IRJET- 3-D Face Image Identification from Video Streaming using Map Reduce(Hadoop)

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1613
3-D Face Image identification from video streaming using Map
Reduce(Hadoop)
Mr. MOHAN SHIVAJI KAMBLE1, PROF.VAISHALI LONDHE2
1Department of Computer Engineering,
Ses’s Yadavrao Tasgaonkar Institute of Engineering and Technology.
2 Professor, Department of Computer Engineering,
Ses’s Yadavrao Tasgaonkar Institute of Engineering and Technology.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract—This paper focuses on the massive face
identification system. I want to recognize a face from a lot of
faces at public places. There are two of techniques are used to
hit the goal: one is the 3Dface identification technique andthe
other is the Hadoop detection technique. Face identification
technique find a similar face by 3D face features form mass
face data. The Hadoop is a parallel processing structure;itcan
boost the computation ability. From imulation outcome, it is
demonstrate that our algorithm is an efficient and accurate
method for huge face identification.
Key words—huge, Face identification, Hadoop, Features,
Parallel Processing.
1. INTRODUCTION:
Face identification is important task for several applications
on human being life. There are some examine waspublished
and described it below. I detect faces by using a hierarchical
knowledge-based method. I use three level resolutions in
their algorithm. The coarse-to-fine strategy reduce the
computation is an advantage in this method.
I also use local feature detector and random graph matching
techniques to create a probabilistic method can locate a face
in a scene. By using five features like two eyes, two nostril,
and Nose/lip junction to depict a typical face. I am trying to
define a facial template and relative distances of any pair
official features. This method can detect the testing object is
a human face or not.
I will use a general and complete face detection technique.It
is a valuable method for face detection and surveying
detecting faces in images.
For face image comparison, it is a hard work because it
needs a lot of working out and it cannot achieve 100%
accuracy. If I want to improve the comparison accuracythen
the multiple face indifferent angle can be achieve the goal.
Furthermore ,if I want search a people from public places, it
is more difficult because it is related to real time operation
problems. It is a massivecomputation. However, theparallel
processing technique increases computational capacity. For
a huge data, it needs massive computation ability. It needs
several computer works together to share the data.
Therefore, Hadoop structure is a suitable system for solving
the huge face identification difficulty.
The main aim of this paper is to develop a simple and
effective face identification system on them assize video
string.
II. Face Identification
A. an Object retrieve from the video string The first stage of
the face identification is to object retrieve. Once the
individual face obtained, the face recognize can be carry out
correctly. Figure 1 shows the flow chart of the object
extraction from the video stream. Since thevideoinputvideo
is color string, The system convert the color image into gray
scale image for reducing the computer computation.
Next step is to find out foreground object from the
background image by using multiple background extraction
modules. A media filter used to filter out the noise. An
exclusive bitmapping (EB) technique issued to convert the
image from gray scale to binary even on differently exposed
environments. Usually, the object image extraction will
causing some distortion, thus, image processing (IP)
techniqueslikethe dilation, opening, and closing techniques
are used to remedy the drawback. Finally, a small area
discard stage is used to delete those objects which are
belonging to noises because those areas are so small. In
addition to, by the person structure; I easily extract the
human face location by scale method of object. The system
compare the face location with the previous frame to make
sure the object is brand new or repeated. Once the objects
are all examined, I refresh the previousframe andrecordthe
new objects. At last, system stabilizes the object and run the
next frame. The details flowchart is described in figure 1.
Fig. 1 Object extraction and matching with the resultant
image.

A. Three-D Face Identification Algorithm
The 3D face identification is better than 2Dbecause 3D
method has lots of features. In this paper, I use the skincolor
and principle componentsscrutiny techniquestoachievethe
face identification. At first, I get the left and right imagesthat
include face components. Next, the color image
normalization stage used for further skin color
Detection. In order to obtain the face area, the several stages
include Erosion, Dilation, Labeling and Padding is used to
achieve the goal. However, the normalization stage is
necessary to make the face image standard. Finally, 3D face
identification can be completed by using following three
steps;
I. Merges the left and right image form a 3D image.
II. Execute the principle component analysis to get the
features.
III. Verify the features by comparing the input image and
database.
Figure 2 shows the details of the system.
Fig.2 Shows 3D face identification.
B. The features and decision function
The face features of this paper totally have 15numbers as
figure 3 shown. The important points of the face aredenoted
in the face image as the listed A to J. The fifteen features are
the distance between the each two points shows in figure 3.
For example the distance A and B express the feature of the
distance between two eyes. The feature IJ denotesthe width
of the mouth. The feature EK is the distance between the
center of the two eyes and mouth.
There are two steps for the decision function; one is the
error of the feature that is the difference between the input
image and database under the threshold.
The first step is check the features if one of features is over
the threshold that it is die out. The second step is calculating
the features error that is the error by compare the input
image with the database. Once the accumulationerror is the
minimum then it is the answer.
Fig 3 the sketch of the face features
III. Hadoop System
Since the 3D face identification from a video stream is
calculation utilization, a big data processing technique is
needed to achieve the goal. For our case, the Hadoop
technique is suitable for huge face identification.
A. Massive face identification structure
Massive face identification system is used to obtain main
information from lot of people and public places. For this
case, to find a person from massive people, it is a big data
problem need a technique to reduce the obscure and finish
the job.
Hadoop structure is an efficient parallel processing
technique can solve big data problem. Figure 4 shows the
Hadoop system that was used in the massive face
identification. It is clearly, there are two module blocks; one
is the subscriber verification and the other is subscriber
increase.
Fig.4 Shows Massive face identification structure in
Hadoop.

The task of the Subscriber increase module is receive the
subscriber information. The increase module pass through
the face image get from the camera and is used to add new
subscriber. For accuracy, the photographs include different
light surroundings, different background,differentangleand
facial expression. Those photographs are stored in the
database for the next time for face identificationcomparing.
The subscriber information recorded in the database
includes the name, id, age and photograph. The task of the
subscriber verification module issued to search and identify
the input face image with the database. It makes sure the
exactly people by comparing with all the database image,
and pick up the one is the least features error. This action
needs immense computation; therefore, it is the need of the
Map and reduces techniques under parallel process to
accelerate operation speed.
B. The implement system
The implement system of the Hadoop structure is shown in
Figure 5. It includes two processes; Mapping and reducing.
The nodes of the Hadoop structure maybe are failure. If a
node is on failure state then the task need to be reassigned.If
the task include some side effect then the share stateneedto
restart. For example, the nodes communicate with the
outside node, and then the share state must be hold untilthe
system restart.
Fig. 5 the implement system of the Map/Reduce
stream
In Hadoop structure, just only inherit the Map Reduce Base
class, it can offer map and reduce classes. Besides, the
Hadoop structure can automate operation in distribution
state when it registers Job. For example, I have 100 pictures
want to compare with the 3 million subscribers that each
subscriber has30 pictures in the database. If each picture
with length is 2048 by 2048, then the size is equal to
2048*2048=12.5MB. This is an enormouscomputationtask.
Hadoop adopt paralel operation technique. It usesthemulti-
server of the cluster to compare with several pictures. This
can reduce the time for identification. For example, I set 25
servers in a cluster and let the task of the Map is 25, the task
of the Reduce is 4. If the waiting subscriber is assign to the
25 server then the jobs of each server is 1/25 of all jobs.
The process of the Map is as below; It uses the subscriber id
and picture number as the input key. Use the subscriber
picture as the input value. Simultaneous, use the subscriber
id and picture number use as the output key. The
comparison results of the between the subscriber picture
and input image are used as the output value. The input of
the Reduce process is the subscriber id and picture number.
It target is to obtain a maximum value of the comparison
results by using input key and front half subscriberid.Onthe
other hand, the Reduce process use the subscriber id and
picture number as output key, and use the similarity of the
test image and subscriber as the output value.
According to the Reduce output, the system sorts the
likeness of all the subscribers to get the most similarity
subscriber id and picture number. When the subscribers
increased, the system just addsthe cluster class can support
more operation. This is the advantage of the big data
techniques.
IV. The Problems and Performance
A. The problems
There are some problems about this system as listed below:
(a) Due to 3D face identification have large features, It is
hard to predict which is the exact result.
(b) If the subscribers too big, the system can’t make sure to
obtain the accuracy result in real time process. For example,
there are 23 million peoples and Have90 pictures of each
people then the system had totally more than 2 billion
pictures. It is too big and hard to process in real time.
B. The performance
The performance of the big data system is based on the
parameters and structure of the process. There are some
methods can improve the performance and listed as below:
(A). I use to adjust the numbers of the Map and Reduce to
realize improve performance.
(b). Set a suitable path of the Name Node and Name Node
Federal technique to solve the Name Node breakdown
problem.
(c) Design a suitable H base to raise the performance of the
system inquiry by means of the problem analysis.
(d) According to the features of the face, adjust procedure
design of the Partitioner, grouping, sort and combiner in the
Shuffle of the Map Reduce process to raise the system
performance.

V. CONCLUSIONS
In this article, I use the distribution and parallel techniques
of the Hadoop to raise the computation speed for the
massive 3D face identification. Fromthesimulationresults,it
is established that the computation speedincreases3.5times
under the simulation condition on four computers a cluster
and one hundred subscribers. When the subscriberincrease,
the system only extend the cluster which can achieve the
high performance face identification at big subscriber.
REFERENCES
[1] M. E. Iber and M. L. Stone, “Low altitude wind shear
detection using airport surveillance radars,” IEEE
Aerospace.Electron. Syst. Mag., vol. 10, no. 6, 1995, pp. 3-9.
[2] P. Avis, “Surveillance and Canadian maritime domestic
Security,” Canada. Military J., 2003, pp. 9- 15.
[ 3] J. Lou, T. Tan, w. Hu, H. Yang and S. J. Maybank., “3-D
model based visual traffic surveillance,” Acta Automatica
Sinica, vol. 29, no. 3,2003, pp. 434-449.
[4] R. Cucchiara, C. Grana, A. Prati and G. Tardini, R. ,“Using
computer Vision techniques for dangerous situation
detection in domestic applications,” in Proc. IEEE Workshop
on Intelligent Distributed Surveillance Systems, London,
2004, pp 1-5
[5] "Architecture and algorithmsfortrackingfootballplayers
with multiple cameras," Vision, Image and SignalProcessing,
IEE Proceedings , vol.152, no.2, 8 April 2005, pp.232-241.
[6] C.Mitcheloni, G.L. Foresti and L. Snidaro, “A co-operative
multicamera Systemsfor video surveillance of parking lots,”
IEEE Symp.Intelligent Distributed Surveillance Systems,
London, 2003, pp. 193-205.
[7] M. Valera and S.A. Velastin, “Intelligent distributed
surveillance Systems: a review”, Vision, Image and Signal
Processing, IEE Proceedings, vol.152, no.2, April 2005,
pp.192-204.

IRJET- 3-D Face Image Identification from Video Streaming using Map Reduce(Hadoop)

More Related Content

What's hot (17)

Similar to IRJET- 3-D Face Image Identification from Video Streaming using Map Reduce(Hadoop) (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- 3-D Face Image Identification from Video Streaming using Map Reduce(Hadoop)