IEEE—ICET 2006

2nd International Conference on Emerging Technologies

Peshawar, Pakistan 13-14 November 2006

Image Compression with Different Types of

Wavelets
Javed Akhtar, Dr Muhammad Younus Javed
Department of Computer Engineering, College of E&ME
National University of Sciences and Technology,
Rawalpindi Pakistan
[email protected], [email protected]

Abstract: - Data compression which can be lossy or decoder decodes the coded representation, up
lossless is required to decrease the storage samples and reconstructs the signal using a
requirement and better data transfer rate. One of the synthesis filter bank [1].
best image compression techniques is using wavelet
transform. It is comparatively new and has many The JPEG2000 standard which is the
advantages over others. replacement of JPEG uses wavelet transform
Wavelet transform uses a large variety of instead of DCT (discrete Cosine Transform) to
wavelets for decomposition of images. The state of perform decomposition of the image.
the art coding techniques like EZW, SPIHT (set An M u N image is decomposed using wavelet
partitioning in hierarchical trees) and EBCOT transform. Each row is filtered and then down
(embedded block coding with optimized truncation)
sampled which gives two N u M/2 images. Then
use the wavelet transform as basic and common step
for their own further technical advantages. The each column is filtered and sub sampled which
wavelet transform results therefore have the gives four N/2 u M/2 images. Of these four sub
importance which is dependent on the type of wavelet images, the one obtained by low pass filtering
used. rows and columns is referred as LL image, the
one obtained by low pass filtering the rows and
In this paper, different wavelets have been used
to perform the transform of a test image and their high pass filtering the columns is referred to as
results have been discussed and analyzed. The the LH image and other two are HL and HH
analysis has been carried out in terms of PSNR (peak images. Each of these sub images can be further
signal to noise ratio) obtained and time taken for decomposed in similar manner. A quantized
decomposition and reconstruction. SPIHT coding technique can be applied on few/all of these
algorithm is considered as a basic standard in final sub images to get the compression and
compression field using wavelet transform. In other advantages [3] [4].
addition to wavelet analysis for simple
decomposition, analysis of SPIHT coding algorithm EZW (Embedded Zero Wavelet Trees), SPIHT
in terms of PSNR for different wavelets is also (Set Partitioning in Hierarchical Trees) and
carried out here. This analysis will help in choosing EBCOT (Embedded Block Coding with
the wavelet for decomposition of images as per their Optimizes Truncation) are few very important
application.
wavelet compression techniques introduced
which have their own standing in this field.
1. INTRODUCTION EZW is one of the first techniques which gave
Wavelet transform is being used for image image compression a new direction which was
compression. It is one of its most popular superior to JPEG standard. SPIHT was further
applications. The state of art technique for image improvement in this direction and is considered
compression use wavelet transform for this as a basis for comparison. EBCOT introduces
purposes nowadays. the latest and versatile ideas in this field and is
the basis of JPEG2000.
The practical implementation of wavelet
compression scheme is very similar to that of SPIHT results have also been obtained and
sub band coding schemes. But this technique discussed here as it is the basic algorithm being
uses the wavelets filters instead of sinusoids for used for comparison purposes in this field. It is
decomposition. The outputs of the filter banks one of the quantization techniques which
are down sampled, quantized and encoded. The decompose the LL sub images till certain levels.

1-4244-0502-5/06/$20.00©2006 IEEE 133

It uses a quantization and sort of bit plane 3. RESULTS AND DISCUSSION
coding technique to code all the sub images
[2][5]. It gives very good results as far as All above family wavelets have been used
compression and embedding technique is to obtain the results. A “woman” image of
concerned. size 256 x 256 from wavelet toolbox has
been used for this paper. All the experiments
In section 2, a very brief introduction and
characteristics of different types of wavelets
have been performed on Pentium IV, 512
have been given. Section 3 discusses the results MB RAM with 2.4 GHZ machine.
obtained by using different types of wavelets for MATLAB has been used for implementation
a test image in terms of time taken by a codec of algorithms.
and PSNR for second level decomposition
approximation & SPIHT technique. Finally Second level decomposition approxima-
conclusion is drawn in section 4. tion has been used for simple compression
as it does not give good results for higher
2. WAVELETS DESCRIPTION level decompositions for which SPIHT has
There are many wavelets being used now a day been used [6]. Same image has been used
for decomposition of signals and images. The for SPIHT coding algorithm. The level of
main types/families are daubechies, decomposition used for this coding
biorthogonal, coiflets, symlets and dmey. The technique is level–6 as it does not give good
major characteristics of different types of results for lower decomposition levels [7].
wavelets/families are briefly described here [8]. The bit rate used for this is 1 bpp.
The Daubechies Wavelets are compactly
supported and have highest number of vanishing 3.1. Daubechies Wavelets
moments. The types are db1, db2…db45. The results for these wavelets are given in
Discrete Wavelet Transform and continuous Table 1 and Graph 1. The time taken for dbs is
wavelet transform is possible for these. These increasing with only exception of db10. The
are not symmetrical. The length of the filter is highest PSNR for wavelet decomposition is
2N. The number of vanishing moments are N. given by db25. All others are also comparable as
there is no unusual change in the performance.
The Biorthogonal Wavelets are compactly
For SPIHT result is almost some what same but
supported wavelets for which symmetry and
the highest PSNR is given by db10. Keeping the
exact reconstruction is possible with FIR filters.
results in mind db10 and db25 are recommended
The types are bior1.1, bior1.3, bior1.5, bior2.2,
as these have the inherent property to give better
bior2.4 …etc. Discrete and continuous wavelet
results.
transforms are possible with these.
The Symlets Wavelets are compactly 3.2 Biorthogonal Wavelets
supported wavelets with highest number of The results of these wavelets are given in
vanishing moments. The types are sym1, sym2, Table 2 and Graph 2. The time taken for wavelet
sym3, sym4 and sym5. Discrete and continuous decomposition and reconstruction is not linear.
wavelet transform is possible with these types of The highest PSNR is given by bior6.8. Keeping
wavelets. Filter length is 2N. this in mind and time taken which is also very
reasonable bior6.8 is recommended as these
Coiflets wavelets are compactly supported possess the inherent property to give better
wavelets with highest number of vanishing results.
moments for both psi and phi for a given support
width. Discrete and continuous wavelet 3.3 Coiflets and Dmeyer Wavelets
transform is possible with these.
The results for Coiflets Wavelets are given in
Discrete Meyer wavelet is FIR based approx of Table 3 and Graph 3. The time taken does not
the Meyer wavelet. Meyer wavelet is infinitely have some proper pattern. The coif5 gives better
regular orthogonal wavelet results for wavelet decomposition and SPIHT.

134
 The Dmeyer wavelet results are shown at the 4. CONCLUSION
end of Table 3. It is to be noted that it gives very
As wavelet image compression has
good PSNR for both with slightly more time
taken. It also has the inherent ability to give revolutionized image compression field with
good results. unbelievable results. This involves the state of
art techniques but wavelet decomposition
3.4 Symlets Wavelets remains the initial step for all these including
wavelet packets techniques. Therefore there was
The results of these types of wavelets are given
a need to exploit the inherent ability of wavelets.
in Table 4 and Graph 4. The time taken for
The analysis was carried out keeping in mind the
wavelet decomposition and reconstruction is
fact that if decomposition produces good results
increasing as we go up. The best PSNR is given
it will also give better chances to advanced
by sym10 and sym25 for decomposition and
techniques for further improved results [3].
SPIHT. So these wavelets are recommended for
SPIHT has also been used here for this purpose.
compression of images. There is one thing to
Biorthogonal and Coiflets wavelets have the
note that as we go up beyond sym25 the time
same type of wavelets giving better results for
taken increases very rapidly which may make
both. Generally speaking it is obvious from the
them impractical for certain applications.
results that any wavelet giving good results for
In addition to above the best SPIHT images decomposition will produce good results for
reconstructed for different wavelets families are advanced techniques being used for image
shown in Figure 1. compression

Table 1. Debauchies Wavelets Analysis Table 2. Biorthogonal Wavelets Analysis

Table. 3. Coiflets Wavelets Analysis Table 4. Symlets Wavelets Analysis

135
 Graph 1. Daubechies Wavelets Graph 2. Biorthogonal Wavelets

Graph 3. Coiflets Wavelets Graph 4. Symlets Wavelets

Figure. 1. SPIHT images for (from left to right) db10, cof5, bior6.8 and sym25.

There is need to carry out study that [2] J. M. Shapiro, “Embedded image coding using
zerotrees of wavelet coefficients”, IEEE Trans.
involves different images and different Signal Processing, vol. 41, pp. 3445-3462, 1993.
decomposition levels to get more accurate
[3] Rafael C. Gonzalez and Richard E. Woods, “Digital
results. Image Processing”, 2nd Edition, Prentice Hall Inc,
2002.
REFERENCES
[4] Paul S Addison, “The Illustrated Wavelet Transform
[1] M. Antonini, M. Barland, P. Mathien and I. Handbook”, IOP Publishing Ltd., 2002.
Daubechies, “Image coding using wavelet
transform”, IEEE Trans. Image Processing, vol. 1, [5] Said, A and Pearlman, W. A. “A new, fast and
pp. 205-220, April 1992. efficient image codec based on set partitioning in
hierarchical trees”, IEEE Trans. Circuits and

136
 Systems for Video Technology, Vol. 6, pp. 243-250, [7] Akhtar, J. and Javed, M. Y., “Bit Rate Along With
June 1996. Decomposition Level Analysis of SPIHT
Algorithm” International Conference on Parallel and
[6] Akhtar, J., Javed, M. Y., and Akhtar, M., Distributed Processing Techniques and Applications
“Implementation and Analysis of Wavelet Image PDPTA’05 held on 27-30 Jun 2005 Las Vegas,
Decomposition and SPIHT Algorithm” IEEE USA.
INMIC 2004 Speech, Image and Video Processing
pp. 97-101 December 2004. [8] http://www.mathworks.com.

137