ISSN 2277 – 3916 CVR Journal of Science and Technology, Volume 9, December 2015

Verilog Implementation of Image Compression

Using Discrete Wavelet Transform
P.Sreekanth
CVR College of Engineering
Email: sreekanth.isoft@gmail.com
Abstract: As our use of computer continues to grow, need for Where m and n are integers. Substituting a and b in Eq.
storing large amount of data efficiently increases day by day. 3.1 by Eq. 4.1 the discrete wavelets can be represented by
Image compression has become a necessary step for making Eq.( 4.2).
better use of available storage and transmission bandwidth.
This paper presents image compression using discrete wavelet
m
transform(DWT) implementation using VERILOG. Ȍ m, n (t) a0 -m/2
Ȍ(a 0 t  nb 0 ) (2)
Decompression algorithm is implemented using inverse
wavelet transform (IDWT). Here we used sub-band coding to
implement wavelet transform. By using DWT we get high In general, the wavelet coefficients for function f (t) are

³ f(t)Ȍ (a
compression ratios. The results of VERILOG arediscussed in given by
 m/2 m
this paper.
C m, n (f) a0 0 t  nb0)dt (3)

Keywords: Image compression, Discrete Wavelet Transform

(DWT),Inverse Discrete Wavelet Transform (IDWT),
Compression Ratio(CR) and sub-band coding II. ONE DIMENSIONAL DWT ARCHITECTURE
The discrete wavelets transform (DWT), which
I. INTRODUCTION transforms a discrete time signal to a discrete wavelet
Data compression is the process of converting data representation.The 1-D DWT is given as the inner product
files into smaller files for efficiency of storage and of the signal f (t) being transformed with each of the
transmission. As one of the enabling technologies of the discrete basis functions. [2], [7]

¢f(t) ȥ m,n (t)²

multimedia revolution, data compression is a key to rapid
progress being made in information technology. It would C m,n (4)
not be practical to put images, audio and video alone on
websites without compression. Data compression
algorithms are used in those standards to reduce the
number of bits required to represent an image or video
sequence. Compression is necessary and essential method
for creating image files with manageable and transmittable
sizes. In order to be useful, a compression algorithm has a
corresponding decompression algorithm that, given the
compressed file, reproduces the original file. There are
many types of compression algorithms developed. These
algorithms fall into two broad types: lossless algorithms Figure 1: One Dimensional DWT
and lossy algorithms. A lossless algorithm reproduces the
original exactly. A lossy algorithm loses some data. Text The generic form of 1 D DWT is shown in above
compression must be lossless because a very small figure1. Here the discrete signal is passed through low pass
difference can result in statements with totally different and high pass filters H and G then down sampled by factor
meanings. There are also many situations where loss may 2 completes forward wavelet transform [6], [7] . Here the
be either unnoticeable or acceptable. In image low pass filtering and down sampling means averaging
compression, for example the exact reconstruction of each operation. Low pass signal contains approximation
sample of the image is not necessary. Depending on the coefficients which contain important information of
quality required of each sample of the image, varying original signal. High pass filtering and down sampling
amounts of loss of information can be accepted. means differencing operation. High pass signal consists of
Since the input signal (e.g., a digital image) is detail coefficients which contain very less information of
processed by a digital computing machine, it is prudent to original signal.
define the discrete version of the wavelet transform. To The low pass filter (H) and high pass filter (G) in one
define the wavelet in terms of discrete values of the dimensional forward dwt combined and it called as analyze
dilation and translation parameters ‘a’ and ‘b’ instead of filter bank [6], [7].
being continuous, make a and b discrete.
(1)

DOI: 10.32377/cvrjst0905 CVR College of Engineering 21

ISSN 2277 – 3916 CVR Journal of Science and Technology, Volume 9, December 2015

III. ONE DIMENSIONAL IDWT ARCHITECTURE The above diagram shows forward wavelet transform
of level 1 [8]. In this diagram ‘H’ represents low pass
The 1-D Inverse DWT given as filter, ‘G’ represents high pass filter. The both filters are

¦ ¦C
combined called as analyze filter bank. Here input is
f f image which is 2 dimensional signals denoted by x (m, n).
f (t ) m ,n ( f )\ m ,n ( t ) Then apply one dimensional DWT row wise and down
m f n f (5) sampling divides input image into two bands. Low pass
filtering band contains very important information which
looks like image and high pass filtering part contain very
less information which look like noise. Then similarly
applying one dimensional DWT column wise completes
the level 1 DWT for image. [3]Here input is original image
and output is compressed image.

V. IDWT FOR TWO – DIMENSIONAL SIGNAL

The below figure.4 shows Inverse wavelet transform of
level 1. In this diagram ‘H1’ represents low pass filter,
‘G1’ represents high pass filter. Both filters are combined
called as synthesize filter bank. First we have to apply 1 D
Figure 2 : One Dimensional IDWT DWT column wise to compressed image then apply
followed by row wise .This completes the level 1
The generic form of 1 D IDWT is shown in figure 2. IDWT.Input to this block is output of level 1 DWT image
The low pass filter (H1) and high pass filter (G1) in one means compressed image. This compressed image then
dimensional Inverse DWT combined and it is called as apply to up sampler first then followed by synthesize filter
decomposition filter bank [2], [7]. bank in column wise. This output is again given to up
Here input is output of one dimensional DWT. The low sampler followed by synthesis filter bank in row wise. We
pass signal and high pass signal passe’s through up can interchange the order of columns and rows. The output
sampler first followed by decomposition filter bank next. is reconstructed image. Reconstructed image size is same
By combining the outputs of low pass filter (H1) and high as original image size.
pass filter (G1) ,we can get original signal back. [8]

IV.DWT FOR TWO – DIMENSIONAL SIGNAL

The two-dimensional extension of DWT is essential for
transformation of two-dimensional signals, such as a
digital image. A two-dimensional digital signal can be
represented by a two-dimensional array x (m, n) with m
rows and n columns, where m and n are nonnegative
integers. The simple approach for two dimensional
implementation of the DWT is to perform the one-
dimensional DWT row-wise to produce an intermediate
result and then perform the same one-dimensional DWT
column-wise on this intermediate result to produce the
final result [11]. This is shown in Figure4. 3. [2],[3],[4]. Figure 4: Level One IDWT for Image

VI. HAAR WAVELET TRANSFORM

Consider a, b are pixel values of image. Then forward
HAAR wavelet transform contain two parts which are
smoothing coefficients and detail coefficients, denoted by
S and D respectively[9],[10]

(a  b)/2 ½
¾
S
D (a  b)/2 ¿
(6)

We have to apply above algorithm to all pixel values of

the image; it completes image compression using forward
DWT. Smoothing coefficients obtained by low pass
Figure 3: Level One DWT for Image
filtering and down sampling by factor of ‘2’. Low pass

22 CVR College of Engineering DOI: 10.32377/cvrjst0905

ISSN 2277 – 3916 CVR Journal of Science and Technology, Volume 9, December 2015

filtering coefficients contain very important information of DWT Block consists two sub blocks,Pixel Averaging
image. We can reconstruct whole image with only low Unit and Pixel Subtraction Unit.
pass filtering coefficients. Detail coefficients obtained by
A) Pixel Averaging Unit
high pass filtering and down sampling by factor of ‘2’.
High pass filtering coefficients contain very less
information of image. We neglect these coefficients if want
high compression.
The image reconstruction is obtained by Inverse
Discrete Wavelet Transform (IDWT).

a (S  D)½
¾
b (S  D) ¿
Figure 7: Pixel Averaging Unit
(7)
Pixel averaging unit is to perform average of two
pixels. In above figure, p1 and p2 are input pixel values,
VII. VERILOG IMPLEMENTATION OF DWT clk is clock signal and pout is output pixel value after
performing averaging operation. Here input pixel values in
form of binary and ‘9’ bits are required to represent each
pixel value.

Figure 5: Block diagram for image Compression

Fig. 5 shows Block diagram of image compression

using Discrete Wavelet Transform (DWT) in
VERILOG.Here memory 1 is input memory where we
store input image in text format. Generally image is two
dimensional signal of size (m x n), where ‘m’ is number of
rows and ‘n’ is number of columns. But VERILOG HDL
language does not support two dimensional memories, but Figure 8: Inner view of Pixel Averaging Unit
supports only one dimensional memory. So we need to
convert two dimensional memories to one dimensional Pixel average unit consists two blocks in it as shown in
memory. By using MATLAB, input image is converted to figure 8. The first block is ‘Right shifter’ which helps to
binary text file. If input image is of size (8x8) which is perform dividing operation. Andthe second block is ‘9 bit
two dimensional then by using MATLAB we convert it to binary’ adder which helps to add two right shifted versions
(64x1) binary text file which is one dimensional, used for of pixel values. These two blocks implement
DWT operations in VERILOG. Each pixel value of the p1 p2
image is represented by 9 bits. (  ), same as smoothing coefficient calculation
We apply above HAAR wavelet transform DWT & 2 2
IDWT algorithm in between two pixels of entire image given in equation 6.
row wise followed by column wise completes compression B) Pixel Subtraction Unit
of image using Discrete Wavelet Transform (DWT).
The outer view of DWT Block which is implemented
in VERILOG is shown in fig 6. Here ‘clk’ is clock signal
which is common for all inner blocks of DWT, ‘memrd’ is
used for reading pixel values from memory to perform
operations according to equations and ‘memwr’ is used for
writing the resultant pixel values in memory. Here mem2 Figure 9: Pixel Subtraction Unit
is output memory to display output pixel values.
Pixel subtraction unit is to perform subtraction of two
pixels. In above figure, p1 and p2 are input pixel values,
‘clk’ is clock signal and pout is output pixel value after
performing averaging operation and ‘psub’ is output of
pixel subtraction unit. Here input pixel values in the form
of binary and ‘9’ bits are required to represent each pixel
value.

Figure 6: DWT block

CVR College of Engineering 23

DOI: 10.32377/cvrjst0905
ISSN 2277 – 3916 CVR Journal of Science and Technology, Volume 9, December 2015

Figure 12: Simulation results of DWT block

C) IDWT Implementation In VERILOG

Figure 10: Inner view of Pixel Subtraction Unit

Pixel subtraction unit contain two blocks in it. Right shifter Figure 13: Block diagram for image Reconstruction
is followed by 9 bit sub tractor. The output of pixel
p1 p2 IDWT block includes one binary adder and binary sub
subtractor unit is (  ) where p1 and p2 are pixel tractor to perform operations on the output of DWT block
2 2 (smoothing and detail coefficients) pixel values as given in
values of image, same as equation 6. equation 3 and 4.
The pixel values of reconstructed image are stored in
binary text file in memory as one dimensional. We see the
results using MATLAB by converting one dimensional
memory to two dimensional memory. Have been taken
(64X64) Lena image as shown fig 14(a) After applying
DWT algorithm, the compressed DWT image is as shown
in fig 14 (b). Then has been taken only LL part of image to
reconstruct the original image it is shown in fig 14 (c). The
error between original and reconstructed is as shown in fig
14 (d). Here compression ratio of 4 was obtained.

Figure 11: Connections inside the DWT block

This pixel averaging and subtraction units are connected in

DWT block to complete the Forward DWT for two (a) (b)
dimensional signals (image) for rows and columns
separately. The figure.11 shows the connections of Pixel
averaging and subtraction units in DWT block.

© (d)
Figure 14: image compression results

Taking total parts of fig 14 (b), then the reconstructed

image is shown in fig 15. Here image quality increases but
compression ratio is decreased to 1.

Figure 15: recontructed image using all parts of DWT

Simlarly, The same (64 X64) DWT and IDWT

algorithms to (128 X 128) image with 4 iterations has been
taken. The results are shown in fig 16.Figure 16(a) shows
the original image and figure 16(b) shows the
reconstructed image with compression ratio of 4. The
figure 16 (c) shows the error between orginal and
reconstructed image

24 CVR College of Engineering

DOI: 10.32377/cvrjst0905
ISSN 2277 – 3916 CVR Journal of Science and Technology, Volume 9, December 2015

(a) (b)

©
Figure 16: results for Lena (128X128)

VII CONCLUSIONS
The need for efficient ways of storing large amount of
data is increasing day by day. For example, if we want to
have a web page or online catalog having hundreds of
images, we essentially look for some kind of image
compression to have those images stored. This is because
the amount of space required holding uncompressed image
costs more. In this paper we discussed how image
compression is done in VERILOG HDL language. Image
compression and reconstruction are displayed in this paper.
We observe that as the size of image is increasing, we
get better reconstructed image quality.

REFERENCES
[1.] RatchakitSakuldee and SomkaitUdomhunsakul, “Objective
Performance of Compressed Image Quality Assessments”,
World Academy of Science, Engineering and Technology
35 2007
[2.] http://en.wikipedia.org/wiki/Discrete_wavelet_transform.
[3.] Jyotishman Das, “Lossless performance of image
compression using 2D DWT”
[4.] Amir Averbuch, “wavelet Compression”, School of
Computer Science Tel- Aviv University
[5.] Kamrul Hasan Talukder and Koichi Harada, “Discrete
Wavelet Transform for Image Compression and A Model
of Parallel Image Compression Schemefor
FormalVerification”, WCE 2007, July 2 - 4, 2007, London,
U.K.
[6.] Chao-Tsung Huang, Po- Chih Tseng, and Liang-Gee
Chen,” Analysis and VLSIArchitecture for 1-D and 2-D
Discrete Wavelet Transform”. IEEETransactions onSignal
Processing, vol. 53, No. 4, April 2005
[7.] Sanjay kumarMitra, “Digital Signal Processing”, Fourth
edition, pp 15-56,2010
[8.] Bryan E. Usevitch, “A Tutorial on Modern Lossy Wavelet
Image Compression:Foundations of JPEG 2000”.
[9.] P. Raviraj and M.Y. Sanavullah,” The Modified 2D-Haar
Wavelet Transformation in Image Compression”, Middle-
East Journal of Scientific Research 2 (2): 73-78, 2007.
[10.] Nidhi Sethi, Ram Krishna and Prof R.P. Arora , “Image
Compression Using HaarWavelet Transform”, Computer
Engineering and Intelligent Systems, ISSN 2222-171

CVR College of Engineering 25

DOI: 10.32377/cvrjst0905