locating a particular student or while analysing
trends. A face recognition system is a computer application capable of identifying or verifying a person from a
digital image. One of the ways to do this is by comparing selected facial features from the image and a face
database. Automated Attendance System is the advancement that has taken place in the field of automation
replacing traditional attendance marking activity.
Generally an Attendance Management System which is developed using bio-metrics, but in our case we collect the
facial data base using face detecting algorithm and Classifying them using Eigen faces approach.
2. Literature Survey
Henry A. Rowley, Shumeet Baluja, and Takeo Kanade have developed neural network-based approach to detect
frontal views of faces in gray-scale photographs. The algorithm and training methods are ubiquitous, and may be
used to recognise various types of faces as well as related objects and patterns. This saves the time-consuming task
of manually selecting non-face training examples, which must be picked to cover the full non-face picture space.
They discovered that the technology can detect 90.5 percent of faces.
Henry Schneiderman and Takeo Kanade described a statistical method for 3D object detection. They used a product
of histograms to express the statistics of object and non-object appearance. The combined statistics of a selection of
ABSTRACT
Facial Recognition is the most used technology nowadays. Apart from Bio-metrics, Iris scan, fingerprint recognition methodologies, facial
recognition is emerging recognition methodology these days. One of the most effective applications of this methodology is automated attendance
using facial recognition, which is contact less, secure, and effective unlike in tradition way (manual attendance) it saves more time. Methodology
used in this project involves Viola-Jones algorithm for face detection and Eigenfaces approach for feature selection and classification. In Viola-jones
algorithm inputs are taken as captured images of individual persons and produce a dataset containing cropped images of individual and these dataset
is directed to Eigenfaces approach as input and training of data occurs through the process of calculating eigen vectors for each eigenface. At the time
of testing, Euclidean distance between eigen vectors of testing image and eigen vectors of trained eigen faces determines the matched individual.
Facial recognition can also be done with PCA, which has 79.6 percent accuracy, and LBPH, which has 90.23 percent accuracy. However, when
employing the Eigenfaces technique, the accuracy is 93.07 percent. MATLAB software with Computer Vision Toolbox and Deep Learning Toolbox
is used for this work.
Keywords: Viola-Jones algorithm, Eigenfaces, Eigen vectors, Facial recognition, Recognition accuracy.

83
wavelet coefficients and their position on the object are represented by each histogram. They used a variety of
histograms to demonstrate a wide range of visual qualities. They built the first algorithm that can consistently
recognise human faces with out-of-plane rotation using this model. They used a test set of 208 photos containing
441 faces, 347 of which are in profile view, and the greatest accuracy observed was around 92.7 percent.
Paul Viola and Michael Jones developed a frontal face identification system that was completely different from
earlier systems in terms of its ability to recognise faces quickly. This method employs a new picture representation
known as an integral image, which aids in the rapid evaluation of features. A method for building a classifier that
involves utilising AdaBoost to choose a small number of significant features and adding gradually more composite
classifiers in a cascade structure to rapidly boost the detector's speed. Finally, the detection rate was discovered to
be 90%.
Ajay Kumar Bansal and Pankaj Chawla demonstrated the concept of the performance of face recognition using the
Principal Component Analysis (PCA) technique. The tests were done on the Otorhinolaryngology (ORL) database,
the Indian face database, and the Georgia Tech face database, which all have different expressions, poses, and facial
characteristics. In the experimental setup, the number of training photos for all databases was altered from 80% to
40%, which means that originally 80% of total images were used in training and 20% for testing, and then the ratio
was varied as 60/40, then 40/60. The comparison clearly reveals that PCA accuracy is 92.50 percent, 74.17 percent,
and 61.25 percent.
Mrunmayee Shirodkar, Varun Sinha, Urvi Jain, and Bhushan Nemade have presented a system for students that
include features such as face detection, feature extraction, feature detection, attendance analysis, and monthly
attendance report production. For face detection, the suggested system uses image contrasts, integral images,
Ada-Boost, Haar-like features, and a cascade classifier. Faces are recognised using an improved Local Binary
Pattern (LBP) that is utilised to recognise student faces utilising a database comprising photos of students. That
technology assisted them in achieving outcomes with a higher accuracy of 88.08 percent and less time consumption
thanks to a distinct mix of algorithms.
Suman Kumar Bhattacharyya and Kumar Rahul used LDA for Facial Recognition. They've tried using the ORL
face database. By using a linear discriminant criterion, the LDA approach overcomes the limitations of the
Principle Component Analysis (PCA) method. This criterion aims to maximise the ratio of the determinant of the
projected samples' between-class scatter matrix to the determinant of the projected samples' within-class scatter
matrix. The final experimental findings show a 92.5 percent recognition rate or correctness.
3. Face Recognition System
This work has three main steps:
(1) Building database using Viola Jones Algorithm.
(2) Feature Extraction using Eigen faces and Classification using Euclidean distance.
(3) Updating Attendance.

84
3.1. Building database using Viola-Jones Algorithm
First step of our project is to build database i.e. obtain only face from image. In simple terms we need a face locator.
Basically a face locator needs to tell whether a photo of optional size contains a human face and give if that is valid
or not and also it should tell where the face is located in the image. For this, we are using Viola Jones algorithm
which is named after two Computer Vision Researchers Paul Viola and Michael Jones. They proposed this method
in 2001 in their paper “Rapid Object Detection using a Boosted Cascade of Simple Features”.
This algorithm works well for frontal faces rather than faces looking upwards, sideways or downwards. First the
image is converted into a grayscale because it is easier to work with gray scale images rather than colored and there
will be lesser data to process with gray-scale images. This algorithm looks at many smaller sub-regions and tries to
find a face by looking for specific features in each sub-region. It needs to check many different positions and scales
because an image can contain many faces of various sizes.
Fig.1. Detection of Haar features
As you can see in above figure Viola-Jones outlines a box and searches for a face within the box. It is essentially
searching for these Haar-like features. The box moves a step to the right after going through every tile in the picture.
As you can see if face features are detected the box turns green else it is red.. Generally we can change the box size
and step size according to our needs. If we take smaller steps, then many number of boxes detect face-like features
(Haar-like features) and the data of all those boxes when aggregated helps the algorithm to determine where the
face is located in image.
Fig.2. Block diagram for building database using Viola-Jones

85
Fig.3. Database for each user
Each folder has consists of c number of images for each user.
3.2. Feature Extraction using eigenfaces and Classification using Euclidean distance
Here we create a folder for each user and store the facial images of that user in that folder. We have created 10
folders and named them as s1, s2, and so on upto s10. Each folder has 150 images of that particular user. These 150
images are of different pose. For example, s1 consists of 150 facial images of person 1 and similarly s2 contains
150 images of person 2. These 150 images are later used for testing and training. These facial images are collected
using Viola Jones algorithm. Each image should be of same size. We have to take the size of image as 92 pixels
width and 112 pixels height. The output of the detection part will be like given in the below figure.
Fig.4. Block diagram for Recognition and Updation

86
Fig.5. Flowchart
Here we create a folder for each user and store the facial images of that user in that folder. We have created 10
folders and named them as s1, s2, and so on upto s10. Each folder has 150 images of that particular user. These 150
images are of different pose. For example, s1 consists of 150 facial images of person 1 and similarly s2 contains
150 images of person 2. These 150 images are later used for testing and training. These facial images are collected
using Viola Jones algorithm. Each image should be of same size. We have to take the size of image as 92 pixels
width and 112 pixels height. The output of the detection part will be like given in the below figure.
Fig.6. Output of Face detection
3.2.1. Extracting Eigen values
Choose an image of face I (x,y) in two dimensional N *N arrangements. Then convert the image into the dimension
of N2
*1. Treat initially a training set of N N* images and it can be converted into N2
*1 dimensions. Now a training
set of N2
*M dimensions is created. Here the number of samples is M . Let 1, 2 ,3, …..M be the face images in

87
the training set. Next, compute the mean for the images in the training set. At that point, average face of the set is
defined by,
ψ =
∑ 
ⁱ (1)
Here, i is the Normalized Image Coefficient introduced to improve verification chances even for lowest matching
of random image with Eigen face image and is well-defined as square root of the sum of the squares of difference
between images and average face image divided by i.
ᶲⁱ =√∑

, i=1,2,3,…..M (2)
By using the results of the above step the matrix A is computed.
i.e., A 1 2 .....M
The matrix C (co-variance matrix) is formed by multiplying A by its transpose.
C=A*A T
(3)
Therefore the “dimension of the matrix C is N2
* N2
and determining the N2
Eigen vectors and Eigen values is an
intractable task for typical image sizes”. In the event that the quantity of information focuses in the picture space is
not exactly the element of the space (M<N2
). There will be just M 1 , as opposed to N2
important Eigen vectors. In
this way, consider a matrix Q=AT
A of dimension N*N . Next compute the Eigen values and Eigen vectors of the
matrix Q . Consider Eigen values i and Eigen vectors vi of Q such that
AT
Avi= i vi
A(AT
Avi )=iAvi
C.Avi=iAvi
Therefore, the Eigen vectors of co-variance matrix C is Avi
Express the Eigen vectors vi as a linear combination of M training set face images to compute the Eigen faces ui,
Where,
ui=∑ ikk ,i=1,2,3,…M, in particular
u1=v111+v122+v133+……+v1MM
u2= v211+v222+v233+……+v2MM
. . . . .
. . . . .
uM= vM11+vM22+vM33+……+vMMM
After acquiring the Eigen faces, the photographs in the database are predicted into the Eigen faces space, and loads
of the pictures in that space are stored.

88
Fig.7. Original images and corresponding Eigen Faces
The weights of the image on the Eigen face space are calculated by using the formula,
Wi ui
T
() (4)
To recognize the face first calculate the Euclidean distance between the Eigen face of the image and Eigen faces
stored previously.
For an unknown image, calculate  and ^= ∑ wiui where wi =ui
T
. Finally calculate the Euclidean distance
between  and ^.
If the Euclidean distance is minimum and it is below the threshold value, then the unknown image will be
identified. Otherwise the unknown image will not be recognized.
In all of the above cases i varies from 1,2,3,…..,n. We have done deep investigation and examination of key
features, which will play primal role in framing concluding the results of image identification over the data base.
The distinct image processing formulae has been framed and tested with small sample of images. Finally we
concluded last one is the best for identification of random image, even it is having very less common features with
mean/original image. So we concentrated and continued on the last one i.e. to prove the image metrics using this
parameter . This will surly give greater Euclidean distance and Mean square error in order to match the random
image with Eigen face image. Random image is the image of particular person at any moment. So using this
formula, it is easy to trim down the error values like Euclidean distance and Mean square error.
Fig.8. Block Diagram for Calculation of Eigen Faces

89
3.3. Updating Attendance
After recognizing the face, the last step is to update the attendance in the excel sheet in the corresponding column
based on the class number.
4. Performance Metrics
Confusion matrix can be defined as “visualizing the performance of your prediction model in tabular form”. Each
entry in a confusion matrix implies the number of predictions made by the model whether the classification is done
correctly or incorrectly. Classification based on alone accuracy can be misleading if you have an unequal number of
observations in each class or if we have more than two classes in our dataset.
True Positive (TP):
It is concerned with the amount of times the classifier correctly predicts the positive class to be positive.
True Negative (TN):
It is concerned with the amount of times a classifier correctly predicts a negative class as negative.
False Positive (FP):
It is concerned with the number of times the classifier has predicted the negative class as positive.
False Negative (FN):
It is concerned with the amount of times the classifier incorrectly forecasts the positive class as negative.
Confusion matrix should always be used as an evaluation criterion for machine learning models. It provides a model
performance metric that is both simple and effective. Here are some of the most prevalent confusion matrix-derived
performance measures.
Accuracy: It is the percentage of total samples correctly categorised by the classifier. Use the formula
(TP+TN)/(TP+TN+FP+FN) to calculate accuracy.
Misclassification Rate: It gives the information about what fraction of predictions were incorrect. It is also known
as Classification Error. You can calculate using (FP+FN)/(TP+TN+FP+FN) or (1-Accuracy).
5. Results & Conclusion
In this approach, automated attendance system using facial recognition is described and proposed model used
Viola-Jones for face detection and to build our database which is used for further process. And using Eigen Faces,
features are extracted and using Euclidean distance as classifier, input face is recognized and finally updated the
attendance sheet accordingly. The proposed model has a database with 10 users, and for each user model has taken
150 images. These 150 images of each user are used for both training and testing. And 80% of data is used for
training and 20% for testing. So, for each user 120 images are taken for training and 30 images for testing. And the
system has considered 5 users and used their data present in database for calculating TP, TN, FP, FNs.
Face recognition: This is one of the matched results in which input image is captured instantly and the output
image is recognized from the training dataset.

90
Fig.9. Correctly, Incorrectly recognized result and Confusion chart
Updating Attendance: This is finally updated automated attendance excel sheet in which, we can get total number
of present days for each student and can analyse the attendance percentage for each student.
Fig.10. Excel sheet before updating
Fig.11. Excel sheet after updating
Declarations
Source of Funding
This research did not receive any grant from funding agencies in the public, commercial, or not-for-profit sectors.
Competing Interests Statement
The authors declare no competing financial, professional and personal interests.

91
Consent for publication
Authors declare that they consented for the publication of this research work.
References
[1] Ajay Kumar Bansal, Pankaj Chawla, (2013). Performance evaluation of face recognition using PCA and
N-PCA. International Journal of Computer Applications, 76(8): 14-20.
[2] Paul Viola, Michael J. Jones, (2014). Robust real-time face detection. International Journal of Computer Vision,
57(2): 137-154.
[3] Akash Singh, Shreya Bhatt, Abhishek Gupta, (2021). Automated attendance system with face recognition.
International Journal of Engineering Applied Sciences and Technology, 5(12).
[4] V. Jalaja, G.S.G.N. Anjaneyulu, (2020). Face recognition by using eigen face method. International Journal of
Scientific & Technology Research, 9(3): 961-966.
[5] Jaspreet Kaur, Anand Sharma, (2017). Performance analysis of face detection by using viola-jones algorithm.
International Journal of Computational Intelligence research, 13(5): 707-717.
[6] Ashish Choudhary, Abhishek Tripathi, Abhishek Bajaj, Mudit Rathi and B.M Nandini, (2016). Automatic
attendance system using face recognition. International Journal of Modern Trends in Engineering and Research,
3(4): 362-365.
[7] Marijeta Slavkovic, Dubravka Jevtic, (2012). Face recognition using eigenface approach. Serbian Journal of
Electrical Engineering, 9(1): 121-130.

Automated Attendance System Based on Facial Recognition using Viola-Jones Algorithm

More Related Content

Similar to Automated Attendance System Based on Facial Recognition using Viola-Jones Algorithm (20)

More from Associate Professor in VSB Coimbatore (20)

Recently uploaded (20)

Automated Attendance System Based on Facial Recognition using Viola-Jones Algorithm