2017 International Conference on Advanced Computing and Communication Systems (ICACCS -2017), Jan.

06 – 07, 2017, Coimbatore, INDIA

Illumination Invariant Face Detection Using

Viola Jones Algorithm
Mangayarkarasi Nehru 1, Dr. Padmavathi S 2
1 2
First year M.Tech (CVIP), Assistant Professor,
Department of Computer Science and Engineering, Department of Computer Science and Engineering,
Amrita School of Engineering, Coimbatore, Amrita School of Engineering, Coimbatore,
Amrita Vishwa Vidhyapeetham, Amrita Vishwa Vidhyapeetham,
Amrita University, India. Amrita University, India.
1 2
mangainehru@gmail.com s_padmavathi@cb.amrita.edu

Abstract: In the recent days, there has been a wide a framework model that was placed under the Computer
advancement in human computing systems. It always Vision toolbox which used the vision. Cascade Object
remains a challenge to make the computer system Detector. This included certain parameters like Use ROI ,
behave like how a human senses things. Computer Scale Factor ,Min Size, Max Size , Merge Threshold,
Vision has been a pioneer in making things more Classification Model. Initial step is to train the cascade
automated and better for humans. This paper presents classification model for which the Classification Model is
a study based approach for detecting human faces used. The size of the smallest and largest object that can
using the Viola Jones algorithm. We train our be detected uses Min Size and Max Size parameters[1]. In
computer to automatically identify the human faces case where there are more humans found there multiple
from the given images irrespective of the illumination detections found surrounding the object whose criteria
conditions. Based on the experimental results we have will be defined by Merge Threshold. As a result this
discussed about the Viola -Jones Cascade Object parameter creates a bounding box around the target object.
Detector which uses various filters and the features to Sometimes there might be misclassifications leading to
detect the various parts of the face. false detections. This can be suppressed by increasing the
threshold value. It is possible to tune this property. On set
Keywords – SURF, Feature Detection, Face Recognition, this property to 0, [2]all the detections made can be
Face Detection, Viola-Jones algorithm returned without the threshold being performed. For multi
scale object detection we use the parameter Scale Factor
I.INTRODUCTION which has the value greater than 1.0001.It resolution is
scaled between the Min Size and Max Size.
The main objective is to design a system that detects
and recognizes the human faces from the input image or There are three basic methods used by Viola-Jones
video. It is possible to detect various parts of the human method [1]which is as follows:
body based on the facial features present. It is always easy
for a human to detect the faces from the given set of 1. Haar-like features are used for the feature
images and distinguish them appropriately. But for a extraction thus getting an Integral image.
system, it should be properly trained such that when a live 2. A machine-learning technique called Ada Boost
dataset is give by the use it should be able to detect the can be used for face detection. Since the
face of a human and also various other features like eyes, classifiers are very complex, using the boosting
nose, mouth etc.. The main idea perceived here is to make techniques the basic classifiers are built.
the computers detect the faces besides the non-facial 3. The final method is Cascade classifier which can
structures present in an image and finally to output the efficiently combine many features. In this, the
image with the detected part selected in it. The resultant classifier has several filters.
challenging part of this paper is to make the system detect
the faces irrespective of the illumination conditions. Once the strong classifiers are cascaded, the face and
non-face regions can be separated. This object detector
II.PROPOSED ALGORITHM also efficiently detects the nose, Upper body, Lips, eyes
and pupil. This paper discusses about the various feature
We use the Viola – Jones algorithm to efficiently vectors that can be used to detect a face using the
detect various parts of the human faces. When every other algorithm[4]. The face detection is controlled by the
people were busy to find the algorithm that would detect cascade object detection framework. The advantage of the
the faces it was Paul Viola and Michael Jones in 2001 proposed idea is that it can detect faces irrespective of
who came up with the most effective algorithm to detect illumination conditions. The Haar feature extraction is the
the human faces and its parts thus providing a framework major part in this process which uses Haar cascade
which was considered as the first ever object detection classifier. These Haar feature can tell us if there is any
frame work that provided object detection in real time. feature present in the given image. Each feature returns a
This algorithm was designed using the Matlab toolbox as single value which is given by the difference of sum of

978-1-5090-4559-4/17/$31.00©2017IEEE
2017 International Conference on Advanced Computing and Communication Systems (ICACCS -2017), Jan. 06 – 07, 2017, Coimbatore, INDIA

the pixels in the white region from the sum of pixels in the 000 features can be calculated which is impossible
black region. For the speedy detection of the face features practically. So, the Ada boost method is used which is a
we consider Haar features as the rectangular region. Some machine learning technique that efficiently finds the best
of the Haar filters that can be used in the detection of features out of 1,60,000.The Ada boost forms a strong
faces are given in fig below: classifier by taking the linear combination of the weak
classifiers which is given as:

F(x) = α1F1(x) + α2F(x) + (2)

Thus, if the image passes through all the stages then it

is a human face. If it fails even in any one of the stages
then it is not a human face. This gives an overall
description of the algorithm and how it is found

III. EXPERIMENTAL RESULTS

In this paper we consider certain constraints and
check if the faces are detected for all the different datasets
and meet the conditions. First we considered the datasets
of dark and fair colored men with and without glasses and
Fig:2.1 checked if face detection is possible. The black and white
image was also considered over here.The results were as
It is calculated as in (1) shown in fig: A bounding box was drawn around the
S(x) = Sum of black region / Sum of White region (1) detected face.

Initially when the image is fed as the input to the

system, the first step is to select the Haar features. It starts
to scan the image from the top left point and checks if any
face feature is present till it reaches the bottom right
corner of the image. In order to detect the faces from the
image we scan it several times using the Haar features.
The rectangular features are calculated rapidly using the
integral image. In this method, we will need only four
values that are present at the each corner of the rectangle
to[5] calculate the sum of the pixels present in any given
region. For an integral image, value at a point (a,b) is
given by the sum of all pixels present above and in the left
of (a,b).

Fig:2.2

Here in the fig: we find the Haar like features. There

is only one region and its black, so it is given as:
f0(x)=A.Lets consider one more filter as in fig:

Fig:3.1

Fig:2.3

This is given as: f1(x)= B-A-C and f2(x)= B+C-A-

D.In Voila Jones algorithm a window with size 24×24 is
used as the base window for evaluating the features in the
given image. In general using this window, about 1, 60,
2017 International Conference on Advanced Computing and Communication Systems (ICACCS -2017), Jan. 06 – 07, 2017, Coimbatore, INDIA

Fig: 3.4

The final criteria were to check for the face detection

of the masked and half-covered faces which gave a
Fig:3.2 positive result as shown in Fig: 3.5

Next case was to check if the face detection is

possible for the side faces as in Fig: 3.3 and also to check
if the faces were if a group of people were there under any
light conditions. The Fig: 3.4 give the resultant matlab
output for the required criteria.

Fig: 3.6
Thus we preserve our results for the further extension
of this project.

Fig: 3.3
2017 International Conference on Advanced Computing and Communication Systems (ICACCS -2017), Jan. 06 – 07, 2017, Coimbatore, INDIA

IV.CONCLUSION AND FUTURE WORKS [10]. R.-L. Hsu, M. Abdel-Mottaleb, and A. K. Jain, Face
Detection in Color Images, IEEE Transaction on
As the face detection is becoming an essential need in Pattern Analysis and Machine Learning, 24 (5) (2002)
696-706.
various application like face recognition, emotion
[11]. C. Zhang and Z. Zhang, A Survey of Recent
recognition and classification, identifying malpractices in Advances in face Detection, Technical Report, MSR-
exams and various other problems. We try to extend this TR-2010-66, Microsoft Research, Microsoft
idea for creating an automated attendance system in Corporation, June 2010.
classrooms where when the student enters the classroom [12]. Viola, Paul, and Michael J. Jones. "Robust real-time face
the camera will detect the faces of each student entering detection."International journal of computer vision 57.2
(2004): 137-154.
the classrooms and will mark attendance for [13]. Yow, Kin Choong, and Roberto Cipolla. "Feature-based
corresponding students after recognizing the faces thus human face detection." Image and vision computing 15.9
updating the attendance for each class in the database (1997): 713-735
automatically.

REFERENCES
[1]. Viola, Paul, and Michael Jones. "Rapid object detection
using a boosted cascade of simple features." Computer
Vision and Pattern Recognition, 2001. CVPR 2001.
Proceedings of the 2001 IEEE Computer Society
Conference on. Vol. 1. IEEE, 2001.
[2]. K. S. Gautam and Kumar, TaSenthil, “Discrimination
and detection of face and non-face using multilayer
feedforward perceptron”, Advances in Intelligent
Systems and Computing, vol. 397, pp. 89-103, 2016.
[3]. TaSenthil Kumar and Saivenkateswaran, Sb, “Evaluation
of video analytics for face detection and recognition”,
International Journal of Applied Engineering Research,
vol. 10, pp. 24003-24016, 2015.
[4]. Fernandez, Ma Christina D., et al. "Simultaneous face
detection and recognition using Viola-Jones Algorithm
and Artificial Neural Networks for identity
verification." Region 10 Symposium, 2014 IEEE. IEEE,
2014.
[5]. Arulkumar, C. V., and P. Vivekanandan.
"Multi-feature based automatic face identification on
kernel eigen spaces (KES) underunstable lighting
conditions." Advanced Computing and
communication Systems, 2015 International
Conference on. IEEE, 2015.
[6]. Hsu, Rein-Lien, Mohamed Abdel-Mottaleb, and Anil K.
Jain. "Face detection in color images." Pattern Analysis
and Machine Intelligence, IEEE Transactions on 24.5
(2002): 696-706.
[7]. W. Zhao, R. Chellap, A. Rosenfeld, and P.J. Phillips,
"Face Recognition: A Literature Survey," ACM
Computing Surveys (CSUR), vol. 35, no. 4, pp. 399-
458, December 2000.
[8]. Agrawal, Samiksha, and Pallavi Khatri. "Facial
expression detection techniques: based on Viola and
Jones algorithm and principal component
analysis." Advanced Computing & Communication
Technologies (ACCT), 2015 Fifth International
Conference on. IEEE, 2015.
[9]. Luh, Guan-Chun. "Face detection using combination of
skin color pixel detection and Viola-Jones face
detector." Machine Learning and Cybernetics (ICMLC),
2014 International Conference on. Vol. 1. IEEE, 2014.