Wavelet Based Face Recognition System

NATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING SCIENCE & TECHNOLOGY (NCETEST-2014)
MARCH 29th -30th, 2014

COLLEGE OF ENGINEERING ROORKEE (COER), ROORKEE, INDIA
Performance Analysis of Wavelet Subband Based

Face Recognition System
1
Mr. B. D. Patel, 2Varun Deep Singh and 3Shubham Agrawal
College of Engineering Roorkee/Electronics & Telecom. Dept., Roorkee, India.

Email: bdpatel@coer.ac.in
2,3
College of Engineering Roorkee/Electronics & Telecom. Dept., Roorkee, India.
Email:2vdeep91@gmail.com, 3agrawalatshubham94@gmail.com
Abstract A new face recognition technique using

Wavelet Subband is proposed. We extract image
features of facial images using Discrete Wavelet
transform (using Cohen Daubechies Function
window) by decomposing face image in subbands 1 to
8. Here we apply DWT on each image in ORL
database to evaluate the performance. The DWT is
computed by successive lowpass and highpass filtering
of the discrete time-domain signal. Then the features
LL3 (i.e the low-low coefficients obtained at 3rd level
of DWT) is analyzed by the classifiers J48 and Naives
classifier. For each person 10 images are used to train
the classifier. The results show that for the entire
wavelets, subband LL3 gives a good classification
accuracy.
Index Terms Classifiers J48 and Naives, Discrete
Wavelet Transform, lowpass coefficients extraction

using Cohen Daubechies Function(CDF), Classifier
performance on LL3 coefficients, Conclusion.
I. INTRODUCTION
Face recognition has been an area of importance
in recent years, because of its benefit of being a
reflexive, non-invasive system to verify
personal identity in a natural and friendly way.
Primary areas of its application are surveillance,
access control and security monitoring systems.
Face recognition is one of the most challenging
tasks in image processing because its
applications requires a high degree of accuracy.
Till date many techniques and methods have
been suggested for accurate face recognition of
an individual. A Comprehensive review of
literature has been presented by Zhao et al. [1]
on this. Methods of face recognition can be
divided into two approaches namely, feature
geometry based and subspace analysis
techniques.
In feature geometry based approach,
recognition is based on the relationship between
human facial features, for example separation

between eyes, width of jaw, distance between
nose and cheeks or face boundary etc. However
these techniques have limited reliability as they
require a correct positioned image of an
individual for recognition.
In Subspace analysis approach we analyse the
intensity variation in given image of face. This
can be achieved by techniques like variance,
applying transforms to extract coefficients etc.
By applying these techniques dimensionality of
a given image is greatly reduced because the
redundant data is removed. Sometimes we
require only the key features to recognize a
given image. This makes subspace methods
more popular for application nowadays.
General face recognition methods consists of
a feature extractor (like PCA, Wavelet
decomposer) to reduce the size of input and a
classifier like Naive, SVM (Support Vector
Machines), Neural Networks, Nearest Distance
Classifiers to find the features which are distinct
for each individual.. Low computation time by
reduction in data is an important advantage.
In this study, we chose wavelet
decomposition as main technique for data
reduction and feature extraction, and analysis of
accuracy of recognition using classifiers
(Naives Bayes and J48). We extract image
features of facial images by use of wavelet
transform (using Cohen Daubechies Function)
by decomposing face image in subbands 1 to 8.
These features are analyzed by using Naives
and Bayes classifier.
The remainder of the paper is organized as
follows: next section introduces to discrete
wavelet transform and the classifiers J48 and
Naive. In Section III we discuss the proposed
method. Results obtained are presented in
NCETEST-2014

section IV. Finally, conclusions are given in

section V.
II. DISCRETE WAVELET TRANSFORM &
CLASSIFIERS
A. Discrete Wavelet Transform
A discrete wavelet transform (DWT) is any
wavelet transform for which the wavelets are
discretely sampled.Discrete Wavelet Transform
is a dominant tool used in image processing and
computer vision, due to its ability to capture
localized time-frequency information of an
image[2]. The application of discrete wavelet
transform decomposes data into different
frequency ranges. Thus frequency components
of image are decomposed into subbands . By
application of DWT on these subbands a more
resolution in frequency domain can be obtained.
1D-Discrete Wavelet Transform is given by:
, =
1
2
(1)
2
Where,
j is the power of binary scaling, k is a constant
of the filter and (t) is called mother wavelet
(here CDF)
B. Classifiers
A classifier can be viewed as a function of
block, which assigns one class to each point of
the input space. The input space is thus
partitioned into different class. There are two
classification algorithms used for classification
of coefficients from image.
Decision tree: J48
A decision tree is a predictive machine-learning
model that decides the target value (dependent
variable) of a new sample based on various
attribute values of the available data. The
internal nodes of a decision tree denote the
different attributes; the branches between the
nodes tell us the possible values that these
attributes can have in the observed samples,
while the terminal nodes tell us the final value
(classification) of the dependent variable. [3]
Naive Bayes classifier
The probabilistic classifier based on applying
Bayes' theorem is Naive Bayes. Depending on
the precise nature of the probability model,
Naive Bayes classifiers can be trained very
efficiently in a supervised learning setting. It
uses the method of maximum likelihood for
parameter estimation.It is probabilistic classifier
with the assumption that the attribute values are
conditionally independent of the given target
values. . The model is
1, 2,, =
Figure-1: Multiresolution scheme after 2 and 3 levels of wavelet

transform on an image.
The DWT uses multiresolution filter banks and

special wavelet filters for the analysis and
reconstruction of signals. The 2-D DWT is
computed by successive lowpass and highpass
filtering of the image. By applying 2D DWT on
an image, the image is decomposed into four
subbands LL, LH, HL, HH subbands,
corresponding to approximate , horizontal,
vertical, and diagonal features respectively.
1
()
(2)
Where is a class variable, 1, 2,, are

attribute, Z is scaling factor Bayesian approach is
to assign most probable target value is
given by:
= max ( )

(3)
III.THE PROPOSED METHOD

For a given face image, feature vectors of faces
are extracted through 2-D Discrete Wavelet
Transform (CDF) at LL3 subband[4]. By
application of DWT we obtain dimensional
reduction of data of a given image. The LL3
subband coefficients are then given to
NCETEST-2014

classifiers J48 and Naives Bayes to examine the

classification accuracy in each classifier. The
images after application of CDF transform are
shown in the figure 2 (White region is LL3).
IV. RESULTS
Here ORL face database [5] from AT&T
(Olivetti) Research Laboratories, Cambridge
has been used. The database contains 40
individuals with each person having ten frontal
images. We have taken images of only 10
individuals for our experiment. Few sample
face images from this database are shown in
figure.
Figure 2. Wavelet decomposition at level 3.
Figure 4. Few face images from ORL face database.
The results obtained in WEKA software are

tabulated as follows:
S.No
1.
2.
3.
4.
5.
6.
7.
Figure 3: Flowchart of proposed method
In classification 70% data is used for training,

remaining 30% in used in testing. For the
application of the wavelet transform on an
image software used is MATLAB, and for
checking classification accuracy (using J48 and
Naive Bayes) we used WEKA software.
8.
9.
Classification
Accuracy
Percentage
data used for
training
Total Number
of Instances
Correctly
Classified
Instances
Incorrect
Classified
Instances
Mean Absolute
Error
Naives
Decision
and Bayes Tree:J48
Classifier Classifier
93.33
83.3333
70%
70%
30
30
28
25
0.013
0.03333
Root
Mean
Square Error
Relative
Absolute Error
0.1155
0.1826
7.3462
18.3655
Root Relative
Squared Error
38.1147
60.2646
NCETEST-2014

V. CONCLUSIONS
We applied wavelet transform on images in
database and compared the performance of two
classifiers (J48 and Naives) using LL3
coefficients .Based on results we observed that
the two classifiers give above 80%
classification accuracy. However we observe
that classification accuracy in Naives Bayes
classifier is more (93.33%), making it more
suitable for classification.
We also observe that at Subband LL3 we
extract good frequency information from an
image. The dimensionality of image is reduced
after application of discrete wavelet transform.
This has the advantage of reducing the storage
size in recognition system; also the recognition
time will also be reduced.
REFERENCES
[1] W. Zhao, R. Chellappa, A. Rosenfeld, and
P.J. Phillips,Face recognition: A literature
survey, CVL Technical Report, University of
Maryland, 2000.
[2]Wikipedia ,the free encyclopedia
http://en.wikipedia.org/wiki/Discrete_wavelet_t
ransform
[3] Chapter 5,CLASSIFICATION METHODS
- University of Minnesota Duluth
http://www.d.umn.edu/~padhy005/Chapter5.ht
ml
[4]Face Recognition based on TwoDimensional PCA on Wavelet Subband by
Kishor S Kinage and S. G. Bhirud
[5] ORL face databases,
http://www.uk.research.att.com/pub/data/orl_fa
ces.zip
NCETEST-2014

Wavelet Based Face Recognition System

Uploaded by

Copyright:

Available Formats

Wavelet Based Face Recognition System

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wavelet Based Face Recognition System

Uploaded by

Copyright:

Available Formats

NATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING SCIENCE & TECHNOLOGY (NCETEST-2014)

MARCH 29th -30th, 2014

Performance Analysis of Wavelet Subband Based

Mr. B. D. Patel, 2Varun Deep Singh and 3Shubham Agrawal

College of Engineering Roorkee/Electronics & Telecom. Dept., Roorkee, India.

Abstract A new face recognition technique using

Wavelet Transform, lowpass coefficients extraction

human facial features, for example separation

NATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING SCIENCE & TECHNOLOGY (NCETEST-2014)

section IV. Finally, conclusions are given in

Figure-1: Multiresolution scheme after 2 and 3 levels of wavelet

The DWT uses multiresolution filter banks and

Where is a class variable, 1, 2,, are

III.THE PROPOSED METHOD

NATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING SCIENCE & TECHNOLOGY (NCETEST-2014)

classifiers J48 and Naives Bayes to examine the

Figure 2. Wavelet decomposition at level 3.

Figure 4. Few face images from ORL face database.

The results obtained in WEKA software are

In classification 70% data is used for training,

NATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING SCIENCE & TECHNOLOGY (NCETEST-2014)

You might also like