R044120124

P. Vijayalakshmi et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 4( Version 1), April 2014, pp.120-124
www.ijera.com 120 | P a g e
An Analysis of 2D Bi-Orthogonal Wavelet Transform Based On
Fixed Point Approximation
P. Vijayalakshmi*, M. Vidya**
*(Department of Electronics and Communication Engineering, Sri Venkateswara College of Engineering,
Chennai)
** (Department of Electronics and Communication Engineering, Sri Venkateswara College of Engineering,
Chennai)
ABSTRACT
As the world advances with technology and research, images are being widely used in many fields such as
biometrics, remote sensing, reconstruction etc. This tremendous growth in image processing applications,
demands majorly for low power consumption, low cost and small chip area. In this paper we analyzed 2D bi-
orthogonal wavelet transform based on Fixed point approximation. Filter coefficients of the bi-orthogonal
wavelet filters are quantized before implementation. The efficiency of the results is measured for some standard
gray scale images by comparing the original input images and the reconstructed images. SNR and PSNR value
shows that this implementation is performed effectively without any loss in image quality.
Key Words— Bi-orthogonal wavelet Transform, Fixed Point Approximation
I. INTRODUCTION
Recent advances in medical imaging and
telecommunication systems require high speed,
resolution and real-time memory optimization with
maximum hardware utilization. The 2D Discrete
Wavelet Transform (DWT) is widely used method
for many medical imaging systems because of perfect
reconstruction property. DWT can decompose the
signals into different sub bands with both time and
frequency information and to arrive high
compression ratio. DWT architecture, in general,
reduces the memory requirements and increases the
speed of communication by breaking up the image
into the blocks.
A wavelet is a mathematical function used
to divide a given function or continuous-time signal
into different scale components. Usually one can
assign a frequency range to each scale component.
Each scale component can be studied with a
resolution that matches its scale. A wavelet transform
is the representation of a function by wavelets. The
wavelets are scaled and translated copies known as
daughter wavelets of a finite-length or fast-decaying
oscillating waveform known as the mother wavelet.
Wavelet transforms have advantages over traditional
Fourier transforms for representing functions that
have discontinuities and sharp peaks, and for
accurately deconstructing and reconstructing finite,
non-periodic and/or non-stationary signals.
Both Discrete Wavelet Transform (DWT)
and Continuous Wavelet Transform (CWT) are
continuous-time transforms. They can be used to
represent continuous-time signals. CWT‟s operate
over every possible scale and translation whereas
DWT‟s use a specific subset of scale and translation
values or representation grid. There are a large
number of wavelet transforms each suitable for
different applications. The discrete wavelet transform
(DWT) has generated a great deal of interest recently
due to its many applications across several
disciplines, including signal processing. Thus
wavelets provide a time-scale representation of
signals as an alternative to traditional time-frequency
representations.
RELATED WORK
Many architectures have been proposed to
perform DWT, but few address the precision of the
coefficients necessary to ensure perfect
reconstruction. The goal of work is to determine the
precision of the filter coefficients (for an orthogonal
wavelet) needed to compute the 2-D DWT without
introducing round-off error via the filter. It needs at
least 14 bits for 1 octave of decomposition if both
DWT and IDWT are fixed-point operation, while 13
bits are enough to come up with the same result if
only the transform DWT is fixed-point [2]
. The
implementation of 2D DWT using fixed point
representation are shown and the results shown that
this can be performed without any loss [3]
. “VLSI
Architectures for the 4-Tap and 6-Tap 2-D
Daubechies Wavelet Filters Using Algebraic
Integers,” presented an algebraic integer (AI) based
multi-encoding of Daubechies-4 and -6 2-D wavelet
filters having error-free integer-based computation. It
also guarantees a noise-free computation throughput
RESEARCH ARTICLE OPEN ACCESS

the multi-level multi-rate 2-D filtering operation.
Comparisons are provided between Daubechies-4 and
-6 designs in terms of SNR, PSNR, hardware
structure, and power consumptions, for different
word lengths [1]
.
II. PROPOSED WORK
Fig 1 shows the block diagram of the
proposed system. Initially the image is given to the
bi-orthogonal wavelet filter, where the image is
decomposed based on sub band coding. The
coefficients which we got from the bi-orthogonal
wavelet filter are quantized using fixed point
approximation. Finally the image is reconstructed.
Fig.1 Block diagram of proposed system
SUB-BAND CODING
Generally image is decomposed based on
sub-band coding. This sub-band coding is the general
form of bi-orthogonal wavelet transform. If the
scaling and wavelet functions are separable, the
summation can be decomposed into two stages. First
step is along the x-axis and then along the y-axis. For
each axis, we can apply wavelet transform to increase
the speed. The two dimensional signal (usually
image) is divided into four bands: LL (left-top), HL
(right-top), LH (left bottom) and HH (right-bottom).
The HL band indicated the variation along the x-axis
while the LH band shows the y-axis variation. The
power is more compact in the LL band. Fig.2 shows
the single level decomposition and Fig.3 shows the
second level decomposition of image along rows and
columns.
Fig.2 First level of decomposition
Fig.3 Second level of Decomposition
III. BI-ORTHOGONAL WAVELET
TRANSFORM
Already known that bases that span a space
do not have to be orthogonal. In order to gain greater
flexibility in the construction of wavelet bases, the
orthogonality condition is relaxed allowing semi-
orthogonal, bi-orthogonal or non-orthogonal wavelet
bases. Bi-orthogonal Wavelets are families of
compactly supported symmetric wavelets. The
symmetry of the filter coefficients is often desirable
since it results in linear phase of the transfer function.
In the bi-orthogonal case, rather than having one
scaling and wavelet function, there are two scaling
functions  ~, that may generate different
multiresolution analysis, and accordingly two
different wavelet functions 𝛹, is used in the
analysis and is used in the synthesis. In addition, the
scaling functions 𝜑, and the wavelet functions 𝛹,
are related by duality in the following equations
(1) and (2)
  0)(~)( ,, dxxx kjkj  (1)
as soon as j ≠ j or k ≠ 𝑘́and even.
  0)()( ,0, dxxx kko  (2)
as soon as kk 
There are two sequences, ng and nh to act
as decomposition sequences and two sequences to act
as reconstruction sequences. If
1
nc is a data set, it
should be decomposed as which is shown in the
equation (3) and (4)
1
2
0
k
k
knn chc   (3)
 
k
kknn cgd 1
2
0
(4)
For reconstruction it is shown as
  
n
nlnnlnl dgchc 0
2
0
2
1 ~~
(5)
If we want perfect reconstruction, so
decomposing and then reconstructing shouldn‟t
change anything. This imposes some conditions
shown in equation (6) & (7)
INPUT
IMAGE
2D BI-
ORTHOGONAL
WAVELET
FILTER
FIXED
POINT
REPRESENT
ATION
RECON
STRUC
TED
IMAGE

n
n
n hg 


~
)1( 1
(6)
lhg n
n
n 

 1
)1(~ (7)
The separation of analysis and synthesis is
such that the useful properties for analysis (e.g.,
oscillations, zero moments) can be concentrated on
the ~ function. The interesting property for
synthesis (regularity) which is assigned to the 𝛹
function has proven to be very useful.
The dual scaling and wavelet functions have the
following properties:
(a) They are zero outside of a segment.
(b) The calculation algorithms are maintained,
and thus very simple.
(c) The associated filters are symmetrical.
(d) The functions used in the calculations are
easier to build numerically than those used in the
Daubechies wavelets.
IV. FIXED POINT APPROXIMATION
The designs of a wavelet transform
processor, to treat the data and filter coefficient
values in fixed-point values. Fixed point has the
advantages of being easier to implement, requires
less silicon area, and makes multiplications faster to
perform. Floating point allows a greater range of
numbers, though floating point numbers require 32 or
64 bits. Fixed-point numbers can be 8 to 32 bits (or
more), possibly saving space in the multiplier.
Fig.3 Original image
The wavelet transform has compact support,
and the coefficients have small sizes that do not take
advantage of the floating-point number's wide range.
Also, using fixed-point numbers allows for a simpler
design. It can be explained in another way as, the
extra hardware that floating point requires is wasted
on the DWT. A standard design for integer
multiplications high speed Booth multiplier is
needed. In the horizontal and vertical filters, decimal
numbers including some bits for the fractional part
represent the values. When the results leave the
filters, the values are truncated to fixed format.
Fig.4 Approximation details of Barbara image
Fig.5 Reconstructed image based on fixed point
representation using daubechies wavelet transform
Fig.6 Reconstructed image based on fixed point
representation using bior wavelet transform
For multiresolution, the DWT coefficient's
range will grow. The increase is upper-bounded by
approximately 2. An extra bit of precision for each
transform octave or dimension will be needed. For a
1-D architecture, also notes that 12 bits of precision

are enough. It is also noted that wavelet transformed
data is naturally floating point, but it itself round it to
the nearest integer values, to minimize data loss. The
chip has 16x16 size in multipliers, but keep 16 bits of
the result, which is chosen by software. Therefore,
fixed-point numbers with an assumed radix is used to
work for computing the DWT.
For the computation of a 2-D DWT or
IDWT, we need to recover the integer part of the
result and the most significant bit of the fractional
part, to allow correct rounding. To produce output
results in a conventional number system, we make a
final substitution of the representation of z in the
polynomial representation of the output data. Since z
is irrational, we will not find an error-free finite
representation within a conventional weighted system
(e.g. binary) therefore, we have to use an
approximation, and the error relating to that
approximation dictates the quality of the output
result.
Fig.7 PSNR graph for Barbara image
Fig.8 Percentage of error reduction graph for Barbara
image
The proposed structure is a lattice design for
the bi-orthogonal 7/3 filter bank using lattice
structures. The design is verified using matlab
programming. They also gives the objective results of
the matlab representation for this design in terms of
the PSNR values when using some of the standard
gray scale images. In addition, the results of the input
gray scale images are compared with the resulting
images of the proposed structure are also shown
using cascade analysis-synthesis.
V. RESULTS
Matlab simulation is done for the given
biorthogonal 7/3 and daubechies 4 and 6 wavelet
structure. Fig.3 is an input image taken for
decomposition and reconstruction based on fixed
point representation. Fig 7 and 8 shows the
comparison of PSNR values and percentage of error
reduction at different times for both daub4, daub8
and bior 7/3 structures.
Fig.9 Noisy Image
Fig.10 Denoised Image using daubechies wavelet
Parameters daub 4 bior 7/3
PSNR(dB) +15.55250 +311.47517
MSE 1.8249e+003 4.6662e-027
Table.1 Comparison between daubechies and Bi-
orthogonal Wavelet

Fig.11 Denoised Image using bior wavelet
Noisy image also denoised by using bior and
daubechies wavelet transform. This is shown in fig
10 and 11. Table.1 shows the comparison between
daubechies 4 and bi-orthogonal 7/3 wavelet. PSNR
and MSE value shows that bi-orthogonal wavelet
transform provides increased performance when
compared to daubechies wavelet transform.
VI. CONCLUSION
The existing daubechies 4 and 6 wavelet is
compared with bi-orthogonal wavelet by finding the
PSNR and MSE values of an image. The coefficients
of both the wavelets are represented in fixed point.
PSNR values and MSE values for each coefficients
are found at different times. The results are simulated
using MATLAB. The results shown that PSNR
values are increased for bi-orthogonal wavelet when
compared to the daubechies 4 and 6 wavelet.
Rounding off the approximation coefficients may
introduce some errors. This can be recovered using
different encoding techniques.
VII. ACKNOWLEDGEMENT
I would like to thank my guide Mrs.
M.Vidya, Assistant professor, who guided me to
complete this project and also thank everyone those
who supported me to achieve my target.
REFERENCES
[1] Shiva Kumar Madishetty, Arjuna
Madanayake, Renato J. Cintra, VLSI
Architectures for the 4-Tap and 6-Tap 2-D
Daubechies Wavelet Filters Using Algebraic
Integers,” IEEE transactions on circuits and
systems, june 2013Feng M.L. and Tan Y.P.
(2004) „Contrast adaptive binarization of
low quality document images‟ IEICE
Electron. Express, vol. 1, no. 16, pp. 501–
506.
[2] Michael Weeks and Qin Wang, “Orthogonal
Wavelet Coefficient Precision and Fixed
point representation” Department of
Computer Science, Georgia University,
2001.
[3] A. Ruiz, A. R. Arnauv and M. Ardinav, “2D
Wavelet Transform using Fixed Point
Number Representation”, University of
spain CICTY, 2001.
[4] K. Wahid, V. Dimitrov, and G. Jullien,
“VLSI architectures of Daubechies wavelets
for algebraic integers,” J. Circuits
Syst.Comput., vol.13, no.6, pp. 1251-1270,
2004.
[5] R. Baghaie and V. Dimitrov, “Computing
Haar transform using algebraic integers,” in
Proc. Conf. Signals, Systems Computers
Record 34th
Asilomar Conf.,vol.1, pp. 438-
442 ,2000
[6] B. K. Mohanty, A. Mahajan, and P. K.
Meher, “Area and power efficient
architecture for high-throughput
implementation of lifting 2-ddwt,” IEEE
Trans. Circuits. Syst. II, Jul. 2012.
[7] K. A. Wahid, V. S. Dimitrov, G. A. Jullien,
and W. Badawy, “An algebraic integer
based encoding scheme for implementing
Daubechies discrete wavelet transforms,” in
Proc. Asilomar Conf. Signals, Syst. Comp.,
pp. 967-971, 2002.
[8] K. A. Wahid, V. S. Dimitrov, and G. A.
Jullien, “Error-free arithmetic for discrete
wavelet transforms using algebraic
integers,” in Proc. 16th
IEEE Symp.
Computer Arithmetic, pp. 238-244, 2003.
[9] M. A. Islam and K. A. Wahid, “Area- and
power-efficient design of Daubechies
wavelet transforms using folded AIQ
mapping,” IEEE Trans. Circuits Syst. II, pp.
716-720, Sep. 2010.
[10] K. A. Wahid, V. S. Dimitrov, G. A. Jullien,
and W. Badawy, “Error-free computation of
Daubechies wavelets for image compression
applications,” Electron. Lett., pp. 428-429,
2003.

R044120124

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to R044120124 (20)

Recently uploaded (20)

R044120124