Proceedings of the Second International Conference on Innovative Mechanisms for Industry Applications (ICIMIA 2020)

IEEE Xplore Part Number: CFP20K58-ART; ISBN: 978-1-7281-4167-1

Face Detection and Recognition System

using Digital Image Processing
Gurlove Singh Amit Kumar Goel
School of Computing Science and Engineering Professor, School of Computing Science and
Galgotias University Engineering
Greater Noida, India Galgotias University
[email protected] Greater Noida, India
[email protected]

Abstract— While recognizing any individual,

the most important attribute is face. It serves as an II. DIFFERENT APPROACHES OF FACE
individual identity of everyone and therefore face RECOGNITION
recognition helps in authenticating any person’s There are basically two prevailing approaches to
identity using his personal characteristics. The the problem of face recognition namely, Geometric
whole procedure for authenticating any face data approach i.e. the feature based and the other one is
is sub-divided into two phases, in the first phase, the photometric approach i.e. the view based. As the
the face detection is done quickly except for those field of face recognition fascinated many researchers
cases in which the object is placed quite far, resulting which there were many contrasting
followed by this the second phase is initiated in algorithms developed, out of which three of them
which the face is recognized as an individual. Then have been widely studied in the literature of face
the whole process is repeated thereby helping in recognition[3][4].
developing a face recognition model which is We can classify the recognition algorithm into two
considered to be one of the most extremely main approaches:
deliberated biometric technology. Basically, there 1) Geometric: This approach mainly deals with the
are two type of techniques that are currently being spatial correlation uniting the profile (i.e. face)
followed in face recognition pattern that is, the features, also we can simply that dimensional
Eigenface method and the Fisherface method. The
layout of the facial attributes. Some of the main
Eigenface method basically make use of the PCA
(Principal Component Analysis) to minimize the geometrical attributes of a human face are nose,
face dimensional space of the facial features. The eyes and the mouth. Based on these attributes
area of concern of this paper is using the digital firstly the face is categorized and then based on
image processing to develop a face recognition these attributes respective spatial intervals and
system. the respective associated gradients are estimated,
thereby advancing the process of face
Keywords— Face recognition system, eigenface
method, principal component analysis (PCA), digital recognition. (Fig.1)
image processing.
I. INTRODUCTION
After the face is detected the main task of face
recognition system starts as to identify the known or
unknown face and act accordingly. Often people are
mistaken by the term face detection whereas face
recognition means on the other hand is to authenticate
a given face data based on the stored face data in the
database. Once the face data matches with the
database then the system is authenticated [1][2]. 2) Photometric stereo: It is a methodology of
There are different approaches that are being computer vision technology which mainly
followed in the whole process of face recognition. recuperates the structure of an underlying object
Each process has its own pros and cons, and has also
from the images that were shot in varying
some limitations which makes it different from other
approaches [5]. circumstances that were affected by the lighting
environment [7]. An arrangement of the surface
standards shown by the slope chart that finally

978-1-7281-4167-1/20/$31.00 ©2020 IEEE 348

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 19:59:55 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Innovative Mechanisms for Industry Applications (ICIMIA 2020)
IEEE Xplore Part Number: CFP20K58-ART; ISBN: 978-1-7281-4167-1

elucidate the retrieved entity’s configuration. 3) Localization: After the process of classification,
(Fig. 2) the bounding box is thus used to localize the
searched human face from the results of the
trained neural network. There are numerous
attributes of face on which work has been done,
some of them are: Position, scale, orientation
and illumination.

Fig. 2 – Photometric stereo image

III. FACE DETECTION

One of the sub divided image frame makes one
class i.e. the one consisting the faces in the image,
which marks the first step towards the process of face
detection[2][9]. It is inconvenient because in spite of Till now we have seen how an image is processed and
the congruity exist among faces but several factors how it is being checked for the presence of any face
like age, skin color and facial expression can vary or not. Now, we will talk more about face detection.
considerably. Then this problem is furthermore After the facial attributes are detected in any image
intricate by the arrival of factors like environment all the rest of objects are ignored and our primary
factors affecting light, risk of imitation and also concern is with the face. Face detection can be also
probability of limited obstruction in image. The face referred as face localization, where main aim is to
detection system that can easily recognize any face find location and size of the known no. of faces
from a given image that too under any circumstance present in an image. Primarily there are two
with any kind of lighting environment is thus approaches that are being followed to recognize facial
considered as the finest face detection system.The part in an image, the first approach is the feature base
function of the face detection system can be further approach and the second approach is the image base
bifurcated into two phases. Phase one consists of approach [6][8][10].
classification, in which the system based on the input 1) Feature based approach:
that was in the form of some random images and if 1.1) Active Shape Model: The ASMs start
the face is present in the image the output comes in
focusing upon the compounded fixed
the form of yes or no. Face localization is the second
phase in which for a given input image it shows a attributes that are higher in physical level
bounding box which comprise the dimensions of appearance, in simple words we can say that
exact location of the face in the image. as soon as any input image is given, the task
The process of face detection system is sub-divided of ASM is to automatically locate the
as follows: benchmark points, which will elucidate the
1) Pre-Processing: Before feeding any image to the appearance of any statistically craved entity
network it is processed properly to lower down in a picture. Some of the physical attributes
the variability. Frontal faces that are comprised that are higher in appearance are eyes, lips,
in the front view of the image is thus obtained by nose, mouth and eyebrows [4]. A training set
cropping the images that contain the human composed of images in which the landmarks
faces[12].On completion of the above step, are manually marked, is used for building a
standard algorithms are used to correct the statistical facial model which is also a part of
lighting of the cropped images. the training stage of an ASM.
2) Classification: To categorize any image as faces 1.2) Low level Analysis: Some of the essential
or non faces, neural networks are implemented low-level visible attributes of a human face
by training on these examples. For the process of are color, motion, intensity and edges etc.
classification, we have combined the MATLAB Amongst these skin color plays an important
NN toolbox along with the basic implementation role of an attribute to detect a face. As
of the neural networks. compared to the processing of all other
facial attributes, color processing is

978-1-7281-4167-1/20/$31.00 ©2020 IEEE 349

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 19:59:55 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Innovative Mechanisms for Industry Applications (ICIMIA 2020)
IEEE Xplore Part Number: CFP20K58-ART; ISBN: 978-1-7281-4167-1

considered better as it is faster. Under a given image the proffered technique

specific lighting circumstances the color is applies 40 various Gabor filters. The
evenly oriented. Now for the motion outcome of this comes out to be images
estimation it is easy as only a translation with varying angles and positions are
model is required. On the other hand, collected. In the next step the fiducial
detecting a human face using the skin color points are marked that are of maximum
attribute has sundry issues like color intensity. According to the distance
depiction of an image that contains the face these points are reduced and after that
that is basically acquired by a camera is the distance amongst the reduced points
derived from numerous features such as is calculated with the use of the distance
moving entities, atmospheric light and many formula. The final step is to compare
more. To a great extent there are basically the distance values with those in the
three types of algorithms that are used for database. If the values match that means
detecting a face namely, HIS, YCbCr, and the face is detected. Gabor filter
RGB. There are three essential steps that equation:
need to be followed in the implementation of
these algorithms, that are
a) Categorize the skin area in the color
scope,
b) Put in a threshold to disguise the skin
area,
c) For detecting the face image draw a
bounding box. 2) Image based approach:
1.3) Feature Analysis: The primary focus of these
2.1) Neural Network: In the recent years neural
algorithms is to systemic attributes that
network is getting more attention in regard
remain into the existence even if the angle of
to the pattern recognition problem like
viewor the lighting circumstances vary, and
Object recognition, OCR and self-driving
then make use of these to detect and pinpoint
robot. Neural network turns out to be very
the human faces. These approaches were
feasible for the problem of face detection.
mainly used for the problem of face
The only drawback of this system is that, it
localization.
requires a very fine-tuned network
1.3.1) Feature Searching:
architecture to get an exceptional
a) Viola Jones Method: This technique
performance. Propp and Samal designed one
was mainly given by Paul Viola and
of the first neural network for face detection,
Michael Jones to detect any object by
which contained 1024 input units and two
minimizing the computational time
output units. Feraud and Bernier came up
thereby achieving the high accuracy rate
with an advanced method for facial detection
in detecting face. This proposed method
with the use of auto associative neural
proved to be 15 times faster than the networks. After this model Lin et al. came
existing systems with an accuracy of
up with new approach of face detection by
around 95%with approx. 17 frames per
making the use of probabilistic decision-
second. This approach makes use of
based neural network. Learning rules and
Haar-like attributes and is built on the feasibility explanation are comprised within
notion of integral image, on the other the Radial Basis Function, which is related
hand boosting algorithm AdaBoost is with theprobabilistic decision-based neural
used to minimize incompletely arranged
network.
attributes. It is basically used on the 2.2) Linear Sub-Space Method (LSSM):
gray-scale images.
Eigen Face Method (EFM): Kohonen took
b) Gabor Feature Method: This technique
the initiative of implementing the Eigen
was proposed by Sharif et al which uses
vectors for the problem of face recognition,
Elastic Bunch Graph Map (EBGM)
by making use of simple neural network; for
algorithm for detecting the face with
recognizing a human face in aligned and
implementation of the Gabor filters. On

978-1-7281-4167-1/20/$31.00 ©2020 IEEE 350

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 19:59:55 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Innovative Mechanisms for Industry Applications (ICIMIA 2020)
IEEE Xplore Part Number: CFP20K58-ART; ISBN: 978-1-7281-4167-1

normalized position. Further advancement in strong with the potential of data reduction and
this was done by Kirby and Sirovich by interpretation.
making use of Linear Encoding. A vector of IV. DIGITAL IMAGE PROCESSING
size m*n represent the images and then the
Curiosity for the digital image processing technique
mean square error is minimized.
basically originated by taking into concern two
primary implementations [7]:
a) Refining the pictorial data for proper
interpretation and explanation.
b) Organizing the scene information for
independent recognition to be done by the
device.

The main focus in the second application area is

fetching the image data in computer process able
conformation [11].
It has a wide area of practical implementation which
include automatic character recognition, in the field
of military for authentication purpose and
automatically authenticating the fingerprints, and
many more applications.
An image can be referred to as a 2D light intensity
function f(x,y), where the spatial coordinates are
denoted by (x, y). The brightness or gray levels of the
image is proportional to the value of f at any point (x,
y). The image processing is broadly categorized into
three types namely,
1) Low level processing (LLP)
2) Medium level processing (MLP)
3) High level processing (HLP)
The first type of processing i.e. low-level processing
only deals with only fundamental operations such as
rotating an image, reading an image, resizing an
image, histogram equalization and also converting the
RGB to gray-level. After this a raw image is derived.
The image thus obtained is a raw image [2][3].
Fetching the area of interest from the outcomes of the
LLP is mainly handled by MLP. The segmentation
process i.e., identification of edges is also handled by
the medium level processing. The task of the high-
level processing is to bring advancement in the
medium-level processing by adding AI to it.
Fig. 4: Eigen Face Method 1) Fundamental steps in image processing:
2.3) Statistical Approach: a) Image acquisition: obtaining a digital
figure in the form of a picture.
Principal Component Analysis (PCA): Immediately b) Image pre-processing: its role is to
identifying the variation amongst the human face is refine the pictures in such a way, so that
not so obvious but then PCA comes into the picture the success rate is high in other phases.
and proves to be the appropriate procedure to deal
with the problem of face recognition. PCA does not c) Image segmentation: its role is to form
work on the age-old principle of categorizing the integral part of the entity from the
geometrical attributes but on the other hand check segregated images.
that which all factors will affect the faces that are d) Image description: its role is to fetch
present in a picture. In the system of face recognition, attributes that are of higher significance
these factors are commonly known as Eigen faces and and plays a vital role in differentiating
the reason behind is because when they are organized
one class from the other.
they look similar to the human face. PCA was used
on a great extent in the field of pattern recognition for
classification problem. PCA proves itself to be very

978-1-7281-4167-1/20/$31.00 ©2020 IEEE 351

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 19:59:55 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Innovative Mechanisms for Industry Applications (ICIMIA 2020)
IEEE Xplore Part Number: CFP20K58-ART; ISBN: 978-1-7281-4167-1

e) Image recognition: based on the data Machines and Home security systems. There are
provided by the description, allocate a many advancements possible in this field as this
label to it. serves as insight into what the future will be in the
field of computer vision.

REFERENCES
[1] https://www.engpaper.com/face-recognition-2018.htm
[2] Manoharan, Samuel. "Image Detection, Classification and
Recognition for Leak Detection in Automobiles." Journal of
Innovative Image Processing (JIIP) 1, no. 02 (2019): 61-70.
[3] https://www.semanticscholar.org/paper/Face-Detection-
using-Digital-Image-Processing-Jindal-
Gupta/a0a9390e14beb38c504473c3adc857f8faeaebd2
[4] https://www.academia.edu/41880923/Recognition_of_faces_
and_Greebles_in_3-month-
CONCLUSION old_infants_Influence_of_temperament_and_cognitive_abiliti
es
Since the no. of Eigen faces to be used is restricted in [5] https://www.researchgate.net/publication/266873152_An_Ap
PCA transformation that’s why the system did not plication_of_Face_Recognition_System_using_Image_Proce
have an accuracy of more than 90% for both manual ssing_and_Neural_Networks
and automatic face recognition. A further work that [6] https://www.researchgate.net/publication/228963208_A_MA
needs to be done is in the field of fully automated TLAB_based_Face_Recognition_System_using_Image_Proc
essing_and_Neural_Networks
frontal view face detection system which when
[7] http://www.ijesi.org/papers/Vol(7)i4/Version-
displayed virtually shows a perfect accuracy. The 5/C0704051728.pdf
real-world performance of this designed system will [8] Rahmani, Dr. Md. Khalid & Ansari, M.A. & Goel, Amit.
be far more precise. In view of attaining a high (2015). An efficient indexing algorithm for CBIR.
accuracy rate the designed and developed system was Proceedings - 2015 IEEE International Conference on
Computational Intelligence and Communication Technology,
not adequately strong. One of the main reasons CICT 2015. 73-77. 10.1109/CICT.2015.165.
behind this flaw is that the sub-system of the face [9] Qiong Cao, Li Shen, Weidi Xie, Omkar M Parkhi, and
recognition system does not exhibit minute changes Andrew Zisserman. 2018. VGGFace2: A dataset for
in degree of steadiness to scale or rotation of the recognising faces across pose and age. In IEEE Conference
segmented face image. The performance of this on Automatic Face and Gesture Recognition.
system can be compared with the manual face [10] Junjue Wang, Brandon Amos, Anupam Das, Padmanabhan
Pillai, Norman Sadeh, and Mahadev Satyanarayanan. 2018.
detection only if we integrate the eye detection Enabling Live Video Analytics with a Scalable and Privacy-
system with the developed system. Aware Framework. ACM Transactions on Multimedia
The other executed applications exhibited an Computing, Communications, and Applications (TOMM) 14,
3s (2018), 64.
exceptional result and returned exceptionally good on
the PCA technique and distorted arrangement. [11] Arsha Nagrani, Samuel Albanie, and Andrew Zisserman.
2018. Seeing Voices and Hearing Faces: Cross-modal
Appropriate use of this developed system of face biometric matching. arXiv preprint arXiv:1804.00326 (2018).
recognition and detection is in the field of [12] Weiyang Liu, Yan-Ming Zhang, Xingguo Li, Zhiding Yu, Bo
surveillance and mugshot matching. This system after Dai, Tuo Zhao, and Le Song. 2017. Deep Hyperspherical
integration with the eye detection system could be Learning. In Advances in Neural Information Processing
Systems. 3953–3963.
mainly serve as an authentication system in the ATM

978-1-7281-4167-1/20/$31.00 ©2020 IEEE 352

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 14,2024 at 19:59:55 UTC from IEEE Xplore. Restrictions apply.