ACKNOWLEDGEMENT

I would like to express my sincere thanks to Prof. Dr. Shashidhar Ram Joshi for
providing precious suggestions about project concept and its inception.
I would also like to thank Prof. Dr. Subarna Shakya and Dr. Surendra Shrestha for
their constant support and encouragement on research activity during Masters
program.

Table of Contents
ACKNOWLEDGEMENT .................................................................................................. I
LIST OF FIGURES ......................................................................................................... III
LIST OF ABBREVIATION............................................................................................ IV
1.

INTRODUCTION...................................................................................................... 1
1.1. Background ........................................................................................................... 1
1.1.1.PCA Algorithm ............................................................................................ 1
1.1.2.Artificial Neural Network ............................................................................ 2
1.2. Motivation ............................................................................................................. 2
1.3. Problem Definition................................................................................................ 2
1.4. Objective ............................................................................................................... 3
1.5. Applications .......................................................................................................... 3

2.

LITERATURE REVIEW ......................................................................................... 4

3.

METHODOLOGY .................................................................................................... 5
3.1. Read the Face Image from Training Database ...................................................... 5
3.2. Preprocessing ........................................................................................................ 5
3.2.1.Average Filtering ......................................................................................... 5
3.2.2.Histogram Equalization ............................................................................... 6
3.2.3.Bi-Cubic Interpolation ................................................................................. 6
3.3. PCA for Feature Extraction .................................................................................. 6
3.4. Artificial Neural Network with Back Propagation ............................................... 7
3.4.1.Input to Feed Forward Network ................................................................... 7
3.4.2.Back Propagation Algorithm and weight Adjustment ................................. 8
3.4.3.Training Neural Network with Input face images ....................................... 9
3.4.4.Testing the ANN .......................................................................................... 9

4.

TIME SCHEDULE .................................................................................................. 10

5.

EXPECTED OUTPUT ............................................................................................ 11

6.

REFERENCES ......................................................................................................... 12

ii

LIST OF FIGURES
Figure 1: Basic Block of Face Recognition ....................................................................... .5
Figure 2 : Architecture of Neural Network .......................................................................... 7
Figure 3: Back Propagation Multilayered ANN ...................................................................8
Figure 4: Time Schedule of Project ........................................................... .......................10

iii

LIST OF ABBREVIATION
PCA

Principal Component Analysis

ANN

Artificial Neural Network

DCT

Discrete Cosine Transform

SOM

Self Organizing Map

BPNN

Back Propagation Neural Network

iv

1.

INTRODUCTION

The human face plays an important role in our social interaction, conveying peoples
identity. The facial recognition process normally has four interrelated phases or steps.
The first step is face detection, the second is normalization, the third is feature
extraction, and the final cumulative step is face recognition. These steps depend on
each other and often use similar techniques. Face Recognition System is a computer
based digital technology and is an active area of research. The Face Recognition
System has various applications like various authentication systems, security systems
and searching of persons etc. These applications are cost effective and consume less
time. Moreover the face database can be easily designed by using any image of the
person.

1.1. Background
Traditionally, pattern recognition is broken down into three areas: segmentation,
feature extraction, and classification. Segmentation is the first step; it is finding
regions of possible signals [1]. The second step is feature extraction and in this step
we search for the most important or significant features of the regions passed by the
segmentation which can be used in the final step, classification. Classification
compares the extracted features to those of previously identified objects and
identifies the object as one of the previously identified classes.
1.1.1. PCA Algorithm
PCA is variable reduction procedure and useful when obtained data have some
redundancy. This will result into reduction of variable into smaller no. of variables
which are called Principal Component. The major advantage of PCA is using it in
Eigen faces approach which helps in reducing the size of the database for
recognition of a test image [2]. The images are stored as their feature vectors in the
database which are found out projecting each and every trained image to the set of
Eigen faces obtained. PCA is applied on Eigen faces approach to reduce the
dimensionality of a large dataset.

1.1.2. Artificial Neural Network

It is a computational system inspired by the formation, Processing Method, and
learning ability of a biological brain. The learning process of ANN is used with
back propagation algorithm. Back Propagation is a feed forward supervised
learning network. The layers of elements make independent computation of data
and pass it to another layer. The computation of processing elements is completed
on the basis of weighted addition of the inputs. The output is compared with the
target value and the mean square error is calculated which is processed back to the
hidden layer to adjust its weights. This process is having iteration for each layer to
minimize the error by repeatedly adjusting the weight of each layer. Hence, it is
called the back propagation. The iteration process carried on until the error falls
below the tolerance level.

1.2. Motivation
Computer vision has been an intriguing sector for research work. To model human
vision is a very challenging job. In the process of acquiring a good and efficient
model for machine learning, Face recognition is general challenge as faces are highly
variable and geometry and appearance not too complicated, However, already
difficult to describe with simple geometric basics or functions. PCA is defined in such
a way that the first principal component has the largest possible variance and each
succeeding component in turn has the highest variance possible under the constraint
that it is orthogonal to the preceding components. ANN is nonlinear model that is
easy to use and understand compared to statistical methods. ANN with Back
propagation (BP) learning algorithm is widely used in solving various classifications
and forecasting problems. Even though BP convergence is slow but it is guaranteed.

1.3. Problem Definition

Face detection is a concept includes many sub problems. Humans recognize faces
many times a day with apparent ease, automating this process has challenged
researcher. A system that automatically recognizes faces would be useful for several
reasons. From a security perspective, an automatic face recognition system could
enhance current access control systems by authenticating a users identity. Examples

of such access control systems are secure computer systems, bank automatic teller
machines, and automatic card readers. In fact, any organization or system that permits
access based on a person's identity would find a face recognition system useful. Other
security applications for a face recognition system would be criminal identification
and scanning airports for terrorists.

1.4. Objective
The main objectives of the project is given below

To train the neural network to recognize the face from image using Back
Propagation Algorithm.

Formulation of face recognition based on feature extractor using PCA that can
be used for authorized user verification.

1.5. Applications
This project finds its applications in the following sectors:

Face detection in biometric

Photography

Computer vision

Pedestrian detection

Person identification.

Human-computer interaction.

Security systems

2.

LITERATURE REVIEW

Presently there are several methods for face recognition. PCA with SOM is the
better technique than 2D-DCT with SOM. It has discussed the face recognition
method combined with PCA and the multi-layer network which is one of the
intelligent classifications was suggested and its performance was evaluated [2]. As a
preprocessing algorithm of input face image, this method computes the Eigen faces
through PCA and expresses the training images with it as a fundamental vector. In
this reviewed various techniques for face detection such as feature extraction.
Classification schemes and databases for face recognition [3]. It gives focus on
CCA was applied to the classical PCA features to form the coherent features for
recognition, but it is applicable to other holistic face recognition features such as
independent component analysis and discrete cosine transform features, which
might improve the recognition performance further.
In that [5] has explored a noble face recognition algorithm which integrates the
principal component analysis; back propagation neural network and discrete cosine
transform to improve the performance of face recognition algorithms.PCA is used to
reduce the dimensionality of face image and the recognition is done by the BPNN
for efficient and robust face recognition. DCT is an exact and robust face
recognition system used in compression due to its compressed representation power.
DCT reduce the amount of time required to recognize a face.
It adopted the global features for extraction using PCA [6] based Eigen faces
computation method and the detection part is completed using multi- layered feed
forward Artificial Neural Networks with back propagation process. The learning
process of neurons is used to train the input face images with 1000 iterations to
reduce the error. In that, face recognition task is completed with improved accuracy
and success rate even for noisy face images. It [7] has given brief introduction of the
PCA and the self SOM neural network which are the heart for the design and
implementation, these are the final algorithms used for the design of an efficient
high-speed face recognition system. In that 87] given brief introduction about
feature extraction Using DCT.DCT reduces the dimension of data to avoid
singularity and decreases the computational cost of PCA. In this, various feature
extraction approaches are considered and a new efficient approach is proposed.
4

3.

METHODOLOGY

Read the face

Preprocessing

images from
raining Database

Extract the

Training and

feature using

Testing using.

PCA

ANN

Figure 1: Basic Blocks for Face Recognition

3.1. Read the Face Image from Training Database

The face images for training and testing of the neural network will be used from the
various data base such as Yale face Database, PIE Database which will be subjected
to preprocessing remove noise and to normalize.

3.2. Preprocessing
In preprocessing stage, initially average filtering is applied to produce blurry effect,
and contrast of the image is enhanced through histogram equalization process. Then
image size is reduced using Bi-cubic Interpolation down sampling method in order to
make it light but efficient and best suited for PCA subspace processing and neural
network classification stage.
3.2.1.

Average Filtering

After retrieving images from the database, each image undertakes smoothing process.
The mean/average filter is applied in order to produce the blurry effect, as resizing
(down sampling) is included in later stages while maintaining the quality of face
image. The 5*5 mask is used for average filter process.
R=1/25
This equation calculates the average value of the pixels, whereas 'z' is the mask, and 'i'
are mask elements. The output 'R' is then convolved with image to produce filtering
effect. A 5*5 mask used in the implementation calculates average of 25 neighboring
pixels in image.

3.2.2.

Histogram Equalization

After average filtering stage all images undertake the process of histogram
equalization. This equalization process is done in order to have uniform distribution
of intensities and to enhance contrast of the images. Mathematically histogram
equalization can be expressed as:
Sk=T(rk)=
In this equation n is the total number of pixels in an image, nj is the number of pixels
that have gray level rk.
3.2.3. Bi-Cubic Interpolation
The process of histogram equalization is followed by image resizing process. The key
advantage of resizing through bi-cubic interpolation is that it produces smoother
surfaces than any other interpolation technique. Reducing the actual resolution of the
image Bi-cubic Interpolation takes into account 16 pixels in the rectangular grid, takes
weighted average of pixels and replaces them with a single pixel, and it is that pixel
which has actually got the flavor of all the replaced pixels. This reduction is done to
reduce redundant information and to reduce mathematical complexity in PCA
process. Here the actual resolution of an image is also reduced to ease complexity in
convergence and significant effects on time required for training the neural network.

3.3. PCA for Feature Extraction

The original images cannot be used directly as the input of the face recognition
module due to the big size 1-D vector come from each 2-D image. The extraction
feature should be able to represent the unique identity of each image as well as be in
small size. Principle Component Analysis process can generate a set of Eigen faces of
a large size of images database. Each image face in a particular database can be
presented by the linearly combination of Eigen faces of that database.PCA projects
the data along the directions where variations in the data are maximizes. The
algorithm is follows as:

Assume the M face images contained in the database as P1, P2,and

P3Pm.

Calculate the average image, , as: = Pl /M, where 1< L<M, each image
will be a column vector the same size.

Compute the covariance matrix as C = PTP where P = [O1 O2 O3.Om].

Calculate the Eigen values of the covariance matrix C and carry on only k
largest Eigen values for dimensionality reduction as k = m n=1(UKT Onn).

Eigen faces are the eigenvectors UK of the covariance Matrix C corresponding

to the largest Eigen values.

All the centered images are projected into face space on Eigen faces basis to
compute the projections of the face images as feature vectors as:
w = UTO = UT(Pi - ), where 1<i<m.

3.4. Artificial Neural Network with Back Propagation

For Face Recognition purpose, we use learning process of ANN with back
propagation algorithm. Back Propagation is a feed forward supervised learning
network.
3.4.1. Input to Feed Forward Network
Firstly, we select the parameters for required Neural Networks operation i.e. the
number of input layers, hidden layers and output layers. These input neurons get the
inputs signal from the training database of face images. Each input has its own
weights.

Figure 2: Architecture of Neural Network [1]

3.4.2. Back Propagation Algorithm and weight Adjustment

The input layer processes the data to the hidden layer which computes the data
further and passes it to the output layer. Output layer compare it with the target value
and obtain the error signals. These errors are sent back for updating the weights of
each layer to minimize the error as shown in figure 2.
Target

Input

ANN with weight

Compare

Output

between neuron

Adjust
Weights
Figure 2: Back Propagation of Multilayered ANN

Back Propagation Algorithm works as follows:

-

Set all weight to random value range from -1.0 to 1.0.

Set an input pattern (binary values) to the neurons of the nets input layer.

Active each neuron of the following layer:

Multiply the weight values of the connections, leading to this neuron with
the output values of the preceding neurons.

Add up these values.

Pass the result to an activation function, which computes the output value
of this neuron.

Repeat this until the output layer is reached.

Compare the calculated output pattern to the desired target pattern and compute
a square error value.

Change

all

weights

values

of

each

weight

using

the

formula:

Weight (old) + Learning Rate * Output Error * Output (Neuron i) * Output

(Neuron i + 1) * (1 Output (Neuron i + 1)).
-

Go to the first step.

8

The algorithm end, if all output pattern match their target pattern.

3.4.3. Training Neural Network with Input face images

After setting the parameters of neural networks we train with Eigen faces of the
input images via input layer, hidden layer and output layer. Each Eigen faces image
distance will be compared with each other. The errors at the output layer are sent
back to the previous layers that are hidden layer and update the weights of these
layers which minimize the error. The momentum and learning rate parameters
counts the updates from previous iterations and recalculates the new updated
output.
3.4.4. Testing the ANN
For face recognition, the Eigen faces images of the test face image will be intended
by feature extraction based on PCA. This Eigen faces image is given to the trained
neural network. The tested Eigen faces is compared with the Eigen faces of the
trained neural network for finest match result using the Log- sigmoid function
values.

4.

TIME SCHEDULE

Material Collection
Algorithm Design
Implement Algorithm
Run algorithm for test Images
Output Analysis
Documentation
0

10 11 12 13 14 15

Weeks

Figure 4: Time Schedule of Project

10

5.

EXPECTED OUTPUT

There will be an artificial neural network based feature extractor that can be used for
authorized user verification in a practical work environment.

11

6.

REFERENCES

[1] Ashvini E. Shivdas; Face Recognition Using Artificial Neural Network.

[2] K.Rama Linga Reddy, G.R. Babu, Lal Kishore, Face Recognition Based on
Eigen Features of Multi Scaled Face Components and an Artificial Neural
Network
[3] Jong-Min Kim, Myung-A Kang a Study of Face Recognition using the

PCA

and Error Back-Propagation 2010IEEE.

[4] Afsaneh Alavi Naini, Fatemeh seiti, Mohammad Teshnelab, Mahdi Aliyari
Shoorehdeli Face Detection Based on Dimension Reduction using Probabilistic
Neural Network and Genetic Algorithm 24-26 IEEE(2009).[3] Hua Huang
and Huiting He Super-Resolution Method for Face Recognition Using
Nonlinear Mappings on Coherent Features. IEEE Transactions On Neural
Networks, VOL. 22, NO. 1, JANUARY2011
[5] Sumanpreet Kaur1, Rajinder SinghVirk2 DCT Based Fast Face Recognition
Using PCA and ANN International Journal of Advanced Research in Computer
and Communication Engineering Vol. 2, Issue5, May2013
[6] Navneet Jindal 1, Vikas Kumar Enhanced Face Recognition Algorithm using
PCA with Artificial Neural Networks International Journal of Advanced
Research in Computer Science and Software Engineering.
[7] Rohit Jain1, Rajshree Taparia Design of Face Recognition System by Using
Neural Network with Discrete Cosine Transform and Principal Component
Analysis International Journal of Advanced Computer Research (ISSN (print):
2249-7277

ISSN

(online):

2277-7970)

Volume-2

Number-4

Issue-7

December-2012.
[8] Saeed Dabbaghchia, Masoumeh P.Ghaemmaghami b, Ali Aghagolzadeh a
Feature extraction using discrete cosine transform and discrimination power
analysis with a face recognition technology Journal of Pattern Recognition 43
(2010) 14311440.

12