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

net/publication/330566339

Efficient Real Time Attendance System Based on Face Detection Case Study
"MEDIU Staff"

Article in International Journal of Advanced Computer Research · July 2017

CITATIONS READS

21 1,459

2 authors:

Huda Mady Shadi M S Hilles

10 PUBLICATIONS 145 CITATIONS

Istanbul Okan University
39 PUBLICATIONS 374 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Shadi M S Hilles on 23 January 2019.

The user has requested enhancement of the downloaded file.

International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

Efficient Real Time Attendance System Based on Face Detection

Case Study “MEDIU Staff”
HUDA.H.Mady 1, Shadi M.S. Hilles2,
1
Computer Science department, Faculty of Computer and Information Technology, Al-Madinah International University
[email protected], [email protected]
Received 1 May 2017; accepted 15 July 2017

Abstract

Recently, using biometrics applications for person’s authentication has received more attention by computer vision researchers.
Biometrics applications play an important role in reliable person authentication for recording the attendance of employees in the large
organizations. Face detection is considered as an optimal solution for enrolment attendance among all biometrics applications such as
Fingerprint. The main objective of this paper is to instigate a develop attendance system automatically in Al-Madinah International University
(MEDIU) and achieve higher both accuracy and performance. To develop system of face detector under challenging of face the proposed
system accuracy using real life videos by establishing Data Base for MIDIO staff. This paper suggests a combination detector based of voila
Jones which deals with challenges in real time of video. In the conducted study the system was capable of determine one face or more in
real time extremely rapidly while achieving high detection rates with accurate and effective performance. The detection results of 100 frame
per video from the data base was satisfying, with the highest result was 96% and the lowest one was 63%.

Keywords: (face detection, voila Jones, challenging of face, MEDUE, Video Database)

1. INTRODUCTION process can be done from a distance and in a disturbed

environment. It usually involves the following three steps:
Biometrics-based techniques have emerged as a first step is detection and pre-processing of a face in a
promising approach in recent years to identify and verify complex background, and localization of its exact position.
individuals because of manual labour in traditional Second: extraction of facial features, such as eyes and nose,
conservation and management. The attendance sheets are followed by normalization to align the face with stored face
boring because they consume a lot of time for the lecturer. images. Thread: face classification or matching.
Recognition techniques are existent, such as fingerprint and
iris Recognition. These techniques are intrusive, and their Localization and detection of faces’ identity are
success depends highly on user cooperation because users challenging demands due to facial appearances variations
must position their eyes in front of an iris scanner or put their considering fundamental factors, such as facial expressions,
fingers in a fingerprint device.[1] and makeup. External factors include form changes, lighting
conditions, and partial blockage. These factors cause
Enrolment real time automated attendance due to Human difficulties in developing a good model face. Significant
face determination and recognition by face detection and efforts were exerted in the field of computer vision and
recognition algorithm is a vital research area in the computer pattern Recognition, but the results were good only in
vision community which are non-intrusive, inexpensive constrained settings and challenges emerged [4]. A
technique, can be captured at a distance and in a covert combination of simple to complex classifiers based on
manner[2]. However, localizing and recognizing face are cascade detector was proposed to overcome this problem
challenging in facial appearance because of the significant [5].
variations in intrinsic factors, e.g. facial expressions,
variations in Illumination, scale, poses, occlusion, complex 2. RELATED STUDIES
background, and different. Human race; [3] .Various efforts
have been exerted in computer and vision and pattern Recently, face detection has attracted significant attention
recognition, but the reliability significantly affects and its researches have rapidly developed by computer
performance. Nowadays, Human face determination is a engineers, since this field has many possible applications in
vital research area in the computer vision community. It can computer vision communication and automatic access
be widely used for video indexing, human computer control system. Face detection is a vital part of face
interaction, surveillances, and robotics. There are several recognition as it is the first step of automatic face
related difficulties in human face perception. recognition. However, face detection in attendance system
The face detection and recognition process represents a videos is not a simple step due to presence of many facial
number of advantages compared to different other biometric appearance variations, such as occlusion, pose variation
technologies e.g. the fingerprint or iris as it has become as a (front, non-front), illuminating condition, image orientation
natural process and nonintrusive. Along with that, this and facial expression.[6]
International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

K.-K. Sung and T. Poggio [7] Obtained an example-based propagation and self-organizing feature map SOFM as
learning method for detecting of human faces vertical frontal presented [21, 22] which used vector quantization for image
views in complex image background, in which contain a face compression, SVM, and boosting. The vast majority of
and non-face distributed through it. The classifier has been processed patterns do not include faces; thus, single of
trained to detect the non-face and distance metric in the classifier-based systems, such as PNN [9] and SVM [10, 11].
image, in different features vector measurements to ensure
the success of the proposed system. Thus the main problem in face detection Algorithm is the
classification between non-face and face, which has the
P. Viola and M. J. Jones [5] Described a framework for following two characteristics: the number of face is much
face detection, which can process images at a high speed lesser than that of non-face and the distribution of non-face
while achieving high detection rates. Three main pattern is very prevailing, so it’s difficult to classify non-face
contributions are presented in the current study. The first and face by one method the companion between two method
contribution is an introduction that represents a new image or more affective performance; furthermore the computing
called ‘integral image ’which is computed quickly. The capacity is high and large demand for real time is
second one is a simple and effective classifier that is built required[12].
using the AdaBoost learning algorithm to identify a few
critical visual features from a large set of potential features. This research used Viola and Jones algorithm based of
The third contribution is an approach to combine classified Group of classification model, which are trained using
features in a ‘cascade; which allows the deposition of the boosting to detect a specific type of object the profile face
background regions of an image quickly. This method used and frontal face CART classifiers to detect faces from
allows the face detection system to be approximately 15 query images. These tow classifiers are trained by extracting
times faster than previous approaches. features from a set of defined image based on Haar-Like
The Viola Jones algorithm has four stages: Haar Feature feature extractor. These cascade detectors are trained to
Selection, Creating an Integral Image, Adaboost Training identify the faces and reject non-face [13]
and Cascading Classifiers. [8]. In a face detection stage,
must use a delicate numerical descriptive such that it sets of Y. Muttu [14] Suggested Viola Jones algorithm was
human face apart from other objects in a given image, this applied for face detection and modified the LBP algorithm
algorithm called Adobos. A conclusive element of the Viola to increase algorithm performance (reduce the time) in the
Jones algorithm is a technique to compute a rectangle feature extraction stage. Viola Jones algorithm was applied
features very speedily. This technique uses an intermediate for face detection. Radical basis function NN presented
representation of the image, that so called an integral image. better classification accuracy than back propagation.
The cascade object detector uses the Viola& Jones algorithm However, further work is needed to improve the accuracy of
to detect people's faces, noses, eyes, mouth, or upper body. facial expression Recognition.
Nevertheless, the Training Image samples can be used to
train a custom classifier to be used within this System. More H. Yang and X. A. Wang [12] Implemented Face
importantly, the classifications of cascade which is Recognition system based on viola and Jones face detection
considered as a specified model and a character vector that method, histograms of oriented gradients (HOG) to feature
sets the classifications for the detector, [8]. The cascade extraction and support vector machine (SVM) as applied as
classifications model which is considered as the main nine classifier. This paper designed multi stage cascade to
designed classifications. Cascade Classifications Model detected face, the result showed better performance with
such as Frontal Face (CART), Frontal Face (LBP), Upper CMU/MIT database that was 94% of detection rate. This
Body, Eye Pair, Single Eye, Single (CART), Profile Face, paper applied support vector machine (SO-SVM) which is
Mouth, Nose. trained off line to detect specific faces to break the self-
learning loop typical in tracking. In this experimented
There are different researched that found a number of Deformable Part Model DPM landmark detector was used to
solutions recently with a wide and sophisticated techniques evaluate the accuracy of the face detection output. This
to detect faces. Starting from the simple edge-based system has achieved great tracking execution with different
algorithms to complex level as the composite high-level. The scenarios on an online video.[15]. This system has achieved
use of algorithms has presented a classified features for the great tracking execution with different scenarios on an
process of face detection and manage a great number of online video.
challenges and uncertainties in the process. And due to the
lack of standardized experimental work there is no a B. Du, S. Shan, L. Qing, and W. Gao [16] studied
comprehensive or comparative assessment or evaluation. As histogram equalization (HE), histogram specification (HS),
the comparisons can be only made when data is provided. logarithm transforms (Log), and Gamma intensity correction
[3]. Face detection challenges from Image or videos are still (GIC), and self-quotient image (SQI) were compared and
not overcome yet. Difficulties range from dealing with analysed. These methods were evaluated on three different
intense real time different face appearance to handle database which are CAS-PEAL, FERET and CMUPIE
variations factors. Various approaches are utilized to detect database. The results depicted that the three methods: HS,
faces, such as Artificial Neural network ANN based on back- HE and GIC improved authentication performance with
International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

different degree of illuminations variations. On the other extracted using the fast algorithm recently proposed by [20].
hand, it was found that Log and SQI might reduce the Therefore, combining two detectors based of Viola and
authentication rate for facial images without much Jones are presented combines estimates from many facial
illumination variations. detections in order to reduce error rate. Effective and robust
design face detection system must be able to handle face
Contrast Enhancement In the process of the contrast challenges like cluttered scenes, face size mutation, face
enhancement the techniques do expands the real range of the modification, location, orientation, diversity in
brightness of values in the image itself. The image can illuminations, facial expressions, pose, occlusion, different
effectively work in a manner that needed for the analysis to backgrounds and variations among different, occlusion,
be made. However, the values of density in the scene have different backgrounds and variations among different
to be pulled for the farther apart in the expanding for a better nationalities and races. In this chapter the suitable algorithms
and higher range of process. The effect to do that will have been used to evolve and expand the face Recognition
increase the contrast visually between the two different areas system for the achievement of better accuracy with more
in the densities, and that will empower the analysis to advance performance. For this the created MEDUE Staff
differentiate between these areas initially and having a better Database (MEDUE-S-V-DB) has been creation and used as
difference of density [17]. a case study. Proposed Flowchart of MEDUE-S-V-DB
creation showed in figure (1).
The histogram equalization methods used for gray-scale
transform defined to be as a known method for the image
processing in order to contrast adjustment using the method
of the histogram. Consequently, often used as the
illumination pre-processing. This method is significantly
affecting the process of making the needed and useful image
for the background and for the foreground even though both
are dark or light. The main advantage of this method is that
it works directly and fairly on the technique and use the
invertible operator. Therefore, the histogram works on
Equalization Function is very well known as it is easily can
be managed to recover and the calculation effect is not to be
computationally intensive in the process function.(B. Du,
[16], [18].

[1] Proposed attendance system for the student to

improve the method of the educational system. Facial
detection is performed by the Jones Viola algorithm to detect
and manage objects under environments, which are
uncontrolled and controlled. The result in proposed
attendance system showed 100% accurate under controlled
environment, By contrast, the effect under uncontrolled
environment is quite low 60 %. This approach can be
effective in automatically recording student attendances in
actual

It is clear that, a significant progress in the Automated

Attendance Management System in real-time is one of the
most challenges associated with person’s detection, this area
many novel methods of face detection and modeling systems Fig. 1.flowchart of MEDUE-S-V-DB creation
have been proposed to resolve each variation listed above
However, accurate and robust face detection still offers a
number of challenges which are variations in illumination, Create MEDUE-S-V-DB:
scale, Occlusion, facial expression, pose problems .Thus,
methodologies need to be deeply studied in order to MEDUE-S-V-DB has been for the administrations of
overcome the face recognition problems with considering experiments in face Recognition and detection system under
accuracy and performance balancing.[19] [8][11]. real time and in order to study the effect of different factors
to execute definiteness and performance. MEDUE-S-V-DB
was designed under continuous motion effects, and
3. RESEARCH METHODOLOGY variations in term of pose, illumination conditions, facial
expressions, orientation and occlusion. As a matter of
This paper presents progress toward an integrated, robust, consideration during this study the nationality variation of
real-time face detection and system. Faces are detected and the staff were taken. MEDUE-S-V-DB contains 20 videos
International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

for 10 peopl.Two videos were recorded for each person. In Viola Jones algorithm that indeed considered as a useful
idition,one video was created included three person . Figure algorithm for face detection are as follow: robust which is
2. Is showed bellow the flowchart of MEDUE-S-V-DB very-high detection rate the true-positive rate, The opposite
creation. one is the very-low false-positive rate and Real time which
is for a practical application at least two frames per-second
and this is must be processed and only Face detection with
no recognition- The main aim is to distinguish faces from
non-faces. Cascade architecture is object detector uses the
Viola Jones algorithm to detect people's faces, noses, eyes,
mouth, or upper body. Slow but accurate classifiers are used
in the later stages for classifying face-like patterns. Thus, the
complexity of classifiers can be adapted to correspond to the
increasing difficulty of the input patterns. Several techniques
are employed to compute various features under varying
Fig. 2. MEDUE-S-V-DB scales and locations effectively, which are crucial in real-
time performance.
Video Acquisition (for training and testing):
Image Pre-Processing:
The training videos contain the various faces with
different poses (B), illuminations (C), facial expressions (D), Pre-Processing is applied typically to transform a source
orientations (A) and occlusions (C). Then for the current image into a new image, which is fundamentally similar to
study, the nationality variation of staff (E) was taken into the the source image, to overcome problems due to low-quality
consideration to train classification algorithm. The testing image. In the Pre-Processing stage, colure images may be
videos were testing videos frames that contained one face or converted into gray scale images, and images are resized to
more, which were compared to a database to find the most the required size and filtered using a low-pass filter to
suitable match. The training video is divided into 100 Illumination normalization can be conducted using
frames which offer the services of cropping and resizing of histogram equalization .In below Figure 4 is shown Pre-
the faces for training the classifier to match faces of the Processing facial images.
unknown person. The figure (3) below shows the alteration The pre-processing stage will be executed for every
among the faces of MEDIU-S-DB. Testing videos is divided detected face as shown in figure 12 in original frame (A).
into 30 frames; these frames are selected every 1 second, to For the pre-processing purpose, the detect faces were
evaluate the system efficiency for identify the people. Figure cropped (B), then converted all images to grayscale(C), after
(3) is showed Variation Faces from MEDIU-S-DB. Testing that, resized the images of faces to remove the noise. (D).
videos is divided into 30 frames; these frames are selected (HE) have been used for normalization (E). Pre-processing
every 1 second, to evaluate the system efficiency for identify one frame is showed in figure (4).
the people.

Fig. 4 Pre-processing a frame

Figure 3. Variation Faces from MEDIU-S-DB Research Procedure:

Face detection is the process to determen people in
Face Detection by Viola-Jones Algorithm:
photographs, figures, images or video frameworks by
Face detection is an advance technology that determines comparing the appearance of face in video frame (Query
the position and variance of the sizes among human faces in frame) to a faces data set .Figure 6 showed the suggest
digital images or video frames. In this research the cascade Digram of Face detection System and this process can be
detector uses the Viola-Jones algorithm to detect faces and a done as the following:
trained classification model for detection was proposed. Procedure 1. The data was obtained from created
Frontal CART and Profile face classifiers was used,. MEDIU-S-DB, as twenty videos were taken for 10 members
Viola Jones Algorithms is applied for fast and easy face (2 videos for everyone) in different face recorded with
detection in real time. Cascading can be performance: In variation of affecting factors.Every single video was divided
trained the cascade of classifiers, a false positive rate of a into a hundred frames.the other 10 videos were used to
cascade showed in Equations (1 & 2). The characteristics of evaluate the proposal .
International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

Procedure 2. Face detection by Viola-Jones algorthem To Whosesoever F is the false positive rate for the cascade
achieve the required aims to design a robust real - time face
detection system under continuous motion, and variations in
term of pose, illumination conditions, facial expressions,
orientation, occlusion and nationality variation, Frontal classifier, K is the number for classifiers, and fi is the false
CART and Profile face classifiers were coupled based on positive rate of the i classifier in this examples that get
Viola&Jones algorithm. system started with detect face through to it. Rat of detection is:
K
D   di (2)
from video frames.this system ued MATLAB 2016b
Toolboxes was used for coding the desired system, data set i 1
based of videos and images.
Procedure 3. Pre-processing the images of faces such as Where D is the rate of detection in the cascaded classifier,
(cropping, convert image from color to gray-scale, resizing K is the number of classifiers, and di is the rate of detection
faces and normalization the faces by histogram in the i classifier on the examples that get through to it. [5].
equalization).
6. Conclusion:
4. RESULTS AND DISCUSSION
Face detection is a very vital topic in computer and in the
Experiment 1: Accuracy of detection system: world of information technology, during the last few years
among the researchers ‘community. It has bound the
The preliminary experiment was testing of the accuracy of researchers together due to different reasons but leading
system detection; seven videos from MEDUE-S-V-DB were towards the same interest. The utmost reasons are that it can
tested as a sample and observed it under continuous motion, be used in commercial life, law enforcement and security
and variations in term of pose, illumination conditions, facial applications that require Recognition. The other possible
expressions, orientation, occlusion and nationality variation. reason could be its convenient and universal procedure as
Then, every video was divided to 100 frames to find the compared with Biometric applications. To achieve the
percentage of detection to show the robustness of the required aims to design a robust real - time face detection
proposed detection system. The results show the efficincy of system under continuous motion, and variations in term of
the system to detect face with different challenges as shown pose, illumination conditions, facial expressions,
in figure (5) orientation, occlusion and nationality variation, Frontal
CART and Profile face.

7. References:
[1] N. D. V Santhi, Studies in Computational Intelligence
660 Intelligent Techniques in Signal Processing for
Multimedia Security. .
[2] L. Introna and H. Nissenbaum, “Face Recognition
Technology - A Survey of Policy and Implementation
Fig. 5. Detection with different challenges
Issues,” Lancaster University Management School, 2010.
[3] D. Hutchison, Biometric Recognition. 2015.
Experiment 2: develop a Face Detection system: [4] X. Zhang and Y. Gao, “Face recognition across pose:
were applied to achieve one of the main aims using three A review,” Pattern Recognition, vol. 42, no. 11, pp.
staff members in one video, then this video inserts into the 2876–2896, 2009.
system to detect three faces and recognize them at the same [5] P. Viola and M. J. Jones, “Robust Real-Time Face
time.The results showed that the system was abled to detect Detection,” International Journal of Computer Vision,
more than one faces crosponding person as shown in figure vol. 57, no. 2, pp. 137–154, 2004.
(6). [6] Y. Mao, H. Li, and Z. Yin, “Who missed the class? -
Unifying multi-face detection, tracking and recognition in
videos,” Proceedings - IEEE International Conference on
Multimedia and Expo, vol. 2014–September, no.
Septmber, 2014.
[7] K.-K. Sung and T. Poggio, “Example-Based Learning
for View- Based Human Face Detection,” IEEE Trans.
Pattern Analysis and Machine Intelligence, vol. 20, no. 1,
pp. 39–51, 1998.
[8] E. Multimedia, No Title. .
[9] L. Huang, A. Shimizu, Y. Hagihara, and H. Kobatake,
Fig. 6. Detect three faces “Face detection from cluttered images using a polynomial
neural network,” Test, vol. 51, pp. 197–211, 2003.
5. Equations [10] C. Zhang, Z. Zhang, O. M. Way, and R. Wa,
“Improving Multiview Face Detection with Multi-Task
International Journal of Contemporary Computer Research (IJCCR), Vol.1 Issue.2 (July, 2017)

Deep Convolutional Neural Networks,” pp. 1036–1041,

2014.
[11] Q. Tao, S. Zhan, X. Li, and T. Kurihara, “Author ’ s
Accepted Manuscript Robust face detection using local
CNN and SVM based on kernel combination Reference :
To appear in : Neurocomputing,” Neurocomputing, 2015.
[12] H. Yang and X. A. Wang, “Cascade face detection
based on Histograms of Oriented Gradients and Support
vector machine,” in P2P, Parallel, Grid, Cloud and
Internet Computing (3PGCIC), 2015 10th International
Conference on, 2015, pp. 766–770.
[13] S. Zafeiriou, C. Zhang, and Z. Zhang, “A survey on
face detection in the wild: Past, present and future,”
Computer Vision and Image Understanding, vol. 138, no.
March, pp. 1–24, 2015.
[14] Y. Muttu, “Effective Face Detection , Feature
Extraction & Neural Network Based Approaches for
Facial,” pp. 102–107, 2015.
[15] . Comaschi, S. Stuijk, T. Basten, and H. Corporaal,
“ROBUST ONLINE FACE TRACKING-BY-
DETECTION Eindhoven University of Technology , The
Netherlands TNO Embedded Systems Innovation ,
Eindhoven , The Netherlands,” 2015.
[16] B. Du, S. Shan, L. Qing, and W. Gao, “Empirical
comparisons of several preprocessing methods for
illumination insensitive face recognition,” in Acoustics,
Speech, and Signal Processing, 2005.
Proceedings.(ICASSP’05). IEEE International
Conference on, 2005, vol. 2, p. ii-981.
[17] A. . Fallis, Guide To Image Preocessing Ebook, vol.
53, no. 9. 2013.
[18] G. Hemalatha and C. P. Sumathi, “Preprocessing
techniques of facial image with Median and Gabor
filters,” 2016 International Conference on Information
Communication and Embedded Systems, ICICES 2016,
pp. 1–6, 2016.
[19] V. Pali, S. Goswami, and L. P. Bhaiya, “An
Extensive Survey on Feature Extraction Techniques for
Facial Image Processing,” in Computational Intelligence
and Communication Networks (CICN), 2014
International Conference on, 2014, pp. 142–148.
[20] M. J. P. Viola, “Rapid Object Detection Using A
Boosted Cascade of Simple Features,” Proceedings of the
2001 IEEE Computer Society Conference on Computer
Vision and Pattern Recognition, no. 1, pp. 511–518,
2001.
[21] Hilles, S. M., & Maidanuk, V. P. (2014). Self-
Organization Feature Map Based On Vq Components To
Solve Image Coding Problem.
[22]Майданюк, В. П., Мазін-Хіллес, Ш., & Мельник,
С. В. (2004). Кодування зображень з використанням
SOFM. Інформаційні технології та комп'ютерна
інженерія, 1(1), 49-52.

View publication stats