See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341316377

Attendance Authentication System Using Face Recognition

Article  in  Journal of Advanced Research in Dynamical and Control Systems · March 2020

DOI: 10.5373/JARDCS/V12SP4/20201599

CITATIONS READS
0 674

1 author:

Dr.Sabeenian R.S
Sona College of Technology
90 PUBLICATIONS   334 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Machine and deep learning in image processing platform View project

tumor segmentation View project

All content following this page was uploaded by Dr.Sabeenian R.S on 20 May 2020.

The user has requested enhancement of the downloaded file.

Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Attendance Authentication System Using

Face Recognition
Dr.R.S. Sabeenian, Head of Department, Department of ECE, Sona College of Technology, Salem, Tamil Nadu, India.
E-mail: sabeenian@sonatech.ac.in
J. Harirajkumar, Associate Professor, Department of ECE, Sona College of Technology, Salem, Tamil Nadu, India.
E-mail: harirajkumar.j@sonatech.ac.in
Lizzie D’ Cruz, Lecturer (SG), ME (Pursuing), Department of ECE, Sona College of Technology, Salem, Tamil Nadu, India.
E-mail: lizzieshajan@gmail.com
Abstract--- Face Recognition, a biometric technology, uses distinguishable facial features to ascertain the presence
of a person, had made a major breakthrough in the field of computer vision. Capturing attendance without human
intervention with utmost secrecy and reliability is the need of the modern era. Attendance plays an important role in
improving teaching learning process, inculcates healthy life habits apart from improving their academic prospects.
The paper proposes implementation of different types of face detection such as haarcascade, histogram of oriented
gradient, convolution neural network and dlib with its comparison. Further, two recognition methods, Convolution
Neural Network (CNN) and local binary pattern histogram (LBPH) have also been implemented. To curb manual
attendance taking, issue of fake or proxy attendance, Raspberry pi based attendance authentication system using face
recognition has been implemented with (LBPH) method. The system captures the image through the camera, detects
and recognized the faces from the image with an accuracy of 95%. If the image matches with the student dataset,
then student’s attendance is entered in the MySQL database. APACHE web server provides application
programming interface to the database. The message is send to parents through Gmail for absent students through
send Mail Transfer protocol. Graphical User Interface using Tkinter has been developed for demonstration of the
system.
Keywords--- Convolution Neural Network, Raspberry pi, MySQL, APACHE Web Server, Graphical User Interface,
TKinter.

I. Introduction
Face recognition is a process, which uses biometrics to ascertain and endorse the presence of the face of a human
by mapping the facial features from the image or video. It compares the given face from the image with the dataset
available to obtain the recognition. The process involved in the face recognition is creation of dataset, detection
encoding and recognition.
Applications of face detection are in surveillance, tracking, photography etc. Dataset can be created through
webcam or can be stored manually. The process of face detection involves finding out the human face from the
image. The face detection algorithm attempts to detect eyes, eyebrows, mouth, nose, nostrils and iris. The facial
regions are summarized and validated to detect the face. Feature extraction is carried out in face encoding. Face
recognition comprises the process of identification and verification of face. In the identification process, the system
ascertains whether the person’s face is available in the database and in the verification process the system checks the
biometric profile available in the face print with the biometric profile of the given face captured live or through a
video. Face detection [1] can be knowledge based methods, feature invariant method, template matching method or
appearance based methods whereas the face recognition methods are, feature from global appearance like principal
component analysis (PCA) and independent component analysis (ICA), and features from local appearance like local
feature analysis or Gabor wavelet method. The other classification of face recognition is based on the similarity
measures with Euclidian distance in neural network. Now a day’s lots of hybrid methods are also developed for
better accuracy. The essential components of face recognition are illustrated with the block diagram given below.

Fig.1: Components of Face Recognition System

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1235
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Face recognition has eclectic application area like schools and college attendance authentication system, access
control, law enforcement, secure payment transaction, security, identification of criminal, unlocking of phone,
diagnoses of diseases, etc.
The second section of the paper illustrates, the different algorithms followed in attendance system with face
recognition. In the third section, various face detectors has been implemented and their comparisons has also been
enumerated. Further, recognition using CNN and LBPH has also been implemented. In the fourth section, classroom
attendance authentication system using face recognition has been proposed using raspberry pi 3 B+. APPACHE web
server is used for storing the MySQL database. Gmail is used for message transfer to parents. Further the confusion
matrix provides the classification report wherein the precision, recall, f1-score and accuracy are obtained on tested
datasets. The fifth section illustrates the results obtained in the face recognition. The concluding section gives the
summary of the proposed work with future plan.

II. Literature Review

The paper [1] proposed a system on different categories of face detection such as knowledge based, feature
based, template matching, and appearance based, and part based with comparison. The paper also mentioned about
the different types of face recognition method and their comparison. Further it is also mentioned that the methods,
PCA, and linear discriminant analysis (LDA) are used for the dimension reduction. PCA is used widely for data
compression as it processes data for noise reduction and redundancy. LDA is offering provision of maximum
distance between different classes. Problems of face recognition, pose problem and lighting in image has been
identified.
The paper [2] proposed a raspberry pi based system for capturing attendance of students using face recognition.
The system uses local binary pattern histogram based method along with controlling door control option. Web server
is included for accessing the attendance remotely. MySQL database is used storing the attendance. Laravel and php
framework is used for creating the web server, for connectivity to database. The system provides a login for every
lecturer for accessing the attendance with face recognition accuracy as 95% on dataset of 11 students.
The system proposed in [3] involved various methods such as viola and jones algorithm, illumination invariant
and principle component analysis. The system compares the face detected with dataset of students to mark the
attendance. In the system, camera captures the image which is then changed to gray scale, quantized using histogram
technique. The different processing steps of the system are enrollment, capturing of image, conversion into gray
scale, histogram normalization, noise filtering and skin classification.
The proposed system [4] has the feature of face detection, facial feature extraction, normalization and quality
score. Further the different parameters of face recognition, pose estimation, sharpness, resolution, and brightness
were also discussed.
In the system [5] proposes an attendance capturing system. The names of recognized faces of students are entered
in the database. The Eigen face method is used for detecting the face. the message is send to the parents for absent
students. The detected face is marked with a circle in the face.
The system proposed in [6] elaborates Viola jones algorithm based face detection which is followed by feature
extraction. Preprocessing is performed by scaling, histogram, equalization. Image is converted into gray scale. PCA,
LDA classifier is used for face recognition. After extraction of features, the same is compared with detected face and
if found similar then the attendance is entered for the recognized face.
The model proposed in the paper [7] used Eigen face and principle component analysis algorithm integrated with
MATLAB GUI. Attendance is obtained in Excel sheet by comparing the dataset with the given image. The system
captures the image of the students and prepares a dataset which is followed by image cropping, resizing, conversion
from RGB to gray level, and image enhancement. In the principle component analysis, image available in the dataset
is converted into multivariate dataset. Through dimension reduction technique, the face structure is decomposed into
orthogonal component called Eigen faces by discarding the not useful information. The PCA technique is used to
extract the facial feature of full frontal image of face. The algorithm prepares training dataset which is converted into
column vectors. Normalization of vector is done to obtain the unique feature of face. Covariance matrix is obtained
for finding Eigen vector, followed by average face vector calculation. Best Eigen faces are selected for representing
image dataset, called weighted dataset.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1236
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

The model proposed in paper [8] uses an improved AlexNet conventional Neural Network and webFace dataset.
In the model, face recognition and RFID card is used for capturing attendance. The campus card is used by the
student to allow only enrolled students in the class and prevent the other personnel to enter in class. After RFID
verification, the feature extraction is done which is compared with the image dataset and present is entered in the
excel sheet against the name of the student.

The system indicated in the paper [9] proposes a face recognition technique which uses an Arduino
microcontroller which compares the image with the available dataset. In this model, if the students go out of the class
then the attendance is put on hold for 15 minutes and attendance confirmation is done only if re-entry in the class is
done within the given time of 15 minutes. The attendance is captured periodically by checking the in-out time.

The system proposed in paper [10] uses three types of face recognition method such as Eigen face, Fisher face
and Local Binary Pattern. The mobile application prepared by the above mentioned models are used for students,
teacher and parents. All the three method has been tested for different Euclidian distance and the accuracy rate is
analyzed.

The system proposed in [11] uses a face recognition model which is integrated as face recognition API with
RFID based system which is used for gathering the video from the camera. RFID reader device is used to validate the
accuracy of face recognition model.

The system depicted in paper [12] employs a modified Viola jones algorithm for face detection and alignment
free face recognition. Attendance in Excel sheet is entered if the face in the image and dataset are similar.

The system proposed [13] consist of four parts such as capturing of image, group segmentation and selection of
faces, comparison and face recognition and attendance updating, in the database. The image is captured through the
camera which is followed by segmentation and face detection using Haar cascade algorithm. Identification of face is
done through histogram values. Attendance is entered in the database if the image in the dataset is found similar to
the image captured by camera.

The paper [14] presents, new and better updates on yolo face detector. New network has been trained for better
accuracy and speed.

III. Comparative Analysis of Face Detection and Recognition

A. Haar Cascade Face detection

Haar Cascade classifier, based on Haar wavelets, has been invented by Viola and Jones in the year 2001. Haar
wavelets method is used to analyze pixels in the image to square shaped function Haar features are image features
used in face recognition.

These features are adjacent rectangle regions in the detection window. Haar cascade is usually trained with
positive and negative pictures, then combination of these are taken for representing the face i.e. positive pictures
possesses a face whereas the negative picture does not contain face for face detection. The features are obtained by
computing the subtraction between the sum of bright pixels with the sum of dark pixels.

Adaboost learning algorithm is used for selection of important features from the larger set and removes the
redundant features. All the features are tested one by one for detecting the face. If any features does not appear in the
block then detection stops.

Some of the haar features are edge feature, line feature and four rectangle features. Edge features are used for
detecting the edges whereas line feature and four rectangle feature is used for detecting slanted line. For detecting the
presence of the haar like features in an image the value of delta should be closer to one.
n n
1 1
 = dark- white =
n
 I(X ) -
dark n
 I(X )
white

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1237
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Fig.2: Haar Like Features- edge, Line, & Four Rectangle

B. Histogram of Oriented Gradients (HoG) Face Detection

Histogram of oriented gradient method is invented by Geitgay in the year 2016, entails image in gray scale and it
compares each pixel with its neighboring pixels with the intention to obtain the direction where the picture is darker.
This is indicated by white arrow, like this pattern is prepared, if the face matches with the pattern, then the face is
detected.

C. DNN Face Detection

The detector system uses a single shot multibox detector with Res net -10 architecture as its backbone. The
model of face detector is implemented using 8 bit quantized version using tensorflow. Image is converted into blob
and passed through network using forward function. The output of the face detector is a 4 D matrix.

D. CNN in dlib Face Detection

The system uses maximum margin object detector with CNN based feature. The detector works in different
orientation of faces. It gives a very easy training process. Faster face detection with GPU compared to with CPU.

E. Yolo V3 Face Detector

The Deep learning face detection method uses a yolov3 algorithm which delivers better performance and has fast
face detection speed. The face detection uses pre-trained you only look once v3 (yolov3) weights and configuration
files. Darknet 53 network is used to extract the facial features from the image.

Fig.3: Face Detection with Different Type of Detectors

The advantages and disadvantages of different types of face detectors are tabulated below.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1238
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Table 1: Comparison of Different Face Detector

S. Detection Method Advantage Disadvantage
No
1. Haar Cascade  Simple architecture  Not detecting the face in the image for non-
 Possible to detect faces at different frontal conditions.
scales
2 DNN face Detection  Detection is possible at non frontal  Not detecting the face in the image for non-
condition. frontal conditions.
 Possible to detect faces at different
scales
3 Histogram of Oriented  Fastest method when used in CPU.  Side face is not detected which is extreme
Gradient  Possible to detect frontal and slightly non frontal.
non frontal face.  Face size more than 80x 80 is not detected.
4 CNN face detection  Perform detection in frontal and non-  Face size more than 80x 80 is not detected.
frontal faces.  Slower face detection with CPU compared to
 Training process is fast. with GPU.
 Faster face detection with GPU
compared to with CPU.
5 Yolo V3 Face Detection  Smaller face image can also be  Difficult to implement using raspberry pi.
detected.
 Fast and good for real time processing.
 Better accuracy
 Frontal and side wise face is also
detected.

F. Face Recognition Using Local Binary Pattern Histogram

The method requires gray scale image for processing the training set. Firstly the face is divided into 7 x 7 equal
sized cell. Each pixel in the cell has its weighted decimal value. Each centered pixel is surrounded with 8 pixels. The
weighted decimal value can be represented as binary by comparing the decimal value in the neighbor with the
centered pixel. These binary values of each cell are called threshold.
In this approach threshold value is obtained by comparing the center pixel with its neighbor pixels and if the
value is different than 0 is taken as threshold value and if the value is similar than 1 is taken as threshold value. The
histogram of each region is taken and all the histograms are concatenated to form feature vector. For recognizing the
face histogram of current face is matched with the histogram of training data and labeling is done for matched faces.
In the face recognition process, the histograms are created for the new image. For matching of input image with
available dataset image, two histograms are to be compared by calculating the distance between the two by Euclidean
distance.
The calculated distance is termed as confidence. Lower confidence delivers better result. The threshold and
confidence value is used to estimate the recognition accuracy.

Fig.4: Concatenation of LBP Values of Histogram [Adopted from [2]]

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1239
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

G. Face Recognition Using Convolution Neural Network

A feed forward neural network, Convolution neural network, comprises of four types of layers, first layer is
convolution layer, having independent filters, which is followed by pooling layer, then rectified linear unit (Re LU
layer), and fully connected layers.
Convolution layer, convolved with the image that provides feature extraction. Next to the convolution layer is
pooling layers, in charge of reduction of the dimension which in turn shortens the training time and controls over
fitting. Next layer is rectified linear unit, gives better convergence. Re LU layer is followed by fully connected layer,
used for the prediction of label to define the face image.
The face recognition using CNN method consists of face detection, face embedding for quantify the captured
face and classifier for training on the embedding.
The face detector locates the faces in an image. Further from the image, 128-d feature vectors are obtained.
Training is done on the face embedding process for optimization, using SVM classifier. The model proposed for
execution of face embedding is FaceNet Model.

Fig.5: Face Detection & Recognition

IV. Implementation of Raspberry Pi based Attendance Authentication System Using Face

Recognition
The proposed system is developed for implementing in the Institute for taking attendance in the particular
classroom using raspberry pi. Raspberry pi does not have enough memory to handle the CNN based face detector so
local binary pattern histogram method is used for face recognition. The system has the following advantage.
 Manual attendance taking is removed by lecturer and which requires less time for maintaining records.
 Increases the time for teaching learning process.
 Possibility of false attendance or proxy attendance is removed.
 Preventive measures will become easy to take for students having less attendance.
 Timely communication to stakeholder by System driven alert mail.
 Easy in retaining of records and have zero redundancy.
The flow diagram of the proposed system is given below.

Fig.6: Flow Diagram of Attendance Authentication System

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1240
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

A. Raspberry pi 3 B+
Raspberry pi 3B + is a credit card sized computer with 64 bit quad core Arm cortex- A53 CPU works at 1.4 GHz.
It contains built-in enhanced dual band 2.4GHz and 5 GHz Wi-Fi and 4.2/BLE Bluetooth connectivity with faster
Ethernet. 1GB of RAM is available in raspberry pi. It comes with 2.5A power supply with micro USB cable and
noise filter. Raspbian OS is flashed onto a micro secured SD card, which runs the raspberry pi using python
programming language.

Fig.7: Raspberry pi Setup for the Model

Features
 14th March 2018, the day raspberry Pi 3 B+, an advanced version of Raspberry Pi 3 B model was introduced
by Raspberry Pi foundation. It is a tiny single computer board, capable of doing some functions like a computer
or laptop, comes with CPU, USB ports, GPU, I/O pins, Bluetooth, Wi-Fi, USB and network boot.

Fig.8: Different parts in the Raspberry pi

 USB and Network Boot and Power over Ethernet facility is the extra feature available in B+ version along with
two extra USB ports. The Raspberry pi 3 B+ model, SoC (system on chip) combines both CPU and GPU on a
single package which turns out to be 3 times faster than Pi 2 models.
 The model has a better processing speed with better range in wireless challenging environments, dual-band
WiFi 802.11ac runs at 2.4GHz and 5GHz and Bluetooth 4.2 is offered with BLE support.
 The metal shielding on the top side of board acts as a heat sink and drains out extreme heat if it is exposed to
the high pressure or temperature.
 The gigabit Ethernet based on USB 2.0 interface comes with 300 Mbit/s speed, faster than 100 Mbit/s of Pi 2
model, with four pin header resides near 40 pin header, allows the Power over Ethernet (PoE), provides

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1241
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

necessary electrical current to the data cable used device thereby reduces the number of cables required for the
installation of a device.

Fig.9: Block Diagram of Raspberry pi based Attendance System

B. Pi Camera for Raspberry Pi

The pi camera module is an 8 megapixel Sony IMX219 image sensor supported with fixed focus lens, capable of
driving 3280×2464 pixel static images and it also supports 640×480p90, and 1080p30, 720p60 video. The image can
be stored locally or transmitted to another location. It interacts with the raspberry pi through MIPI camera serial
interface protocol. The proposed system uses raspberry pi 3B+, pi camera module and laptop. The system provides
attendance of the present student using face recognition. The attendance log is saved in MySQL database.

C. Generation of Image for Dataset

The pi camera module captures the image on clicking of image generation option in the GUI. 60 images of the
face is captured and stored in the folder with the name of the person.

D. Face Recognition Using Local Binary Pattern Histogram

The image is captured by pi camera module and face is detected from the image using haar cascade method. The
detected face is converted into gray scale which is followed by conversion of PIL image into numpy array. The
feature extraction of face is carried out which is then compared with the data available dataset. For the matching
face, ID and name is obtained from the dataset folder and displayed with the recognized face.

E. Creation of MySQL Database with APACHE Web Server

To establish a smooth and secure communication between server and the client through HTTP protocol,
APACHE is used. It is reliable and stable open source software and can be used free of cost for commercial use also.
It is very flexible due to its module based structure. APACHE web server has the ability to support multiple
programming languages. It works in both window and UNIX. The programming language PHP operates with
APACHE for the creation of dynamic web content. MySQL is the database for storing the attendance log of the
students in a easy queried format.
In the proposed model two data base is created, first one is the student’s record which has all students name,
registration number, father’s name and email Id of parents. When the face will be recognized, the name will be
entered in the second MySQL database, Attendance log. For sending mail, comparison between the students record
and the attendance log entry is carried out and mail will be send to the parents email Id whose wards entry is not
available in the attendance entry.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1242
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Fig.10: Process Flow for Functioning of Creation of Database and Web Server

F. Sending of Message through Mail

The sending the mail through Gmail, first Multipurpose Internet Mail Extension (MIME) is created. The sender
and receivers credentials are stored in the header of MIME. The less secure app access is activated for access of
Gmail. Python script is written for sending the mail to the parents. The SMTlib module available in python package
is used for sending the mail with send mail transfer protocol (SMTP). SMTP is basically a communication protocol,
a set of communication guidelines used to allow the software to send electronic mail via internet. The SMTP session
can be initiated with a valid port number and security features. Every server have different port number. The python
cod for sending is executed through the terminal and the Gmail will be received by the parent.

Fig.11: Process Flow of Sending Message through Gmail

G. Graphical User Interface using Tkinter

Graphical user interface is a desktop APP which helps the user to interact with the code available in the
computer. Some of the applications of GUI are calculator, games like chess and solitaire etc. Tkinter is offered with
the python package. It is python’s standard framework for graphical user interface (GUI). This is an open source and
is accessible by python license. Some other packages available for GUI framework are PyQT, Kivy, Jython,
WxPython etc.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1243
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

To create a GUI first the Tkinter module is imported. The second step towards creation of GUI is initialization of
window manager which is performed by Tkinter. Tk( ) which is followed by renaming of GUI window. A blank
window will be created with minimum, maximum and close buttons.
Different types of widgets like button, label, canvas, check button and entry are used for developing GUI with the
help of python script.
The widget for the development of the GUI is listed below.
Button – The event proposed is triggered when clicked.
Label – This widget is used for inserting text into the window.
Canvas – canvas is used to draw shape to the GUI and also supports various drawing techniques.
Entry – creation of input field is done through this widget. The text can be entered in the field provided for
processing.
Menu – menu is used to create menu in the GUI.
Check button - It is a toggle button, records all true false or on-off state.
Frame – The grouping of the widget is done through this widget, is used as a container in Tkinter.
Combo box – IT is a dropdown menu used for displaying all the elements which is defined in the python code.
Message box – Displaying of message is performed through importing this library.
Radio button – It suggest multi choice option to the user.
Spin box – This widget comprises two tabs, used for up and down scroll.
List Box – It provides the option to select a list like java, python, C ++ to the user from which different features can
be accepted.
Scrollbar – talk about the slider controller for implementing the listed widgets.
Geometry Management- this is used to establish the frame or the window. There are three prominent type of
geometry measurement classes are available.
Pack ( ) – It arranges the widget in a block manner.
Grid ( ) - It arranges the widget in a table like structure.
Place ( ) – It is used to place the given widget in specified place.
The flow diagram of developing the GUI is given below.

Fig.12: Flow Diagram Depicting the Development of GUI

The image of the developed graphical user interface is shown below. This desktop App will capture the image
and prepare the dataset, train the image and recognize the face.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1244
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

Fig.13: Display Screen of GUI Using Tkinter

The work flow of the system is given below.
 First the registration number and name of the student is entered the entry field provided.
 The capture image button is clicked for generating the dataset. The dataset will be stored in the folder with
name of the student.
 The captured data is trained by clicking the train image button.
 The face recognition button is pressed for recognizing the face.
 The attendance of the student is displayed the field provided.

Fig.14: Flow Diagram Depicting Functioning of GUI

H. Confusion Matrix
The confusion matrix also called as error matrix is used to analyze the performance of the algorithm on tested
dataset. The confusion matrix is prepared by Sklearn library with the help of the function confusion matrix. Python
script is written for obtaining confusion matrix array, the output array without normalization.
The classification is also obtained which gives precision, recall f1-score and accuracy. These parameters are
calculated with the help of true positive, true negative, false positive and false negative. The true positive outcomes
are the correct prediction of positive class by the system, true negative is nothing but the outcome in which the

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1245
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

system predicts the negative class, false positive is an outcome which is incorrectly predicted the positive class by
the system and false negative predict incorrectly the negative class. True positive is denoted by TP, true negative is
represented as TN, false positive and false negative are denoted by FP & FN respectively.

1) Precision
Positive predictive value is termed as precision. The ratio between actual positives to the total positive provides
precision.
Precision= (1)

2) Accuracy
The ratio between the sum of true prediction (true positive and true negative) and the total predictions (true
positive, false positive, true negative, and false negative) is termed as accuracy.
Accuracy = (2)

3) F1 – Score
F1 score, a statistical measure to the rate performance, conveys a balance between sensitivity and precision. It is
a mean of an individual’s performance, based on recall and precision, also called F score.
F1- score (3)

4) Sensitivity
Sensitivity is also called recall. It is obtained by dividing the true positive by the sum of true positive and the true
negatives.
Sensitivity = *100 (4)
The tested data of eight students is used for making the confusion matrix. The recognition has been carried out
for eight students. The face recognition of Liz and Anita had confusion with Aishwarya and Benita. Remaining all
faces have been recognized with 100% accuracy. The output array and the classification report obtained in the
terminal are depicted below which shows 95% accuracy.

Fig.15: Confusion Matrix Array and Classification Report

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1246
Received: 20 Feb 2020/Accepted: 18 Mar 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

V. Result and Discussion

The face recognition has three parts such as image capture, image training, and recognition of face. The images
of students are captured three webcam and stored in the folder with name of the student. The training image of
student is given below.

Fig.16 (a): training Image

The recognition result shows face of the student with name and ID of the student.

Fig.16 (b): Face Recognition

The attendance is entered in MySQL database for the present students. Intimation send to parent for absent
students.

Fig.16(c): Attendance Log of Students in MySQL Database

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1247
Received: 20 Feb 2020/Accepted: 18 Mar 2020
 Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 04-Special Issue, 2020

VI. Conclusion
Different types of face detection using Haarcascade method, DNN face detection using single shot multibox
detector, HOG and CNN face detector implemented and comparative study carried out. Implemented two method of
face recognition, local binary pattern histogram and convolution neural network. Raspberry pi based face recognition
implemented using local binary pattern histogram algorithm with an accuracy of 95%. Apache web server provides
the Application programming Interface API for launching the MySQL database which maintains the Attendance log
of students. The parents are informed about the absent students through Gmail with the assistance of SMTP.
Performance can be improved by using raspberry pi 4 model for handling heavy face recognition algorithm along
with additional feature of recording of Lecture as a future work.

References
[1] Mallikarjuna Reddy, K.Naresh, S.Shiva Kumar, Madar Bindu, “Comparison of various detection and face
recognition algorithms”, International Journal of Engineering Research in Computer Science and
Engineering, vol 5, issue 4, April 2018.
[2] Omar Abdul Rhman Salim, Rashidah Funke Olanrewaju, Wasiu Adebayo Balogum, “Class Attendance
Management System using Face Recognition”, 7th International Conference on Computer & communication
Engineering, September 2018.
[3] Priyanka Wagh, Roshani Thakare, Jagruti Chaudhari, Shweta patil, “ Attendance System based on Face
Recognition using Eigen face and PCA algorithm”. 2015, International conference on Green Computing and
Internet of Things, 8-10 October 2015.
[4] Shubhobrata Bhattacharya, Gowtham Sandeep Nainala, Prosenjit Das, Aurobinda Routray, “Smart Attendance
Monitoring System (SAMS): A Face Recognition based Attendance System for classroom environment”, 2018
IEEE 18th International conference on Advanced Learning Technologies,9-13 July 2018, Mumbai, India.
[5] E. Varadharajan, R.Dharani, S. Jeevitha, B. Kavinnathi, S. Hemalatha, “ Automatic Attendance Management
System using Face detection”, South Asian Journal of research in Engineering Science and Technology, vol 3,
issue 10 April 2018.
[6] Nilest D. Veer, B.F Momin, “ An Automatic Attendance System using Video Surveillance Camera”, IEEE
International Conference on recent trends in Electronics Information Communication Technology, May 20-21,
2016.
[7] Rekha E, Dr. Ramaprasad, “An Efficient Automated Attendance Management System Based on Eigen Face
Recognition”, 2017 7th International Conference on Cloud Computing, Data Science & Engineering.
[8] Wenxian Zeng, Qinglin Meng, Ran Li, “Design of Intelligent Classroom Attendance System Based on Face
Recognition”, 2019 IEEE 3rd International Technology, Networking Electronics and Automation Control
Conference(ITNEC), Chengdu,China, 2019, pp.611-615.
[9] Soniya V, Swetha Sri R, Swetha Titty K, Ramakrishnan R, Sivakumar S, “Attendance Automation Using Face
Recognition Biometric Authentication”, 2017 International Conference on Power and Embedded Drive
Control (ICPEDC).
[10] Refik Samet, Muhammed Tanrivendi, “Face recognition based mobile Automatic Attendance Management
System”, 2017 International Conference on Cyber worlds, Chester UK.
[11] Marko Arsenovic, Srdjan Sladojevic, Andras Anderla, Darko Stefanovic, “Face Time-Deep Learning Based
Recognition Attendance System”, 15th International Symposium on Intelligent System and Informatics, Sept.
14-16, 2017, Subotica, Serbia.
[12] Nazare Kanchan Jayant, Surekha Borra, “ Attendance Management System using Hybrid Face Recognition
Techniques”, 2016 Conference on Advances in Signal Processing.
[13] Jenif W.S.D Souza, Jothi S, Chandrasekar, “ Automated Attendance Marking and Management System by
Facial Recognition using Histogram”, 2019 5th International Conference on Advanced Computing &
Communication System.
[14] Joseph Redmon, Ali Farhadi, “yolov3: An incremental Improvement”. Published in ArXiv 2018.

DOI: 10.5373/JARDCS/V12SP4/20201599
ISSN 1943-023X 1248
Received: 20 Feb 2020/Accepted: 18 Mar 2020

View publication stats