You are on page 1of 4

Image Compression Based on Wavelet Transform

Liu Bo, Yang Zhaorong


The higher education key laboratory for Measuring &Control Technology and Instrumentations of Heilongjiang Province
Harbin University of Science and Technology
Harbin China
fen__dou@163.com


AbstractAccording to the basic theory of wavelet, this paper
used matlab software simulation to study image compression
problem. Compared with traditional methods, it is more simple
and effective.Wavelet image compression technology solves the
contradiction of time and frequency domains localization
which can not be achieved by the Fourier transform. For
improving the image compression ratio and compression speed
this study has actual reference value.
Keywords- wavelet analysis, image compression, matlab
I. INTRODUCTION
The wavelet transform is established and developed on the
basis of the Fourier transform. Bit there are some differences
between them. The difference is that the Fourier transform is
overall domain analysis which use a separate time or frequency
domain representation of the signal characteristics
[1]
. The
wavelet transform is localized time-frequency analysis, it joint
time domain and frequency domain to represent signal
characteristics. Wavelet transform overcome some defects of
the Fourier transform, It can more effectively extract time
domain and frequency domain localized information from the
characterization of signal and achieve multi-scale refinement, it
is widely used in Time-frequency analysis of the various
fiekds,is known as the mathematical microscope.
in recent years, wavelet transform rapid develop, image
compression is an important aspect of the application of
wavelet transform. Its characteristic is high compression ratio
and compression speed, compressed signal and image
characteristics can be maintained unchanged, and anti-
interference in the transmission
[2]
. There is a lot of method
based on wavelet transform compression.
II. WAVELET TRANSFORM
Wavelet transform is based on the Fourier transform of the
concept of changing the time window. It uses the time domain
and the scale domain. It is based on the needs of low and high
frequency information to adjust the length of time window. The
scale and time is inversely proportional
[3]
. Effective solution to
the problem of non-stationary signals in high and low
frequency localized. It is a fixed window size, but its shape can
change. It has a time window and frequency window can be
changed in the time-frequency domain analysis method.It has a
higher frequency resolution and lower time resolution in the
low-frequency points, and it has a high time resolution and
lower frequency resolution in the high frequency part. It is
called a mathematical microscope
[4]
. Wavelet transform with
adaptive signal.
Wavelet transform the meaning of
( ) t
is the basic
wavelet function. Do it the displacement of the

. It under
different scales

and the signal


( ) t
to be analyzed within
the plot.
1
( , ) ( ) *
t
WT t dt

| |
=
|
\ .
}
>0
Equivalent frequency-domain representation:
( ) ( , ) ( ) *
2
j
WT X e d

=
}

In the formula,
( ) X
and
( )
are the Fourier
transform of
( ) t
,
( ) t
.
Wavelets in the wavelet transform is not unique. Wavelet
basis function
( ) t
is the diversity. In practical applications,
we select the
( ) t
is very important. Analysis of the same
issue will produce different results because we use different
wavelet bases
[5]
. Here are a few examples of wavelet bases:
1) Haar orthogonal wavelet bases
1
( ) t


| s s |
|
= s s
|
|
. \
0 1/2
-1, 1/2 1
0, others

As shown in figure 1.

Figure 1. Harr wavelet
145
2012 International Conference on Measurement, Information and Control (MIC)
978-1-4577-1604-1/12/$26.00 2012 IEEE May 18-20, 2012
2) Daubechies
It is a famous French scholar Daubechies proposed by the
binary wavelet and dbN is the general term.N denotes the serial
number of the wavelet.As shown in figure2.The expression is:
( )
1
1
0
N
N k k
K
k
P y C y

+
=
=


1 N k
K
C
+
is the binomial coefficient in the formula.There are:
2 2 2
0
| ( ) | cos sin
2 2
N
m P

| | | |
=
| |
\ . \ .
,
1
0
0
1
( )
2
N
jk
k
k
m h e

=
=




(a)Scaling function (b) Wavelet function
Figure 2. Db10 wavelet
3) Mexican Hat (mexh) wavele
Because its waveform is like a Mexican hat, hence the
name.As shown in figure 3.Its function is Gaus,the second
derivative:
( ) ( )
2
2
2
1
t
t t e

=
( )
2
2
2
2 e



=


Figure 3. Mexican Hat wavelet

It has good localization in time domain and frequency
domain. And that
( ) 0
R
t dt =
}
. It does not exist a scaling
function, so I do not have orthogonality. Later development of
the formation of a continuous wavelet transform, two-
dimensional wavelet transform.
III. WAVELET TRANSFORM AND IMAGE COMPRESSION
Between pixels, the image data signal is generally a
correlation, they are adjacent to the strongest correlation
between the pixel rows or columns, and their correlation
coefficient is the law of exponential decay. Image related to the
existence of a theoretical basis for image compression
[6]
. In
order to use some method to remove redundant information, to
achieve the purpose of compression. Image compression
method based on wavelet transform has been gradually
replaced based on discrete cosine or other descendants coding
techniques,while a new image compression standard.Wavelet
image compression process shown in figure 4.

Figure 4. Figure based on wavelet transform image compression process
A. Matlab image compression
There are many ways based on wavelet transform image
compression, the success of the wavelet transform zero-tree
compression, wavelet variable vector compression.
Dimensional wavelet transform is more simple and effective.
1) The use of two-dimensional wavelet transform for
image compression
In the two-dimensional wavelet transform, Images by
wavelet transform decomposition of a series of sub-images at
different frequencies to get the low-frequency coefficients, the
level of high frequency coefficients, vertical high-frequency
coefficients,the diagonal high frequency
[7]
.As shown in
figure5and6.Most of the points on the high-frequency sub-
image values are close to zero, the higher the frequency
performance of the more obvious. Because the image is the
main low-frequency part, you can use the wavelet transform
decomposition to remove the high frequency part of the image
while retaining the low-frequency part. So, also known as two-
dimensional wavelet transform compression to retain the low-
frequency information compression
[8]
.
Defined as two-dimensional discrete wavelet
transform

1 2
1 2 1 2
1 2 1 2
1
( , , , ) ( , )
X b X b
a a b b
a a a a


=

1 2 1 2
( , , , ) a a b b
is the wavelet coefficients of the
signal X.
1 2
, a a
,said telescopic extent.
1 2
, b b
said the location
of the translation.

said the mother wavelet transform


function.
After wavelet transformation, the image is divided into high
and low frequency information. We get a better image to
remove high-frequency information storage , the use of inverse
wavelet transform according to the specific requirements of the
compression qualitymultiple.We add the appropriate amount
of detail, and reconstruct a new image, and can reduce the
146
distortion, as high compression than compress images to reach
the demand shown in figure 7.

Original image

Before compression to the size of the images

Figure 5. Original images and data tables


Decomposition of the low and high frequency information
Figure 6. Each component of the decomposition diagram

(a)First compressed image (b) Second compressed image



Figure 7. The two compression comparison with the original image
information
It can be seen from the results of the first compression in
order to extract the original image in the wavelet
decomposition of the first layer of low-frequency information,
and we can get the compression is relatively small . The second
compression is to extract the first layer decomposition of low-
frequency part of the low-frequency part of the second layer of
low-frequency part
[9]
. Theoretically, we can extract the wavelet
decomposition of N-tier. You can get any compression ratio of
the image. As shown in Figure 8.

L
L
H
L
3
HL2
HL1
L
H
3
H
H
3
LH2 HH2
LH1 HH1
Figure 8. image can be separated two dimensional multiresolution
decomposition of the three levels

The higher the level of picture decomposition, it's the
greater the distortion. We directly use the low-frequency part
of the compressed image. The images are blurred to the naked
eye can not distinguish between the image content. Image with
the hierarchical increase in the number, the compression rat io
is decreasing, the visual effects will be the worse
[10]
. The image
is difficult to distinguish from the visual content.
2) the wdencmp method of image compression
processing
a) [XC,CXC,LXC,PERF0,PERFL2]=wdencmp('gbl',X,'
wname',N,THR,SORH,KEEPAPP)
b) [XC,CXC,LXC,PERF0,PERFL2]=wdencmp('lvd',X,'
wname',N,THR,SORH)
c) [XC,CXC,LXC,PERF0,PERFL2]=wdencmp('lvd',C,
L,'wname',N,THR,SORH)
The function wdencmp for one or two dimensional signal
denoising or compression. The wname using a wavelet
function. gbl (the global abbreviation) each layer with a
threshold for processing. lvd said on each deal with different
thresholds. N denotes the number of layers of wavelet
decomposition. THR is the threshold vector. Format (2) (3) on
each floor are required to have a threshold. The length of the
threshold vector THR for N. SORH said that the Select soft
threshold and hard threshold (respectively, 's' and 'h).
Parameters KEEPAPP value of 1 when the low frequency
coefficients without thresholding processing. The contrary, the
low frequency coefficients thresholding. XC is the signal de-
noising or compression. XC wavelet decomposition structure
[CXC, LXC]. PERF0 and PERFL2 norm percentage of
recovery and compression. Percentile indicate that the noise
reduction or compression of the energy components to retain. If
[C, L] is the wavelet decomposition structure of X,
LL HL1
LH1 HH1
Name size Bytes
Original 1368*2089 2857752
First 691*1052 1902319
Second 353*533 1505192
147
( )
( )
2
100
2
CXC
PEREL
C
-
=
,
(vector norm) C

Wdencmp method compression effect as shown in figure 9.


Figure 9. Wdencmp compression effect is as follows
IV. SUMMARY
This paper briefly introduced the Fourier transform,
wavelet transform and a few commonly used wavelet functions,
and Matlab two-dimensional discrete wavelet transform image
compression technology.Wavelet time-frequency analysis and
multi-scale analysis of the more prominent features in the
image compression. It has high compression ratio and
compression speed. It also has a good signal to noise ratio. It is
largely to reduce the image storage space .Reduce the network
communication to share bandwidth. It kept the original
characteristics after compression, suitable for the demanding
image compression. As an algorithm, the wavelet transform has
good prospects in this field.

ACKNOWLEDGMENT
Firstly, thank for my mentor Liu Bo, because of his
helping, finally, thanks for the other members in the laboratory.

REFERENCES

[1] Saffor A,Bin Ramli A.R,Nq K.N. Wavelet-based compression of
medical images: Filter-bank selection and evaluation[J]. Australasian
Physical and Engineering Sciences in Medicine,2003,26(2):39-44.
[2] Salama Paul,Rizkalla Mather E,Ekbauer Michael. VHDLimplementation
of the fast wavelet transform[J], Journal of VLSI Signal Processing
Systems for Signal,Image, and VideoTechnology,2006,42(3):223-239.
[3] Fras Matjaz, Planinic Peter,Digital video quality at transmission over
The noisy channel (DCT and DWT-based codec) (Kakovost digitalnega
videa pri prenosu prek umnega kanala (kodek na osnovi DCT inDWT))
[J].ElektrotehniskiVestnik/Electrotechnical Review,2008,75(1-2):70-75.
[4] Zou Yi,Pan Wei,Ao Lu. Efficient JPEG 2000 image compression
scheme for multihop wireless networks[J]. Journal of Donghua
University (EnglishEdition),2010,27(2):229-232.
[5] Nasri Mohsen,Helali Abdelhamid,Sqhaier Halim et al. Efficient JPEG
2000 image compression scheme for multihop wirelessnetworks[J].
Telkomnika,2011,9(2):311-318.
[6] Khash man Adna,Dimliler kamil. Image compression using neural
networks and Haar wavelet, WSEAS Transactions on Signal Processing,
2008,4(5):330-339
[7] Debnath jayanta Kumar,Rahim Newaz Muhammad Syfur. A modified
vector quantization based image compression technique using wavelet
transform, Proceedings of the International Joint Conference on Neural
Networks, 2008,2 (8):171-176
[8] Li Hui,Takei Masahiro. Application of wavelet packet image
compression technique to particle image velocimetry, International
Journal of Applied Electromagnetics and Mechanics,2001,5(4):383-392
[9] Davis Geoffrey. Self-quantized wavelet subtrees: A wavelet-based
theory for fractal image compression, Data Compression Conference
Proceedings,2005,19(5):232-241
[10] Cheng Zhao,Zhang Tianxu,Lu Haifeng, Image compression algorithm
using image restoration based on wavelet analysis, Proceedings of SPIE
- The International Society for Optical Engineering,2009,5(2):131-139

148

You might also like