Automated Personalized Face Detection Using Viola Detector

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 – 0882

32
Volume 6, Issue 1, January 2017

Automated Personalized Face Detection using Viola Detector

1
Mr. Amit Kumar,2Mr. Anil Rao
Department of Computer Science Engineering,IET, Alwar, Rajasthan
1
amitcse86@gmail.com,2anil.alw@gmail.com

Abstract • Representation of Integral image for feature extraction

In correspondence with rapid population magnification all • Ada-Boost for face detection

over the map the survival, facial detection technology plays a • Cascade classification for fast rejection of non-face sub-

more preponderant paramount and inductively authorizing role windows.

in our circadian lives, namely in a well-kenned aspects as The projected technique of face detection in this paper is

antihero investigations, notarize systems, ingression systems, divided into two steps. In the first step, Viola and Jones

etc. Face detection perspicuity is influenced by antithetical algorithm is used to detect faces in the image that is stored

variations in relegating a target repine in an image; this selected from hard disk drive, and the result of this step will be

manage include pose variation and facial occlusion. The the applicant face in the image [2].

overall aim of this research is to design a strategy to increment So face detection process takes more time when is a time the

the efficiency & optimize the results to detect the region of size of the image to be processed increases.

culled face in an image. The performance of Automated


Personalized Viola Detection Algorithm in terms of efficacy II. RELATED WORK
and time taken to find the desired position will be analyzed. Face Detection activities have increased popularity since the
tardy nineties. To improve detection rate many researchers
Keywords: Face detection, Haar featureViola-Jones, Integral presents different techniques for identifying human faces. But
image, OpenCV, Adaboost
Viola-Jones face detection using Adaboost algorithm has
shown a better outcomes with15 frames/second (fps) on

I. INTRODUCTION images of resolution 320×288 as compared to precedent


technique while retaining the detection rate. [1]. Software
Face detection algorithm is used to determine and display the
technique which work with multi-threading in the optimized
region of the face in applications such as video surveillance,
OpenCV implementation [7] in CPU centered system is
security, smart card application etc. In latest technology the
efficient to achieve 1.78 fps on VGA size images [8] and 14.2
vogue and interest of face detection play a consequential role
fps on preferably minuscule images of resolution 256x192
in terms of amended detection speed and precision due to its
[9].Besides that, Cho et al. [10] aside from shown a FPGA-
demand and latest applications. In particular, the essential
predicated face detection system with Haar classifiers utilizing
work by Viola-Jones has made face detection virtually
buffers and distinctive frame grabbers to expedite the
achievable with high detection speed by utilizing a well-
processing, which is capable of work with 6.55 fps for VGA
trained classifier [1]. Paul Viola and Michael Jones presented
image. Viola using KNN classifier algorithm show the
a technique for face detection which doesn’t consume much
accuracy of the projected face and emotion reorganization
computation time. This method implements face detection
system is 94.5 to 97 % [5].A framework of occluded face
techniques which decrease computation time than any other
detection based on Adaboost and Haar features enhances the
antecedent methods.
detection rate at about 99%, which meets the desires for in
The algorithm divided into mainly 4 Process:
ATM security applications [4].
• Haar like Features Selection

www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 – 0882
33
Volume 6, Issue 1, January 2017

A. Haar Features • The nose bridge region is brighter than the eyes.
The Haar features are the impuissant features predicated on Useful domain knowledge:
the gray image. The features are used for image intensity • Location - Size: eyes & nose bridge region
expression of rectangular region adjoining some location in • Value: darker / brighter
the image, and calculate the intensity difference. Haar-like • Rectangle features:
features treated as digital image a feature that is used in object Computed result = Σ (pixels in black region) - Σ (pixels in
recognition. These features were used in the first time face white
detector. Viola and Jones [8] developed Haar-like features by region)
adapting the idea of using Haar wavelets. The size of these
rectangular features is not fixed, but the black and white
square shape and size are always the same. The conception of
Haar-like feature is to consider the nearby rectangular regions
at a definite location in a detection window, sums up the pixel
intensities in each region and compute the distinction between
these sums. This difference is then utilized as a feature Figure 2: Extracting Haar feature from image
A.
replication to categorize subsections of an image. Figure 1
B. Integral Image Representation
shows three kinds of rectangle Haar features. The measure of a
In an integral image we can measure the value at pixel(x,y) is
two-rectangle feature is the distinction between the sums of
sum of pixels above and to the left of (x,y). Integral image is
the pixels within two rectangular sections [15]. These
used to calculate sum of all pixels inside any given rectangle
rectangular sections have the similar size and shape and
using only four values at the corners of the rectangle.
formed either horizontally or vertically adjacent. A three
rectangle feature measures the sum within two outer
rectangles subtracted from the sum in a center rectangle.
Conclusively a four-rectangle feature measures the distinction
between diagonal sets of rectangles [3].

Figure 3: Integral Image Representation

The sum of the pixels for rectangle D can be


computed as shown below. The value of the integral image
For location 1 is the sum of the pixels in rectangle A.
For location 2 is A+B
Figure 1: Haar like feature
For location 3 is A+C
All human faces contain specific related properties. These
For location 4 is A+B+C+D.
properties are used to build definite features called as Haar
The sum of values for D can be computed as 4+1-(2+3).
Features.
D=A+(A+B+C+D)-(A+B+A+C)
The related properties for a human face are:
• The eyes region is blacker than the upper-cheeks.

www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 – 0882
34
Volume 6, Issue 1, January 2017

processing. So Cascade classifier performs operation in


several stages to identify a face image and discarded irrelevant
images.

Figure 4: image with corresponding result of integral image

C. Adaboost for aggressive feature selection


As indicated that there can be nearly 16000+ feature values
Figure 5: Cascade classifier process
contained by a detector at 24x24 base resolution which need to
be measured. But some features are relevant and irrelevant so
we need to discard irrelevant features. Only relevant feature III. METHODOLOGY
will be used to identify a face. The fundamental quandary to be solved is to implement an

So we use Adaboost algorithm to find only the best features algorithm for detection of faces in an image. This quandary

among all these 16000+ features. Then a weighted can be facilely resolved by human beings. However it is very

combination of all these features is used in identifying and arduous for a machine to solve this task prosperously in facile

deciding any given window has a face or not. manner. In order to make this arduous task more facile Viola

These features are also called as weak classifier. Adaboost Jones limits them to complete view frontal upright faces. That

constructs a strong classifier as a linear combination of these is, with the aim of detecting the whole face must point towards

weak classifiers. the camera and it should not be tilted in any side.

F(x)= α1f1(x)+ α2f2(x)+ α3f3(x)+… Conventional Viola Jones has several problems which are
mentioned below:
Where f1(x), f2(x)…is a weak classifier and F(x) is strong
 In terms of speed and accuracy existing Viola Jones
classifier. The term α1, α2, α3 is distributed weight over image
algorithm is quite slow.
[3].
 Existing Images does not work well with tilde images and
An image should contain one or more faces; it is confirm that
distortion images.
an extreme large amount of the assessed sub-windows would
 Viola Jones does not detect Black Faces.
still be non-faces. So the algorithm should focus on discarding
 It does not work well with illuminating conditions.
non-faces rapidly and expend more on time on possible face
 Feature Selection for calculating of Integral Image need to
regions. Therefore a single strong classifier designed from
be optimized.
linear combination of all best features.
 Multiple detection of faces(Due to overlapping sub
windows)
D. Cascade Classifier
To solve these problems, we will implement an Automated
Viola-Jones algorithm works with cascaded classifiers.
Personalized Viola Detector.
Cascade classifier performs operations in several stages each
containing a strong classifier. The function of each stage is
used to identify whether a sub window is face or not. A given IV. PROPOSED METHOD
sub window is instantly discarded as not a face if it fails in any In this proposed method an image consist of multiple or single
of the stage [3]. face will be taken as input. In terms of speed and reliability
If the input region fails to pass the threshold of a stage, the for face detection from an image we will calculate ratio of sum
cascade classifier will immediately reject the region as a face. of black rectangle and sum of white rectangle and with this I
If a region passes all stages successfully, it will be classified will utilize some better features except those which are
as an applicant of face, which may be distinguished by further utilized in precedent algorithm. These features are better in

www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 – 0882
35
Volume 6, Issue 1, January 2017

identifying ebony and skewed faces. They additionally work face and it is more useful for face recognition and forms our
for tilde faces in some images. classifier. Now we are searching for a database that contains
To implement the haar cascade classifier, Open Source colorful and complex images to run our program to implement
Computer Vision Library (OpenCv) [11] will be used. It is Automated Personalized Viola Detector.
formerly presented by Paula voila and Michael jones [12].
Haar features are the important part of the haar cascade
REFERENCES
classifier for the face detection process. Haar features are used
[1] Adrian Wong Yoong Wai, Shahirina Mohd Tahir, Yoong
to identify the presence of feature in input image. Each
Choon Chang”GPU Acceleration of Real Time Viola-Jones
features is calculated by subtracting the sum of pixels under
Face Detection”IEEE International Conference on Control
white rectangle from the sum of pixels under black rectangle
System, Computing and Engineering, 27 - 29 November 2015,
and produce a single value. The integral image approach is
Penang, Malaysia
used to compute rectangle features.
[2] Mohammad Da'san(1), Amin Alqudah(1) and Olivier
Debeir(2)”Face Detection using Viola and Jones Method and
Neural Networks”International Conference on Information
and Communication Technology Research (ICTRC2015)
[3] Vaibhav. V. Dakre, Pravin. G. Gawande. "An efficient
technique of multimodal biometrics using fusion of face and
iris features", 2016 Conference on Advances in Signal
Processing (CASP), 2016
[4] Hua Wang1, Xin Gu1, Xiao Li1, Zhe Li1, Jun
Ni2”Occluded Face Detection Based on Adaboost
Technology”2015 Eighth International Conference on Internet
Computing for Science and Engineering
[5] Dolly Reney”An Efficient Method to Face and Emotion
Detection”2015 Fifth International Conference on
Communication Systems and Network Technologies
[6] P. Ithaya Rani”Robust Real Time Face detection
automatically from video sequence based on Haar
features”2014 International Conference on Communication
and Network Technologies (ICCNT)
Figure 6: Automated Personalized Viola Process [7] Dice Holdings, Inc., (2015). Open Source Computer
Vision Library, [Online]. Available:
V. CONCLUSION &FUTURE WORK http://sourceforge.net/projects/opencvlibrary/.
In this work, an incipient modified Viola Jones Algorithm has [8] J. P. Harvey, “GPU Acceleration of Object Classification
been proposed to detect black and distorted faces in an image Algorithms Using NVIDIA CUDA,” Master’s Thesis,
which can work in any of the illuminating conditions. The face Rochester Institute of Technology, Rochester, NY, United
detection approach predicated on” Automated Personalized States, 2009.
Viola” which uses incipient pixel predicated features which [9] A. Dutta et.al, “Real Time Face Tracking and Recognition
are different from subsisting Viola Jones predicated methods. (RTFTR),” Tribhuvan University-Institute of Engineering,
Face database called as CMU typically contain grayscale Latipur, Nepal, 2009.
images and it is not well for colorful algorithms. Databases [10] J. Cho et.al, “FPGA-Based Face Detection System Using
such as FERET, grimace, face only have images with single Haar Classifiers,” Proceeding of the ACM/SIGDA

www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 – 0882
36
Volume 6, Issue 1, January 2017

International symposium on Field Programmable Gate Arrays.


New York, NY, USA: ACM, 2009, pp. 103–112.
[11] L. Sirovich and M. Meytlis. Symmetry, probability, and
recognition infacespace.PNAS - Proceedings of the National
Academy of Sciences,106(17):6895–6899, April 2009.
[12] S. Lawrence, C. Giles, A. Tsoi, and A. Back, ―Face
Recognition: A Convolution Neural Network Approach, IEEE
Trans. On pp. 98 - 113, 1997.
[13] Paul Viola and Michael Jones”Rapid Object Detection
using a Boosted Cascade of Simple Features”Conference on
Computer Vision and Pattern Recognition2001
[14] Yi-Qing Wang”An Analysis of the Viola-Jones Face
Detection Algorithm”IPOL Journal Image Processing On
Line, Vol-4, 128-148
[15] Hongguang Sun. "Fast Face Detection Based on
Enhanced AdaBoost", Lecture Notes in Electrical
Engineering, 2012

www.ijsret.org

You might also like