Comparative Analysis of Conventional, Real and Complex Wavelet Transforms For Face Recognition

International Journal of Computer Applications (0975 8887)
Volume 39 No.6, February 2012

6
Comparative Analysis of Conventional, Real and
Complex Wavelet Transforms for Face Recognition

Ankit A. Bhurane
Dept. of Electronics and
Telecom Engineering,
SGGS IE &T, Nanded, India.

Sanjay N. Talbar

Preeti N. Gophane

ABSTRACT
Face recognition has been studied for many years in the
context of biometrics and is one of the most successful
applications of image analysis and understanding. Various
methods, approaches and algorithms for recognition of human
faces were proposed. In this paper, independent, comparative
study of conventional discrete wavelet transform (DWT), real
dual-tree discrete wavelet transform (R-DT-DWT), and
complex dual-tree discrete wavelet transform (C-DT-DWT)
based features for face recognition is carried out. In 2005,
Delac et al. [26] presented an independent comparative study
of PCA, ICA, and LDA on the FERET data set where it was
concluded that no particular distancemetric combination is
the best. In this paper we intend to bring further conclusions.
Unlike the contribution by Delac et al., our conclusions are in
context of DWT, R-DT-DWT, and C-DT-DWT. Moreover,
these approaches are tested on nine different databases at
different levels and under three different distance metrics,
which allowed us to compare their performance
independently. Our simulation results show that no particular
distancemetric combination is the best across all standard
benchmark face databases. However, the overall performance
for city block distance measure was found to be better as
compared to the Euclidean and cosine distance. Also, the
performance for R-DT-DWT and C-DT-DWT based features
were found equivalently efficient in many cases. So taking
redundancy into consideration, it may be suggested to opt for
R-DT-DWT for face efficient recognition.
Keywords
Real Dual-tree Discrete Wavelet Transform (R-DT-DWT),
Complex dual-tree discrete wavelet transform (C-DT-DWT),
Face Recognition Technology (FERET).

1. INTRODUCTION
Face recognition has always been a hot research topic for last
several years. It is greatly motivated by the needs of military,
public security, scene surveillance including commercial and
law enforcement applications [1]. Turk and Pentland used the
principal component analysis (PCA) technique to develop the
first successful and well known Eigenface scheme for face
recognition which is widely used due to its simplicity and
effectiveness. Later various methods for the face recognition
were proposed. Over the last decade wavelets have become
powerful and flexible tools for computation and data
reduction. For good direction selectivity, real dual-tree
discrete wavelet transform (R-DT-DWT), and complex dual-
tree discrete wavelet transform (C-DT-DWT) were introduced
in [2], [3].
The aim of this paper is to provide an independent,
comparative study of conventional discrete wavelet transform
(DWT), real dual-tree discrete wavelet transform (R-DT-
DWT), and complex dual-tree discrete wavelet transform (C-
DT-DWT) with accompanied distance metrics and in
completely equal working conditions. In order to perform a
fair comparison, same input images are the input into all three
algorithms. In this paper, all the three algorithms are tested on
three different distance metrics. Distance metrics used are L1
(City block), L2 (Euclidean), and cosine. This paper is
organized as follows: Section II gives short introduction to
DWT, R-DT-DWT, and C-DT-DWT. In Section III
experimental results are presented. Finally, conclusions are
drawn in Section IV.
2. PRELIMINARIES
A simple face recognition system is as shown in the Figure 1.
It includes feature extraction, similarity measurement and
decision making. Given a stored database of facial images one
has to train an automated system to identify or verify a person.

Fig 1: Block Diagram of face recognition system [27].
2.1 Conventional two dimensional wavelet
discrete transform (2D-DWT)
A conventional two dimensional wavelet discrete transform
(2D-DWT) can be considered as equivalent to filtering the
input image with a bank of filters, whose impulse responses
are all approximately given by scaled versions of a mother
wavelet. The separable (row-column) implementation of the
2D-DWT is characterized by three wavelets,
1
( , ) ( ) ( ) (LH wavelet) x y x y | =

(1)

Test Image
Training Set
Input Probe
Face Image
Feature
Extraction
Face Feature
Database
Similarity
Measure
(Comparison)
Decision
Making
7
2
( , ) ( ) ( ) (HL wavelet) x y x y | =

(2)

3
( , ) ( ) ( ) (HH wavelet) x y x y =

(3)

The LH wavelet is the product of the low-pass function ( ) |
along the first dimension and the high-pass (actually a
bandpass) function ( ) along the second dimension.
Similarly the HL and HH wavelets can be treated. In this
paper, the filters given by Abdelnour et al. [4] were used.
2.1.1 Problems with Conventional Wavelets
The problems with conventional wavelets are as follows:
- Shift Invariance: Real DWT are very sensitive to shifts.
Small shifts in the input signal can cause major
variations in the distribution of energy between DWT
coefficients.
- Poor Directional Selectivity: DWT coefficients reveal
only three spatial orientations (horizontal, vertical and
diagonal).
- Lack of Phase: DWT analysis of real signals lacks the
phase information that accurately describes non-
stationary signal behavior.
2.1.2 Comparison with Gabor approach
Apart from many techniques proposed in the literature, Gabor
wavelet-based representation of face images is an excellent
solution covering most of the above properties. However,
Gabor wavelet-based representation is computationally very
complex and the memory requirement for storing Gabor
features is very high. These problems were tried to overcome
using sub-Gabor [15], simplified Gabor wavelets [16],
optimal sampling of Gabor features [17], etc. But none of
these approaches succeeded effectively. An approach to both
shift invariance and directional selectivity was given by
Simoncelli et al. [18]. However, the dual-tree complex
wavelet transforms approach proposed by Kingsburry [19],
[20], gives perfect reconstruction and greater directional
selectivity. Thus, it provides an excellent alternative to Gabor
wavelets with the potential to overcome the above-mentioned
shortcomings of the Gabor wavelets.
Sankaran et al. [21] and Celik et al. [22] used the Complex
Dual-Tree Discrete Wavelet Transform (C-DT-DWT) and
Gabor wavelets for facial feature extraction. Complex
Wavelet Transform based face recognition was also discussed
in [23]. In [24] and [25], orthogonal neighborhood preserving
projections (ONPPs) and supervised kernel ONPP with C-DT-
DWT was used for face recognition.
2.2 2-D Dual-Tree Discrete Wavelet
Transform (DT-DWT)
The 2-D dual-tree discrete wavelet transform (DT-DWT) of
an image is implemented using two critically-sampled
separable 2-D DWTs in parallel as shown in Figure 2. One of
the advantages of the dual-tree DWT (DT-DWT) over
separable 2D DWT is that, it can be used to implement 2D
wavelet transforms that are more selective with respect to
orientation. The filters used in the first stage of the dual-tree
DWT are different from the filters used in the remaining
stages [3].
Here, we have used Farras filters for the first stage and
Kingsburys Q-shift filters for remaining stages.

Fig 2: 2D Dual-tree discrete wavelet transform.
2.2.1 2-D Real Dual-Tree Discrete Wavelet
Transform (R-DT-DWT)
In real dual-tree discrete wavelet transform (R-DT-DWT), the
sum and difference for each pair of subbands gives rise to
wavelets in six different directions. At each level, the
wavelets are strongly oriented at angle of 15, 45, and
75 as shown in Figure 3.

Fig 3: Six Wavelets associated with the real 2D dual-tree
DWT.
2.2.2 2-D Complex Dual-Tree Discrete Wavelet
Transform (C-DT-DWT)
The 2-D complex dual-tree discrete wavelet transform (C-DT-
DWT) has twice as many wavelets as that of R-DT-DWT
(two wavelets in each direction). The wavelets are oriented in
the same six directions as those of the 2-D R-DT-DWT. In
each direction, one of the two wavelets can be interpreted as
the real part of a complex-valued 2D wavelet, while the other
wavelet can be interpreted as the imaginary part of a complex-
valued 2D wavelet as shown in the Figure 4. The extra six
wavelets are obtained by swapping the low pass and high pass
filter coefficients of real and imaginary trees.
An illustration of R-DT-DWT operated on a synthetically
generated image is as shown in the Figure 5.

Fig 4: Impulse responses for 2-D C-DT-DWT: First row is
interpreted as the real part and the second row as imaginary
Tree 1

Tree 2
Level 1 Level 2
Level J
Face
Image

h0(n)
h1(n)
h01
g01
h01
g01
LL

LH
HL
HH

LL

LH
HL
HH

LL

LH
HL
HH

g0(n)
g1(n)

h11
g11
h11
g11
LL

LH
HL
HH

LL

LH
HL
HH

LL

LH
HL
HH

8
part of the complex wavelet. The third row shows the
magnitude response.

(a) A synthetically generated image

(b) Thresholded R-DT-DWT subbands at third level

(c) Thresholded C-DT-DWT subbands at level = 3.
Fig 5 : R-DT-DWT and C-DT-DWT subbands of a
synthetically generated image.
3. EXPERIMENTAL RESULTS
3.1 Simulation Software
The software for face recognition was built using MATLAB.
This environment was chosen because it easily supports image
processing, image visualization, and linear algebra. The
software was tested against wide range of databases
3.2 Benchmark Face Databases and
Settings
In order to evaluate the proposed face recognition system, our
experiments were performed on following nine benchmark
face databases:

3.2.1. Face Recognition Technology (FERET)
Database
The FERET database [6] contains 1564 sets of images for a
total of 14,126 images that includes 1199 individuals and 365
duplicate sets of images. In our experiment 50 subjects were
chosen randomly, each having 11 frontal face images varying
in illumination, expression, and pose.
3.2.2. Olivetti Research Laboratory (ORL) face
database
The ORL face database [7] contains ten different images of
each of 40 distinct subjects. For some subjects, the images
were taken at different times, varying the lighting, facial
expressions (open/ closed eyes, smiling/ not smiling) and
facial details (glasses/ no glasses). All the images were taken
against a dark homogeneous background with the subjects in
an upright, frontal position (with tolerance for some side
movement). There are 10 images per subject.
3.2.3. Yale face database
The Yale face database [8] contains 165 images of 15
subjects. There are 11 images per subject, one for each of the
following facial expressions or configurations: center-light,
with glasses, happy, left-light, without glasses, normal, right-
light, sad, sleepy, surprised, and wink.
3.2.4. Japanese Female Facial Expression
(JAFFE) database
The JAFFE database [9] contains 213 images of 7 facial
expressions (6 basic facial expressions + 1 neutral) posed by
10 Japanese female models. Each image has been rated on 6
emotion adjectives by 60 Japanese subjects. In our
experiment, 20 face image of each subject were chosen.
3.2.5. University of Bern face database
The University of Bern face database [10] contains frontal
views of 30 people. For each person 10 gray level images with
slight variations of the head positions (1, 2 right into the
camera, 3,4 looking to the right, 5,6 looking to the left, 7,8
downwards, 9,10 upwards.
3.2.6. Indian Institute of Technology Kanpur
(IITK) face database
The IITK database [11] consists of 11 different images of
each of 40 distinct subjects. For some subjects, some
additional photographs are included. All the images have a
bright homogeneous background with the subjects in an
upright, frontal position. The following orientations of the
face are included: looking front, looking left, looking right,
looking up, looking up towards left, looking up towards right,
looking down. Available emotions are: neutral, smile,
laughter, sad/disgust.
In our experiment, 20 male and 20 female data sets were
chosen and were treated independently.
3.2.7. Computer Vision Science Research (CVSR) -
Grimace face database
The CVSR database [12] contains a sequence of 20 images of
18 different individual, using a fixed camera. During the
sequence the subject moves his/her head and makes grimaces
which get more extreme towards the end of the sequence.
9

Fig 6: Database samples: (a) FERET, (b) ORL, (c) Yale,
(d) JAFFE, (e) University of Bern, (f) IITK [Male], (g)
IITK [Female], (h) CVSR [Grimace], (i) Georgia Tech., (j)
S.G.G.S.

3.2.8. Georgia Tech face database
The Georgia Tech. database [13] contains 15 different images
of each of 50 distinct people taken in two or three sessions at
the Center for Signal and Image Processing at Georgia
Institute of Technology. The pictures show frontal and/or
tilted faces with different facial expressions, lighting
conditions and scale. In our experiments we had used cropped
database.
The benchmark images were treated on "as is" basis. All the
images were initially resized to dimension 6464 for our
experiments. No other preprocessing was done. For each
subject of each database, about half of the images were used
for training purpose and remaining others for testing.
3.2.9. S.G.G.S. Face Database
The SGGS database [5] contains 25 different images of each
of 30 distinct people taken in one session at Shri Guru Gobind
Singhji (S.G.G.S.) Institute of Engineering and Technology,
Nanded, India. For some subjects, some additional
photographs are included. All the images were taken in plain
background, varying the lighting, pose, facial expressions and
facial details (glasses/ no glasses) in robust environment.
Figure 6 shows samples from each of the above databases.
3.3 Feature Extraction
At first level, a face image of size N N is decomposed
into four subband image: LL, LH, HL, and HH. Higher level
decompositions are on LL subband image at previous level.
The feature extraction is done as follows: Input image is
initially resized to an N N image. Further, the DWT, R-
DT-DWT, C-DT-DWT are applied independently up to level
J . At
th
J level, the subband images will have size
2 2
J J
N N | |
|
\ .
. Then, the feature vector can be formed by
concatenating the rows or columns of all the subbands at J
th
level. For example, the feature vector size for conventional
DWT, R-DT-DWT, and C-DT-DWT at 4 J = will be
4
2 2
J J
N N | |

|
\ .
, 2 4
2 2
J J
N N ( | |

| (
\ .
and
2 2 4
2 2
J J
N N ( | |

| (
\ .
respectively. For our
experiments, the input images are initially scaled to 64 64
. Independent analysis of DWT, R-DT-DWT and C-DT-DWT
was done at third and fourth level.
The R-DT-DWT and C-DT-DWT subbands of a face image
are shown in Figure 7.
3.4 Similarity Measures
The similarity measures used in our experiments to evaluate
the efficiency of different representation and recognition
methods include Euclidean distance, city block distance, and
cosine distance. If,
X: mx-by-n data matrix treated as mx (1 ) n row vectors x
1
,
x
2
, ..., x
mx
, and
Y: my-by-n data matrix treated as my (1 ) n row vectors y
1
,
y
2
, ..., y
my
,
then, the similarity measures between the vectors x
s
and y
t
are
defined as follows:
A. City Block distance (L1 distance):
1
n
st sj tj
j
d x y
=
=
(4)

(a) Input face image

(b) R-DT-DWT subbands of input image
10

(c) C-DT-DWT subbands of input image
Fig 7: R-DT-DWT and C-DT-DWT subbands of a face
image.

B. Euclidean distance (L2 distance):
2
( )( )'
st s t s t
d x y x y = (5)
C. Cosine distance:
'
' '
1
( )( )
s t
st
s s s s
x y
d
x x y y
| |
| =
|
\ .
(6)
3.5 Results
For each subject of each database, about half of the images
were used for training purpose and remaining others for
testing. The results are tabulated in Table 1. The efficiency of
recognition can be calculated as follows,

100%

correctly recognized face images
Efficiency
total number of input face images
=

The results show that no particular distancemetric
combination is the best across all standard benchmark face
databases. However, overall performance for city block
distance measure is found to be better as compared to the
Euclidean and Cosine distance in many cases. Also, the
performance of R-DT-DWT and C-DT-DWT based features
are found to be about equally efficient in many cases.
4. CONCLUSIONS
This paper presented an independent, comparative study of
DWT, R-DT-DWT and C-DT-DWT based face recognition
methods with their accompanied three distance metrics (L1,
L2 and cosine), at different levels, and in completely equal
working conditions.
Our comparative research shows that no particular distance
metric combination is the best across all standard benchmark
face databases. This verifies the contribution made by Delac
et al. [26]. In addition, however, overall performance for city
block distance measure was found to be better as compared to
the Euclidean and cosine distance in many cases. Also, the
performance of R-DT-DWT and C-DT-DWT based features
were found to be about equally efficient in many cases.
Although C-DT-DWT has the benefit of being both oriented
and approximately analytic, the overall performance of R-DT-
DWT and C-DT-DWT based features are found to be about
equally efficient in many cases for face recognition. So taking
redundancy into consideration, it may be suggested to opt for
R-DT-DWT for efficient face recognition.
4. ACKNOWLEDGMENTS
The authors would like to thank the teaching, non-teaching
staff members and M.Tech students from Electronics
department of Shri Guru Gobind Singhji Institute of
Engineering and Technology, Nanded (Maharashtra, India)
for their cooperation in creating the SGGSIET face database.
The authors would like to thank those researchers who have
kindly provided the public face databases such as FERET,
ORL, Yale, JAFFE, IITK, CVSR, University of Bern, and
Georgia Tech. Authors would also like to thank Dr.
Kingsbury (Cambridge University) and Prof. Selesnick
(Polytechnic University, NY, USA) for answering the queries.
11
Table 1: Face Recognition Results.

5. REFERENCES
[1] W. Zhao, R. Chellappa, A. Rosenfeld, P. J. Phillips, Face
Recognition: A Literature Survey, ACM Computing
Surveys, 35(4): 399-458, 2003.
[2] N. Kingsbury. The dual-tree complex wavelet transform:
A new technique for shift invariance and directional
filters. IEEE Digital Signal Processing Workshop, DSP
98, paper no. 86, August 1998.
[3] I. Selesnick, R. Baraniuk, and N. Kingsbury. The dual-
tree complex wavelet transform. IEEE Signal Process.
Mag., 22(6):123151, Nov. 2005.
[4] A. F. Abdelnour and I. W. Selesnick. Nearly symmetric
orthogonal wavelet bases. In Proceedings of IEEE
International Conference on Acoustic, Speech, Signal
Processing (ICASSP), May 2001.
[5] A. A. Bhurane. Face Recognition using Dual-Tree
Discrete Wavelet Transforms. M.Tech Thesis, S.G.G.S.
Institute of Engineering and Technology, Nanded,
Maharashtra, India, July 2011.
[6] The color FERET database, USA. Website.
http://face.nist.gov/colorferet/.
[7] AT & T: The database of faces (formerly: The ORL
database of faces). Website.
http://www.cl.cam.ac.uk/Research/DTG/attarchive/faced
atabase.html.
[8] The Yale face database. Website.
http://cvc.yale.edu/projects/yalefaces/yalefaces.html.
[9] The Japanese Female Facial Expression (JAFFE)
database. Website. http://www.kasrl.org/jaffe.html.
[10] The University of Bern face database. Website.
http://www.ph.tn.tudelft.nl/PRInfo/data/msg00010.html.
[11] The IITK face database. Website. http://vis-
www.cs.umass.edu/~vidit/IndianFaceDatabase/.
[12] The CVSR Grimace database. Website.
http://cswww.essex.ac.uk/mv/allfaces/index.html.
[13] The Georgia Tech. face database. Website.
http://www.anefian.com/research/GTdb_crop.zip.

Distance
Measures

Database

Efficiency (%)
J = 3 J = 4
DWT
R-DT-
DWT
C-DT-
DWT
DWT R-DT-DWT
C-DT-
DWT
Euclidean

FERET 74.33 73.33 73.33 73.33 73.66 73.66
ORL 93.50 93.50 93.50 92.50 94.00 94.00
Yale 87.77 88.88 88.88 85.55 85.55 85.55
JAFFE

92.00 91.00 91.00 90.00 91.00 91.00
University of Bern 92.00 91.33 91.33 90.66 92.00 92.00
IITK
IITK (Male) 74.16 70.00 70.83 72.50 73.33 73.33
IITK (Female) 86.66 86.66 85.83 87.50 87.50 88.33
CVSR (Grimace) 96.66 96.66 96.66 96.66 96.66 96.66
Georgia Tech. 71.00 71.50 71.50 70.50 69.75 70.00
SGGS 67.43 66.92 66.92 62.05 62.56 63.07
POSTECH 72.88 73.33 73.11 63.77 64.00 64.00
Average Efficiency 82.58 82.10 82.08 80.45 80.91 81.05
Cityblock
FERET 77.66 79.33 78.33 74.66 75.33 75.33
ORL 95.00 96.00 96.50 95.00 95.55 95.50
Yale 94.44 95.55 95.55 93.33 93.33 94.44
JAFFE

95.00 97.00 97.00 97.00 96.00 96.00
IITK
IITK (Male) 75.00 77.50 77.50 74.16 76.66 76.00
IITK (Female) 86.66 88.33 88.33 89.16 90.00 89.16
CVSR (Grimace) 96.66 96.66 96.66 96.66 96.66 96.66
Georgia Tech. 72.75 72.75 73.00 72.00 72.25 72.75
SGGS 77.94 77.69 77.17 71.53 74.35 74.61
POSTECH 82.44 82.66 83.33 75.11 72.88 73.77
Cosine
FERET 84.00 84.33 84.33 83.00 82.00 82.33
ORL 98.83 92.00 92.00 93.00 83.50 94.00
Yale 91.11 92.22 92.22 87.77 90.00 90.00
JAFFE

96.00 96.00 96.00 98.00 98.00 97.00
IITK
IITK (Male) 69.16 67.50 67.50 70.83 70.00 69.16
IITK (Female) 87.50 87.50 87.50 89.16 89.16 89.16
CVSR (Grimace) 96.66 96.66 96.66 96.66 96.66 96.66
Georgia Tech. 65.75 65.75 65.75 67.75 68.50 68.75
SGGS 72.56 71.53 71.53 69.74 69.48 69.23
POSTECH 72.44 72.66 72.66 62.22 63.33 63.55
12
[14] The SGGS face database. Website. http://sggs.ac.in.
[15] L. Nanni and D. Maio, Weighted sub-Gabor for Face
Recognition, Pattern Recognition Letters, vol. 28, no. 4,
pp. 487492, 2007.
[16] W.-P. Choi, S.-H. Tse, K.-W. Wong, and K.-M. Lam,
Simplified Gabor Wavelets for Human Face
Recognition, Pattern Recognition Letters, vol. 41, no. 3,
pp. 11861199, 2008.
[17] D.-H. Liu, K.-M. Lam, and L.-S. Shen, Optimal
Sampling of Gabor Features for Face Recognition,
Pattern Recognition Letters, vol. 25, no. 2, pp. 267276,
2004.
[18] E P Simoncelli, W T Freeman, E H Adelson and D J
Heeger, Shiftable Multiscale Transforms, IEEE
Transactions on Information Theory, 38(2), pp 587-607,
March 1992.
[19] N G Kingsbury, The Dual-Tree Complex Wavelet
Transform: A New Technique for Shift Invariance and
Directional Filters, Proc. 8th IEEE DSP Workshop,
Bryce Canyon, Aug 1998.
[20] N G Kingsbury: The Dual-Tree Complex Wavelet
Transform: A New Efficient Tool for Image Restoration
and Enhancement, Proc. EUSIPCO 98, Rhodes, Sept
1998.
[21] H. E. Sankaran, A. P. Gotchev, K. O. Egiazarian, and J.
T. Astola, Complex Wavelets versus Gabor Wavelets
for Facial Feature Extraction: A Comparative Study, in
Image Processing: Algorithms and Systems IV, vol. 5672
of Proceedings of SPIE, pp. 407415, San Jose, Calif,
USA, January 2005.
[22] T. Celik, H. Ozkaramanli, and H. Demirel, Facial
Feature Extraction using Complex Dual-Tree Wavelet
Transform, Computer Vision and Image Understanding,
vol. 111, no. 2, pp. 229246, 2008.
[23] A. Eleyan, H. Ozkaramanli, and H. Demirel, Complex
Wavelet Transform-Based Face Recognition, EURASIP
Journal on Advances in Signal Processing, 2008, Article
ID 185281.
[24] Y. Sun and M. Du, DT-CWT Feature Based
Classification using Orthogonal Neighborhood
Preserving Projections for Face Recognition, in
Proceedings of the International Conference on
Computational Intelligence and Security (CIS 06), pp.
719 724, Guangzhou, China, November 2006.
[25] Y. Sun, DT-CWT Feature based Face Recognition
using Supervised Kernel ONPP, in Proceedings of the
International Conference on Computational Intelligence
and Security Workshops (CISW 07), pp. 312315,
Harbin, China, December 2007.
[26] Delac K., Grgic M., and Grgic S. Independent
comparative study of PCA, ICA, and LDA on the
FERET data set. Wiley Periodicals, 15:252260, 2005.
[27] Bhurane A.A. and Talbar S.N. Vision-based
authenticated robotic control using face and hand gesture
recognition, International Conference on on Electronics
Computer Technology (ICECT 2011), vol V-1, pp 6468.
ICECT, IEEE, April 2011.

Comparative Analysis of Conventional, Real and Complex Wavelet Transforms For Face Recognition

Uploaded by

Copyright:

Available Formats

Comparative Analysis of Conventional, Real and Complex Wavelet Transforms For Face Recognition

Uploaded by

Document Information

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

Comparative Analysis of Conventional, Real and Complex Wavelet Transforms For Face Recognition

Uploaded by

Copyright:

Available Formats

International Journal of Computer Applications (0975 8887)

Volume 39 No.6, February 2012

You might also like