IRJET- A Review on Various Techniques for Face Detection

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 10 | Oct 2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 108
A Review on Various Techniques for Face Detection
Nikita Raut1, Varun Dhuldhoya2
1,2Student, Dept. of Computer Engineering, KJ Somaiya College of Engineering, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Face detection has been an intriguing area in the
field of computer vision and machine learning, especially
during the past several years. Face detection is the process of
automatically locating human faces in digital images or
videos. Faces are one of the most prevailing and very specific
objects in visual media which can be tracked with the help of
face detection algorithms. In spite of the inherent complexity
of environment (illumination, pose, color coding, occlusion
etc.), we have seen the development of some intelligent and
accurate face detection algorithms. In this paper we have
made an attempt to study the various methods namely Viola
Jones, Genetic Algorithm, CNN, SVM, Hough transform with
convolution neural network, MMXfeatureextraction, Minmax
with embedding for face detection.
Key Words: face detection, viola and jones, soft
computing, artificial intelligence, neural networks,
machine learning, accuracy
1. INTRODUCTION
The subject of face detection and recognition is as old as
computer vision. Researches on automatic face detection
began from 1960s, but over the past decade, this
groundbreaking technology hasnotjustbecomeviable,it has
become widespread. Although other methods of
identification such as fingerprints, or iris scans can be more
accurate, face detection and recognition remains a major
focus of research because of its non-invasive nature. Face
detection is primarily the specific case of object detection.
Object detection deals with detecting instances of objects
such as people, automobiles, buildings or faces in an image
or a video.
Face detection refers to identifying people’s faces within
digital images or videos. It is the first basic step to all the
facial analysis methods like face recognition, face modeling,
face tracking, face verification etc. First, the face has to be
identified by the computer then only the further evaluation
can be done.
Face detection has a wide range of applications. It is being
used in photography as a way to help cameras autofocus on
people’s faces. Another application is smile recognition,
which helps people take pictures at the perfect moment in
order to capture smiles.
One of the most significant applications of face detection isa
facial recognition. A facial recognition system matches an
individual’s face instantly against a databaseofphotographs
in order to establish identity. Facial recognition tools are
used to secure phones and apps.
Face detection is being used by some companies in order to
spot people walk by a certain area. It can also be used to
predict age, gender and other factors. Biometric techniques
like identifying fingerprints are now being replaced by a
more efficient biometric technique- face detection.
Face recognition is one of the most relevant applications of
image analysis. It’s a true challenge to build an automated
system which equals human ability to recognize faces. One
aspect that has slowed down progress in this area is the lack
of base lining studies whose goal is not just the development
of complete systems but the analysis of how the different
pieces of a system contribute to its success. The goal of this
study then is to perform a systematic comparison of the
different techniques. In upcoming sections, we discusssome
recent approaches like Viola Jones, Genetic Algorithm, CNN,
SVM, Hough transform with convolution neural network,
MMX feature extraction, Min max with embedding which
perform accurate skin cancer detection with the help of
various algorithms and discuss the advantages and
disadvantages of the different techniques used.
2. LITERATURE REVIEW
A. Viola Jones
The Viola Jones Object Detection framework [9] was
proposed by Paul Viola and Michael Jones in 2001, where it
was used for detecting objects. Ever since, it has also been
used for face detection which is a sub-domain of object
detection. In [1], Vikram and Dr.S.Padmavathi present a
system based on Viola-Jones Cascade Object Detector, for
detecting faces in an image and locating the facial features
like eyes, nose and mouth in it. The advantage of this system
is that in addition to detecting faces it also detects the facial
features, which is otherwise a difficult task. In thisalgorithm
the distance between the two eyes is pre-defined,thus when
any image is given as an input this distancewill becompared
with the actual distance and therefore the eyes will be
detected. It detects nose and mouth in a similar way.
The Viola - Jones consists of 3 techniques for the facial parts
detection: 1) The Haar like features for the feature
extraction. The presence of the features in any of the input
image is determined by the Haar features 2) Ada boost
machine-learning method for detectingtheface.3) Cascade
classifier used to combine many of the features efficiently.
The Haar Cascade classifier is the main part of the face
detection. The Haar features have been allowed to convolve
and scan through the images from the top left corner and it
ends the detection of face processing at the right bottom
corner.

Then, the best features among the 1,60,000 features have
been found using Adaboost technique. And finally the
cascade method is used, that will pass the case if theimage is
of a human face, otherwise it will beconsideredasa non-face
region.
B. Skin Color Based Detection
Hakam et al. [5] propose a method to identify human skin
and faces from colored images where the human skin
colored pixels are detected via a matrix that contains a
collection of possible skin color values which represents the
skin color of most people. They have used the circularity
feature to distinguish human faces from other objects with
similar color.
One of the most important information in identifying a
human face is the color. However, the color is not enough to
detect and localize human faces in images since other parts
of the body often have the similar color. To solve this
problem, they have used the geometry of the face since it
differs from other parts of the body.
They created a database containing all the pixels that
represent the colors of the human skin. The bigger the
database, the better the accuracy but slower the
computational speed. The algorithm will compare all the
pixels in the database with each pixel in the possessing
image, and if the pixel in the processing image is close
enough to one of the pixels in the database image, then that
pixel will be labelled as 1, meaning it is a skin color, if not the
pixel will be labelled as 0 i.e. non-skin pixel.
The produced image goes into a set of modifications. The
first modification is removing the small areas which are
present away from the skin in the image and filling the holes
present on the skin. Another process employed is to createa
morphological structuring to remove the dots and the non-
uniformity in the image. . If the background pixels have
values similar to the skin color, then the backgroundand the
faces will be as one object. In order to solve this problem
edge detection was used, where they transformed the input
image into gray level format and then the edges in the image
were found using sobel and canny detectors.Oncewegetthe
objects, circularity is used to find how circular the object is,
so if the circularity value is above the threshold, a face has
been detected.
The main drawback of this method is the creation of the
reference matrix database. Another problem is the
possibility of errors. However, number of errors can be
reduced by considering more features todistinguishtheface
form other similar skin colored objects.
C. Detection using Min-Max Features Enhanced with
Locally Linear Embedding
In their paper Rahmat et al. [12] have improved the
classification accuracy in their previous work by integrating
a technique called Locally Linear Embedding (LLE).
Essentially, the MMX feature extraction method equally
divides an image into three horizontal, and three vertical
face regions. Then, the sum of horizontal and vertical face
regions was calculated and combined to form the final MMX
features. After MMX feature extractionLLEtransformationis
done. LLE resolves a low dimensional representation that
approximates the high dimension features this method is
very efficient and under certain circumstances it is possible
to derive an embedding only from the geometric situation
without looking at scale, distance or connectivity between
interval data.
First parameter is the nearest neighbors (K) to support the
global geometry of the data. Second parameterdmaxdefines
how much data has to be retained by LLE. Next a set of
weights is computed that can reconstruct each data point.
The embedding vector Y, is computed with the
aforementioned weights .After this classification is
performed using multilayer perceptron .Number of hidden
layers is varied using trial and error to get the best results
.Both hidden and output layers are use tangent sigmoid
activation function. The training algorithm used was Scale
conjugate Gradient as it shows excellent results in pattern
classification.
Two MLPS were taken one for LLE and one for non LLE data.
For non LLE data All the hidden units tested produced good
to excellent accuracy (above 80%),withsomevariationfrom
81.8% to 95.2% For LLE data all hidden units tested
produced excellent accuracy with minimum as 94.4% and
maximum as 96.4%
Based on the observations, it was found that MMX was
already a powerful feature representation algorithm even
before applying LLE features andafterapplyingLLEfeatures
the representation of MMX features had improved. This is
because the LLE algorithm canrepresentthefeaturesinsuch
a way that it maximizes the separability between face and
non-face cases.
D. Convolution Neural Networks
Chenghao Zheng et al. [6] propose a real time approach for
face detection by utilizing a single end-to-end deep neural
network with multi-scale feature maps, multi-scale prior
aspect ratios as well as confidence rectification. This system
has been proposed to overcome the major shortcoming of
the existing CNN-based face detectors such as need of high
computational cost and slow calculation. They have
proposed a face detection algorithm based on a single end-
to-end CNN model with similarities to (single shot multibox
detector) SSD. This model utilizes the three major
advantages of SDD 1) Multiscale feature maps which
guarantee the scaleinvariancetosome extent2)Convolution
detectors, which predict the objects 3) Default boxes and
aspect ratios.Forconfidence rectification,regressionmethod
is utilized instead of SoftMax function – a function used by
the SSD, because it is easier to determine a face if the box is
big enough with a higher resolution. In addition to this,
human intuition is used where the easily detected faces fall

in class 1, whereas difficult faces fall in class 0.5. Finally, in
order to improve the speed of the network, they have
reduced the network depth and the number of convolution
maps.
Yingxin Lou et al. [7], present a novel object detection
network based on a pre-trained and multi-feature VGG-16
network. This model has been proposed as animprovement
to faster R-CNN. According to [8], Faster R-CNN has certain
drawbacks like (1) It only adopts Conv5 feature map which
is not accurate enough for object detectionand bounding(3)
Faster R-CNN generates proposals using anchors which is
coarse for enclosing the various objects. (4) Cannot enclose
objects precisely. To overcome these shortcomings, they
have proposed the Pt-Net model. Their model is first
initialized by a pre-trained VGG-16 model and its own CNN,
then the image is provided to the pre-trained CNNs, the
outputs of the selected layers are aggregated into multi-
feature maps and the proposals are generated on Conv5
feature map via particle filter, the proposals are then
mapped to multi feature maps and the maps are cropped,
finally, the face is detected using a novel overlap loss
function.
E. Hough Transform with Convolution Neural
Networks
In their paper, Oshin et al. [9], develop novel feature
extractor method by the combination of Hough transform
and CNN defining face features matrix dimension.Adaptive
cascade DCNN represents a fine CNN model which is able to
extract high quality features which enhancesthepredictions
or initialization. When the initialization is obtained than
deep convolutional neural network provide two level
convolutional networks to rectify every landmark
individually received from local receptive fields. In Deep
CNN technique the major facial key points like eyes, nosetip,
mouth (M) are detected using fine dot point while in Hough
transformation the horizontal central line of higher
threshold is measured for the detection purpose.
Although both ACDCNN and Hough transform -CNN isbased
on CNN and follow the cascade frame work. We show that
the proposed system can achieve batter performance the
initialization results and the final results of the ACDCNN
The score is calculated for each region of interest of
detection and alignment, which matches the information
represented by the spectrum. It can be used for color
identification, inspectionandotherapplication,whichcanbe
used for comparison purpose. Forimageidentification,setof
references of image and their matching score obtained
during the matching process. It is also called as the true
positive rate of detection.
The faces are detected with more accuracy, in the point of
mean error and lower failure rate shows that the applied
method is comparatively more efficient than the earlier
methods. The accuracy of the detection of faces increases
ranging from 90% to 100% .and the accuracy of detection of
alignment of faces is also increases.
The variation in the illumination, illustrates the robustness
of the system. Therefore the future scope of the work is to
develop a computer based fast method for the facedetection
and face recognition. This work can be extended for the
study of other intra subject variations like expression,aging,
occlusion, gender classification and other biometricdevices.
F. Genetic Algorithm
According to [3], features extractedusing neural network are
not sufficient and have low accuracy. Face detection using
neural networks suffers from problems while extending the
classifier for newly added people and learning updated
information about the existing people
In this paper, Mr. Deepak et al. uses genetic algorithm
followed by Scale-invariant feature transform (or SIFT)
algorithm to improve the accuracybyfeatureextraction. The
main characteristics of the algorithm presented are the use
of a combination of classification results and its incremental
nature. The proposed algorithms consists of the following
stages: 1) Taking the input image 2)extracting the features
from image using genetic algorithm based SIFT algorithm
3)Detecting faces using feature vectors.
The first step of the genetic algorithm is selection. The
fitness value for each pixel is calculated and a fitness
function is used to select the child chromosomes. Then PCA-
mutation was performed on the selected pixels. The
mutation is necessary to insure high levels of diversityinthe
population. The next step is SIFT feature extraction. Then,
the SIFT algorithm which is an algorithm in computer vision
to detect and describe local features, will find the points
whose surrounding patches are distinctive
G. Support Vector Machines
Sanjeev et al. [10] propose an improved face detection
method for highly corrupted images. There are many
unwanted elements present in image which are commonly
known as noise that should be removed from an image for
any further processing. Mean and Median filter arenormally
used to reduce noise present in an image and for preserving
useful detail in the image. They have used Adaptive filtering
to reduce noise and preserve useful information in an image
as it is more selective than mean and median filtering.. Once
the noise from an image has been removed the image is sent
to the SVM trainer for identification of faces.
It is observed from this paper that the efficiency of using the
nose structure as an identifying feature for face detection
improves the success rate.
The drawback of this paper lies in the fact of its dependence
on visibility of faces after the filter procedure. Future scope
lies in improving the identificationratio, eitherbycombining
several feature extraction algorithms] or combining several
classifiers or combining the above two methods.

H. Detection using K-means clustering and Local
Binary Patterns (LBP) algorithm
Wisam et al. [11] propose a method where the targeted
images were collected, and K-means clustering was applied
to separate the colors that form these images. According to
this method, a face should be separatedfromitsbackground.
In the proposed method, the targeted imageswerecollected,
and K-means clustering was applied to separate the colors
that form these images. According to this method, a face
should be separated from its background.
The LBP algorithm is applied to the faces in the image and
then the LBP feature is divided into 6 groups3positiveand3
negative.
LBP is the process of returning image information in such a
way that will make it a unique binary pattern.
1) The image is divided into blocks, and the procedure is
applied separately for each block. 2) Each pixel should be
compared with eight neighbors starting from thetopleftina
clockwise or counterclockwise direction. 3) The difference
between center points and surrounding pixels is compared.
If the value of a neighbor pixel is lower than that of a center
pixel, then it should take a value of “0;” otherwise, the value
should be “1.” 4) The histogram for each block is computed,
which will result in a feature vector. 5) The histogram is
normalized (optional). 6) The histogram of all blocks is
concatenated, which will provide a unique binary
description for the entire image. 7) To classifyinformation,a
machine-learning algorithm can be used.
The LBP algorithm is applied to the faces in the image and
then the LBP feature is divided into 6 groups3positiveand3
negative .On the basis of the distributionforthesepoints,the
classifying element is able to separate the face from a non-
face area.
The first procedure of the algorithm involves segmenting an
image to separate the face from the scene by using the K-
means clustering code and reducing the time consumed for
face detection.
As stated previously, the main concept of face detection is to
identify the differences between the LBP for the face sample
and that for the block taken from the image. The similarity
between the two feature vectors is then calculated, and the
face and non-face are segregated. The final result of LBP on
an image can be represented by a vector, which contains a
histogram for every block.
The main concept of the proposed algorithm is to classify
these feature vectors to estimate the object it describes,
whether or not it is a face or non-face object. Results show
the efficiency of the algorithm in detecting faces with a high
rate of accuracy. Detecting faces in many poses, such as
turned and upside-down faces, is our main contribution to
the literature.
Technique
Used
Advantages Disadvantages
Viola-Jones
In addition to
detecting faces it
also detects facial
features
[19]Detector is
most effective
only on frontal
images of face.
It fails to detect
face in an image
with Partial
features. This
condition exists
when face is
slightly rotated.
Skin colorbased Simplicity
Creation of
reference
matrix dataset,
possibility of
errors in many
cases
Min-Max
Features
Enhanced with
Locally Linear
Embedding
Convolution
Neural Network
Improved
Accuracy
Time
consuming fora
huge dataset
Hough
Transform with
convolution
neural network
Genetic
Algorithm
Solves the
problem of
extending the
classifier for
newly added
people and
learning updated
informationabout
the existing
people
Poor results if
poor fitness
function chosen
SVM
K- means
Clustering and
Local Binary
Patterns

3. CONCLUSIONS
In this study, we have examined various techniquesthatgive
high accuracy for face detection. Face detection requires
various stages like pre-processing,segmentationandfeature
extraction. This survey focuses on different strategies like
Viola Jones, Genetic Algorithm, CNN, svm, Hough transform
with convolution neural network, MMX feature extraction,
Min max with locally linear embedding. As per the review,
each algorithm is found to have its advantages and
disadvantages. However, amongst the analysed algorithm,
the Min Max with locally linear embeddingalgorithmhasthe
least amount of disadvantages and thus, outweighs other
algorithms like Viola-Jones, K-mean clustering and neural
networks
REFERENCES
[1] Vikram, K., and S. Padmavathi. "Facial parts detection
using Viola Jones algorithm." Advanced Computing and
Communication Systems (ICACCS), 2017 4th International
Conference on. IEEE, 2017..
[2] Kavitha, J. C., and A. Suruliandi. "Texture and color
feature extraction for classificationofmelanoma usingSVM."
Computing Technologies and Intelligent Data Engineering
(ICCTIDE), International Conference on. IEEE, 2016.
[3] Mishra, Deepak, and Shaila Chugh. "Face detection using
genetic based SIFT algorithm." Green Computing,
Communication and Conservation of Energy (ICGCE), 2013
International Conference on. IEEE, 2013.
[4] Munteanu, Adrian, et al. "Wavelet‐based lossless
compression scheme with progressive transmission
capability." International Journal of Imaging Systems and
Technology 10.1 (1999): 76-85.
[5] Shehadeh, Hakam, Audai Al-khalaf, and Mahmood Al-
khassaweneh. "Human face detection using skin color
information." Electro/Information Technology (EIT), 2010
IEEE International Conference on. IEEE, 2010.
[6] Zheng, Chenghao,MenglongYang,andChengpengWang.
"A Real-Time Face Detector Based on an End-to-End CNN."
Computational Intelligence and Design (ISCID), 2017 10th
International Symposium on. Vol. 1. IEEE, 2017.
[7] Lou, Yingxin, et al. "PT-NET: Improve object and face
detection via a pre-trained CNN model." Signal and
Information Processing (GlobalSIP), 2017 IEEE Global
Conference on. IEEE, 2017.
[8] Ren, Shaoqing, et al. "Faster r-cnn: Towards real-time
object detection with region proposal networks." Advances
in neural information processing systems. 2015.
[9] Misra, Oshin, and Ajit Singh. "An approach to face
detection and alignment using hough transformation with
convolution neural network." Advances in Computing,
Communication,&Automation(ICACCA)(Fall),International
[10] Kumar, Sanjeev, Asutosh Kar, and Mahesh Chandra.
"SVM based adaptive Median filter design for face detection
in noisy images." Signal ProcessingandIntegratedNetworks
(SPIN), 2014 International Conference on. IEEE, 2014.
[11] Alobaidi, Wisam H., et al. "Face Detection Based on
Probability of Amplitude Distribution of Local Binary
Patterns Algorithm."
[12] Hidayat, Rahmat, et al. "Face Detection using Min-Max
Features Enhanced with Locally LinearEmbedding."(2018).
[13] Hamdy, Ahmed, Mohamed Elmahdy, and Maha
Elsabrouty. "Face detection using PCA and skin-tone
extraction for drowsy driver application." Information and
Communications Technology, 2007. ICICT 2007. ITI 5th
International Conference on. IEEE, 2007..
[14] Wong, Kwok-Wai, Kin-Man Lam, and Wan-Chi Siu. "An
efficient algorithm for human face detection and facial
feature extraction under different conditions." Pattern
Recognition 34.10 (2001): 1993-2004.
[15] Sharifara, Ali, Mohd Shafry Mohd Rahim, and Yasaman
Anisi. "A general review of human face detection including a
study of neural networks and Haar feature-based cascade
classifier in face detection." Biometrics and Security
Technologies (ISBAST), 2014 International Symposium on.
IEEE, 2014.
[16] Dang, Kirti,andShanuSharma."Reviewandcomparison
of face detection algorithms."CloudComputing,Data Science
& Engineering-Confluence, 2017 7th International
[17] Hashemi, Seyyed Mohammad Reza, and Maryam
Faridpour. "Evaluation of the algorithms of face
identification." Knowledge-Based Engineering and
Innovation (KBEI), 2015 2nd International Conference on.
IEEE, 2015.
[18] Fitrianingsih, Elis, Endang Setyati, and Luqman Zaman.
"Extraction of Eye and Mouth Features for Drowsiness Face
DetectionUsingNeural Network."Kinetik:GameTechnology,
Information System, Computer Network, Computing,
Electronics, and Control 3.2 (2018): 103-112.
[19] Anitha, C., M. K. Venkatesha, and B. Suryanarayana
Adiga. "High speed face detection and tracking."
International Journal of Soft Computing and Artificial
Intelligence, ISSN (2015): 84-90.

IRJET- A Review on Various Techniques for Face Detection

More Related Content

What's hot (20)

Similar to IRJET- A Review on Various Techniques for Face Detection (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- A Review on Various Techniques for Face Detection