DESIGN OF PORTBLE SECURITY SYSTEM USNIG FACE

RECOGNITION WITH BACK-PROPAGATION ALGORITHAM

AND MATLAB
M. Hareesh Babu 1, M. Bala Naga Bhushanamu2, B. Benarji3, Prof. M. Purnachandra Rao4
1 ,2 ,3 Research
Scholar, 4Professor & Research Guide
1,2,3,4
Dept. of Systems Design, Andhra University
[email protected], [email protected], [email protected], [email protected]

Abstract: In our globally connected world, threats from I. INTRODUCTION

various aspects are going at an alarming rate. These are
controlling with different security systems such as metal In this globalized world, security of a person or
detector, closed circuit cameras, and scanning systems. country is a major issue. An annual budget for the
All these aids are meant to recognize and identify the country and their people security has raising
explosives and others weapons. Here, it is more drastically by every nation. Even though the security
important to identify the particular person or persons, system in root levels are failing at small and non-
who were planning to distract the society or particular
recognitions events, which leads to financial and death
event. This paper is aimed to design that to control the
loses. So, security at these events need to be enhance
threats by identifying the suspected people by a simple
face recognition technique using simple PC or Laptop to control the various loses. All these situations can be
with the help of scientific software MATLAB and its overcome by utilizing the technology such as Artificial
neural network tool box. In general, all major events are Neural Network, Face recognition, Digital Image
fully securitized with well-developed protection systems Processing and Matlab software. The main aim of the
but only problem with non-major and small events, paper was to identify the persons from the face
where security systems are matter of financial issues. So, recognition technique and back-propagation
militants and other destroyers are taking advantage of algorithm, which is a part of neural network. In this
these situations and creating a panic and terror
paper, we have designed as that will collect the still
situations. This paper is also designed like that a PC or
photograph of the person, especially face part of the of
laptop with camera can be a face recognition system to
identify the suspected peoples and most wanted criminal. the person, by the personal computer or laptop’s built
By recognizing the people, we can mostly avoid the in camera and then fed to the MATLAB software,
threats from these people and dangerous situations. where captured image has processed according to
Neural network is a science that has been extensively back-propagation algorithm using neural network tool
applied to numerous pattern recognition problems such box. This image is stored in the database and also
as character recognition, object recognition, and face compared with the previous images from database
recognition, where this paper has programmed for face system, which has taken previously or collected from
recognition with the back-propagation algorithm and
internet database system. If images found to be
simulated with the software MATLAB and its neural
matched, then system immediately intimating that
network tool box. Here, the back propagation plays the
central operation role to get the key features were particular person has already listed or identified in this
extracted from the picture for training the network. database. Here, the data base system will readily stores
Since the major role of the project is mainly focusing on the suspected people and criminals images. In this
the training of the neural network, already extracted key manner, we can identifies the criminals, militant, and
features of the person’s image from the database were suspected persons at a small and non-recognized
taken for training the back-propagation network. Here, events at rural areas by simple using personal
we have taken 7 input units, 6 hidden units and 4 output computer and laptops.
units, contained back-propagation network. The output
unit, 4-output units, generates the 4-bit output which MATLAB is a high-level language and
gives the person identity. interactive environment for computations,
Keywords: Security, Back-propagation, Face recognition, visualizations, and programming. Meanwhile, you
Neural network, Images, MATLAB, Data learning rate, have a chance to analyze data, develop algorithms, and
Weight, Train. create models and applications. This language tool and
built-in mathematical functions will enable you to The face recognition task is a challenging because of
explore multiple approaches and reach a solution variability in the pose, orientation, location, and scale.
faster than with spreadsheets or traditional It is also depends on the facial expressions, lighting
programming languages, such as C or C++ or VC++. conditions, and age excreta. In this paper, we have
Main Key features of the MATLAB are as follows: (1) mostly considered about the three conditions as
Mathematical functions for linear algebra, statistics, follows: (1) Variations in image plane and pose (2)
Fourier analysis, filtering, optimization, numerical Facial expressions. (3) Lighting conditions and
integration, and solving ordinary differential background. Section one due to the faces in the image
equations. (2) Built-in graphics for visualizing data vary due to rotation, translation, and scaling of the
and tools for creating custom plots. (3) Development camera-pose or the face itself. Second sections due to
tools for improving code quality and maintainability the appearance of a face are largely affected by the
and maximizing performance. (4) Tools for building expression on the face. The presence or absence of
applications with custom graphical interfaces. (5) additional features such as glasses, breads, and
Functions for integrating MATLAB-based algorithms mustaches further add to this variability. The final
with external applications and languages such as C, section describes about the lighting conditions that
Java, .NET, and Microsoft Excel. (6) Creating Apps depend on the object and light source properties, and
with graphical user interfaces in MATLAB. affect the appearance of the face. The back ground
which defines the profile of the face is important and
Back-propagation is a common method of cannot be ignored. The whole process involves
training artificial neural networks and an abbreviated training the network with a set of pictures and after the
as "backward propagation of errors." From a desired training process is over, an arbitrary image is given as
output, the network learns from many inputs, similar input. The output is a set of binary values of yes or no,
to the way a child learns to identify a dog from indicating whether the image resembles any trained
examples of dogs. It is a supervised learning method picture or not.
and is a generalization of the delta rule. It requires a
dataset of the desired output for many inputs, making III. TRAINING APPROACH
up the training set. It is most useful for feed-forward
networks. Back-propagation requires that The training approach has taken into three stages. In
the activation function used by the artificial neurons or first stage, the back-propagation network is used to
"nodes" be differentiable. The back-propagation extract various unique features from every image, and
learning algorithm can be classified into two phases: learning vector quantization is used for training the
propagation and weight update. Each propagation network. Here, the back-propagation network is also
involves the following steps: (1) Forward propagation used as auto-associative network and each image is
of a training pattern's input through the neural network resized to 56 x 46 pixels then fed to back-propagation
in order to generate the propagation's output network as an input, which leads to be 2,576 input
activations. (2) Backward propagation of the units, 2,576 output units, and 200 hidden units.
propagation's output activations through the neural Further, each pixel value is converted into a bipolar to
network using the training pattern target in order to accept the values as ‘-1’ and if the pixel value is grater
generate the deltas of all output and hidden neurons. then 0.5 as ‘+1’ (Figs. 1 and 2 ).
For each weight-synapse follow the following steps:
(1) Multiply its output delta and input activation to get
the gradient of the weight. (2) Subtract a ratio
(percentage) of the gradient from the weight. This ratio
(percentage) influences the speed and quality of
learning; it is called the learning rate. The greater the
ratio, the faster the neuron trains; the lower the ratio,
the more accurate the training is. The sign of the
gradient of a weight indicates where the error is
increasing; this is why the weight must be updated in
the opposite direction. Phase 1 and 2 repeats until the
performance of the network is satisfactory.
Fig 1. BPN architecture.
II. DESIGN TECHNIQUE
 frontal and 7. Bottom left frontal. So, the dimensions
of the image are considerably reduced to 7-6-4 from
2576-50-6. Finally, the 6-bit output units of the image
converted into 4-bit output,

Fig 2. Hybrid BPN network. Fig 4. Block diagram for the entitled paper.

In second stage, the back-propagation network is used which gives the person identity. Here also the output
as a string network to act as a database system with a is converted from binary to hexadecimal to address the
2,576 input units, maximum of 200 hidden units, and memory location where the information of a person
6 output units. The output is converted from binary to has stored. To analyze the efficiency of the back-
hexadecimal to address the memory location of the propagation network for varying inputs to identify the
database, where the information about a person is correct person, we trained the network in both
stored. This 6-bit output gives the person identity (Fig. continuous valued and discrete value features (Figs 4
3 ). and 5 ).

Fig 5. BPN network after process the original image.

Fig 3. BPN network for process the original image.
IV. SIMULATION RESULT
In final stage, the key features of the image were only
extracted for the training the network because if we See Figs. 6,7,8,9, and 10.
give the picture as input to the network is not feasible
due to the size of the weight matrix is huge. To train
the network, we have extracted key features from the
database for train the back-propagation network with
7 input units, 6 hidden units, and 4 output units has
used. The seven attributes of the image features are 1.
Frontal or central 2. Upper frontal. 3. Lower frontal. 4. Fig 6. A set of training images.
Top right frontal. 5. Top left frontal. 6. Bottom right
 Fig 7. A set of testing images. Fig 9. Sample dataset for discrete input.

Fig 10. Binary data for persons’ identity.

V. CONCLUSION

Advantages in technology such as face recognition and

Image Processing has leads to overcome many
sensitive problems in security aspects such as
identifying the suspected people and enemies by
capturing the image and identify the validity of a
person by using the extracted features available in the
database. For practical implementation, more attention
should be given to the preprocessing stage to extract
the features directly from the image. Finally, the
Fig 8. Sample Data Set for continuous input. intention of the paper “Design of Portable Security
System Using Face Recognition with Back-
propagation Algorithm and MATLAB” has simulated
in PC and laptop for very low cost and advantage is
that it can mobilized to anywhere.
 VI. ACKNOWLEDGMENT.

We thank to our beloved Guide & Head of the

VIII. AUTHORS BIOGRAPHY.
Department Prof. M. Purnachandra Rao, who gave us
such an opportunity to work on NEURAL Prof. M.Purnachandra Rao received his
NETWORK, ARTIFICIAL INTELGENCE and M.Sc (Tech) in Electronics from Andhra
MATLAB SOFTWARE and his assistance on our University 1980 and he has awarded his
presentation. Ph.D. from Dept. of physics, A.U,
Visakhapatnam. Later on appointed has
VII. REFERENCES. Asst.Prof, Associate.Prof and Professor in
Physics Dept and Head of the Dept for
1. J.R. Anderon, Cognitive psychology and Its Systems Design & USIC.Research Work: Has a 30 years
Implications, 2nd edn. (Freeman, San Francisco, of experience in the field of SODAR and its applications.
1985) Hareesh Babu. M received his M.Sc (Tech)
in Electronics and Instrumentation from
2. D.H. Ballard, Parameter nets. Artif. Intell 22(3), 235- Andhra University 2008. Has a 2 years of
267(194). experience in Embedded Engineer at
Sartourious Mechatronics Ind Pvt.Ltd and he
3. E.A Feigenbaum, A. Barr, P.R.Cohen (eds) The is currently pursuing Ph.D in dept of Systems
Design, A.U, Visakhapatnam. Research
Handbook of Artificial Intelligence (Addision- Work: Has a 3 years of experience in the
Wesley, New York 1989) field of SODAR and its applications and
Matlab scientific software.
4. M.A. Eshera, K.S. Fu. “A Graph Distance Measure Publications:
for image analysis”; IEEE Trans. Syst. Man Cyberm. Design & Simulation of Graphical User Interface for SODAR
System using VC++: ITBI-2013.
SMC 14(3)
Graphical User Interface Design & Simulation for SODAR Signal
5. K.S Fu, Sequential methods in pattern recognition Generation Using Matlab: AP Science Congg-2013.
and matching of pictorial structures. IEEE Trans. Bala Naga Bhushanamu.M received his M.Sc
Comput (1968) in Electronics and M.Phil from Andhra
University. MS from BITS Pilani in 2011 and
6. J.L. Kolodener, Reconstruction memory: a he is currently pursuing Ph.D in dept of Systems
Design, A.U, Visakhapatnam. Research Work:
computer model. Cogn. Sci. 7(4), 281-328 (1938). Has a 3 years of experience in the field of
SODAR and its applications and Matlab
7. www.mathworks.in/help/vision/examples/face- scientific software.
detection-and-tracking.html Publications:
Design & Simulation of Graphical User Interface for SODAR
8. www.mathworks.in/help/nnet/ref/trainlm.html System using VC++: ITBI-2013.
Graphical User Interface Design & Simulation for SODAR Signal
9.http://mathworks.in/discovery/face_recognition.htm Generation Using Matlab: AP Science Congg-2013.
l

10. www.mathworks.in/help/nnet/ref/traingd.html Benarji. B received his M.Tech in Digital

Systems and Signal Processing from GITAM
University in 2010 and he is currently pursuing
Ph.D in dept of Electronics and Communication,
A.U, Visakhapatnam. Research Work: Has a 3
years of experience in the field of SODAR and
its applications and Matlab scientific software.
Publications:
Design & Simulation of Graphical User Interface for SODAR
System using VC++: ITBI-2013.
Graphical User Interface Design & Simulation for SODAR Signal
Generation Using Matlab: AP Science Congg-2013.