See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/333936853

A Comparative Study of DCT and DWT Image Compression Techniques

combined with Huffman coding

Article in Jordanian Journal of Computers and Information Technology (JJCIT) · June 2019
DOI: 10.5455/jjcit.71-1554982934

CITATIONS READS

8 2,160

1 author:

Ashraf Yunis Maghari

Islamic University of Gaza
48 PUBLICATIONS 758 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ashraf Yunis Maghari on 27 December 2019.

The user has requested enhancement of the downloaded file.

 73
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

A COMPARATIVE STUDY OF DCT AND DWT IMAGE

COMPRESSION TECHNIQUES COMBINED WITH
HUFFMAN CODING
Ashraf Maghari
(Received: 11-Apr.-2019, Revised: 27-May-2019, Accepted: 9-Jun.-2019)

ABSTRACT
Image compression techniques have been widely used to store and transmit data which requires storage space
and high transfer speed. The explosive growth of high-quality photos leads to the requirement of efficient
technique to store and exchange data over the internet. In this paper, we present a comparative study to
compare between the Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) algorithms in
combination with Huffman algorithm; DCT-H and DWT-H. The comparison is based on five factors:
Compression Ratio (CR), Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Structural Similarity
Index Measure (SSIM) and compression/decompression time. The experiments are conducted on five BMP gray-
scale file images. We found out that DWT-H coding is comparable to DCT-H coding in term of CR and
outperforms DCT-H in terms of MSE, PSNR and SSIM. The CR average results of the five test images for DCT-
H and DWT-H are 2.36 and 3.17, respectively. Moreover, DCT-H has the average results of MSE = 13.19,
PSNR = 37.15 and SSIM = 0.76, while WDT-H has the average results of MSE = 4.54, PSNR = 42.5 and SSIM
= 0.85. On the other hand, DCT-H outperforms DWT-H in term of execution time for compression and
decompression. DCT-H has an average compression time of 0.358s and an average decompression time of
0.122s, while WDT-H has 2.38s compression time and 2.13s decompression time.

KEYWORDS
Image compression, DCT, DWT, Huffman coding, PSNR, SSIM.

1. INTRODUCTION
In the last decades, the demand for image compression has increased, particularly after the great
development of camera devices and the proliferation of high-quality image and video exchange over
the internet [1]. The applications based on images, such as medical imaging, cameras and video-on-
demand systems contain large amounts of data to transmit. The main idea behind image compression
is to reduce the size of the image in order to minimize the storage space and increase the transmission
speed [2]. Data Compression (DC) is a technique that transforms the original data to its compact form
by the recognition and utilization of patterns existing in the data. It should be able to inverse the data
very approximately to original data [3]. DC techniques are crucially used in many real time
applications, like satellite imagery, Geographical Information Systems (GISs), graphics, Wireless
Sensor Networks (WSNs), …etc. For example, an image without compression contains 1024 pixels of
24 bit with a size of 3MB. The image needs a transmission time of 7 minutes with an ISDN line and
64 Kbit/s. If the image is compressed at 10:1 compression ratio, its size drops to 300 KB and needs
below 6 seconds transmission time [4]. The purpose of compression is to eliminate data redundancy
and irrelevancy in order to decrease the storage and transmission costs while maintaining good quality.
There are two different categories of image compression techniques; lossy and lossless techniques.
Lossy compression is a technique that transforms the original data into more efficient data and cannot
reconstruct the original data without errors. It is also called transform coding [3]. Lossless
compression is a technique that processes the original data without losing any information [5].
Examples of lossy techniques include Discrete Hartley Transform (DHT), Discrete Fourier Transform
(DFT), Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), …etc. [4]. Lossless
compression techniques include Huffman coding, LZW, arithmetic coding, …etc.
Transform coding converts the input data into another kind of representation in which the transformed
values (coefficients) are encoded by compression techniques. DCT and DWT are the most widely used

A. Maghari is with Faculty of Information Technology, Islamic University of Gaza, Palestine. Email: [email protected]
74
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

transform coding techniques which have the ability to compress data using a smaller number of
coefficients [3]. The major drawbacks, however, of DCT are preface of fake contouring effects and
blocking artifacts at higher compression. Similarly, DWT requires huge computational resources.
Mostly, transform-based image compression algorithms follow three step processes:
transformation, quantization and entropy encoding. The quantization may be scalar or vector
and the quantized transformed coefficients are entropy-coded [6]. On the other hand, Huffman
coding as a lossless algorithm has a good compression ratio and a fast compression time. A
previous study showed that Huffman coding is better than RLE and Delta encoding
techniques in term of compression time [7]. Huffman replaces fixed-length code words with
variable-length code words, where low-frequency symbols are expressed with longer
encodings and high-frequency symbols are expressed with shorter encodings. In this paper,
we have analyzed and implemented DCT image compression technique combined with
Huffman coding (DCT-H) and DWT image compression technique combined with Huffman
coding (DWT-H). The two combined techniques have been evaluated using different
performance metrics.
There are many comparative studies which have been conducted to compare between
compression techniques using DCT and DWT [8], [9]. However, their experiments were
conducted on only one or two images with different sizes and did not take into account the
compression and decompression time. The main objective of this work is to empirically
compare between DWT and DCT in combination with Huffman coding in terms of five
factors: Compression Ratio (CR), Mean Square Error (MSE), Peak Signal to Noise Ratio
(PSNR), Structural Similarity Index Measure (SSIM) as well as compression and
decompression time. This study will be helpful for upcoming researchers to approximately
select and develop the required algorithms to be used in a particular situation. Our
experiments are conducted using five different images with the same size (256×256). The size
of (256×256) pixels is chosen to ease the implementation of DCT and WDT, where the image
components are divided into (8×8) blocks for DCT implementation.
The remainder of the paper is organized as follows. We present a background in Section 2.
The related work is described in Section 3. Section 4 presents the methodology used and the
experimental setup. Results and discussion are explained in Section 5. Section 6 concludes
our findings of this research.
2. BACKGROUND
2.1 Lossy and Lossless Image Compression
Lossless compression is commonly used with text files, where the input and output data are the same,
before and after the compression process, while lossy compression is commonly used with different
types of data, such as image, video and audio data. At lossy compression the input data and output
data are not the same, which means that there is a loss of some data at the compression process so that
when we perform the decompression process, we obtain a closer approximation of the input data [7].

2.2 Discrete Cosine Transform (DCT)

Discrete Cosine Transform (DCT) is one of the lossy compression approaches which are commonly
applied in photo compression as jpeg compression. DCT is very approximate to Discrete Fourier
Transform (DFT), but DCT includes a basis of cosine functions and real number co-efficients. Both
DFT and DCT transform data from a spatial domain into a frequency domain. An inverse function is
used to reconstruct the image back [10]. The basic idea of DCT is to convert a signal into basic
frequency components. The image is divided into several blocks. Then, the sum of cosine functions on
different frequencies can be mathematically used to express each block of an image. For example,
Joint Photographic Experts Group (JPEG) [11] is a well-known compression scheme based on DCT.
For DCT, the image is first transformed into an appropriate format for image compression. All the
image components are divided into (8×8) blocks. Every block is encoded using discrete cosine
 75
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

transformation, which is used to exploit the spatial correlation between the pixels [2]. The following
formulae in Equation (1) and Equation (3) describe the transformation function and its inverse.

2.2.1 Definition of DCT

Given a function f(i, j) (an M*N block of an image), the 2D DCT transforms it into a new function
F(u, v). The general definition of the transform is:

(1)

where i, u = 0, 1,…, M-1, j, v = 0, 1,…, N-1 and the constants C(u) and C(v) are defined as:

(2)

The image block dimension is defined as M=N=8.

2.2.2 2D Discrete Cosine Transform (2D DCT)

The 2D transform function is defined as follows:

(3)

where i, u, j, v = 0, 1, …, 7 and the constants C(u) and C(v) are determined by Equation (2).
2D Inverse DCT (2D IDCT):
The 2D inverse function is defined as follows:
(4)

where i, u, j, v = 0, 1, …, 7 and the constants C(u) and C(v) are determined by Equation (2).
After the transformation, most of the information is intense to a few low-frequency components. These
components are then quantized in order to decrease the number of bits required to represent the image.
The quantization step produces many zero components in the bit stream. Therefore, entropy encoding,
such as Huffman encoding, can achieve better compression.

2.3 Image Compression by Wavelet Transform

Discrete Wavelet Transform (DWT) is another popular technique for lossy compression, which is
commonly applied in photo-video compression area. The main idea of DWT is to decompose a signal
into a set of basic functions which transforms a discrete time signal into a discrete wavelet
representation. DWT is mostly based on two-dimensional discrete wavelet transform (2DDWT). It
uses one-dimensional discrete wavelet transform (1D-DWT) row-wise to get low (L) and high (H)
bands. Then, 1D-DWT can be applied column-wise to get four sub-bands, such as LL, LH, HL and
HH. Furthermore, each of these four bands can then be divided into four sub-bands. Some schemes
based on wavelet transform are discussed in [2].
On the other hand, the inverse of DWT (IDWT) should be capable of inversion at minimum
approximately to the original signal. Equation () shows the expanded and translated family of
wavelets ψ [12] which is an orthonormal basis of :

(5)

where Z represents the set of integers. The scale of translations is changed along with the overall scale
2j, so as to keep movement in the lower resolution image in proportion [12]. The simplest wavelet
transform is called Haar Wavelet Transform, which forms averages and differences of a given
sequence of numbers. Theory of Haar WT algorithm can be found in [13].
76
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

2.4 Lossless Entropy Coding Techniques

Coding is a technique that assigns binary digits to the transformed and quantized output. Variable-
length coding, also known as source coding, is used to have a less average length of bits per pixel for
the image, in which it replaces each input symbol with a specific code word. The entropy represents
the average amount of information contained per symbol in the source S. It is the average number of
bits needed to represent the symbols in the source S. The entropy H(S) of an information source with
alphabets is:
(6)

where is the probability that a symbol will occur in S, indicates the amount of
information contained in , which corresponds to the number of bits needed to encode . The most
widely used entropy coding methods in data compression literature are Huffman coding and Shannon–
Fano coding.

2.5 Huffman Coding

Huffman coding is a famous compression coding technique, which replaces fixed-length code words
with variable-length-code words, where low-frequency symbols are expressed with longer encodings
and high-frequency symbols are expressed with shorter encodings [14]. It is a type of optimal prefix
code, which is widely employed in lossless data compression. Huffman coding is uniquely decodable
and consists of two components such as constructing Huffman tree from input sequence and traversing
the tree to assign codes to characters. It is noticed that the two nodes at the same level of the tree will
have the same code lengths. Huffman coding is still popular because of its simpler implementation,
faster compression and lack of patent coverage. Several compression methods, like Deflate, JPEG,
MP3, …etc. use Huffman code as the back-end technique [3].

3. RELATED WORK
There are many studies which have been conducted on image compression techniques. Hnesh and
Demirel [15] proposed a new hybrid model that integrated DWT and DCT-SVD techniques. They
applied their model on jpeg images and compared their results to other experiments that they
performed on original DCT and DCT-SVD separately. They reported that their model results were
better than DCT and DCT-SVD in term of compression ratio.
Other efforts [16] present a new framework for compression of medical images based on compressive
sensing (CS). The researchers proposed a framework that combined the discrete cosine transform
(DCT) as well as the discrete wavelet transform (DWT) with CS. Their framework was applied on CT
and MRI medical images. Their results showed that the CS-and DWT-based compression technique
performs better than other compression techniques in terms of compression ratio and PSNR for MRI
and CT images.
Sharma and Kaur [17] proposed a hybrid model, using DWT, DCT and Huffman coding, for the
purpose of medical image compression. Their hybrid model aims to achieve higher compression rates
by first applying DWT and DCT on RGB components. Then, quantization was applied to calculate
probability index for each unique quantity in order to find out the unique binary code for each unique
symbol to encoding. Huffman coding was then applied on the quantized components. Their results
showed that their hybrid model can effectively improve the compression ratio and PSNR.
All of the above studies applied different algorithms in image compression based on several factors to
achieve a higher compression ratio. There are also some comparative studies that have been conducted
to compare compression techniques. Barbhuiya et al. [8] conducted a comparative study on image
compression using DCT (Discrete Cosine Transform) and DWT (Discrete Wavelet Transform). They
also presented a compression algorithm using DWT and Inverse DWT and applied it on two image
formats; JPEG (Joint Photographic Experts Group) and PNG (Portable Network Graphics) color
images. Their results showed that the DWT technique outperforms the DCT technique in terms of
Compression Ratio, Mean Square Error (MSE) and Peak Signal to Noise Ratio (PNSR).
 77
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

Saroya and Kaur [9] proposed a comparative study between DCT and DWT algorithms based on the
parameters mean square error (MSE) and peak signal to noise ratio (PSNR). Their experiments were
conducted on one image (Lena) without showing any detailed results related to PSNR or MSE. Most
of reviewed comparative studies did not report detailed results and did not take into account the
compression and decompression time. This work, however, compares DWT and DCT in combination
with Huffman coding in terms of four factors; Compression Ratio (CR), Mean Square Error (MSE),
Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) as well as
compression and decompression time.

4. METHODOLOGY AND EXPERIMENTAL SETUP

In this section, we present the idea of image reconstruction, the compression and decompression
models, discuss the performance metrics and then explain the experimental setup.

4.1 Image Compression

The principle of image compression algorithms is reducing the redundancy in the image data and (or)
producing a reconstructed image from the original image with the introduction of an error that is
insignificant to the intended applications. The aim here is to obtain an acceptable digital image
representation, while preserving the essential information contained in that particular dataset [18].

4.2 Compression/Decompression Model (DCT-H)

Due to our methodology, we applied lossy and lossless algorithms on some test images. Figure 1
shows the framework followed to implement our compression/decompression model for DCT-H.

Figure 1. DCT-H, compression model (left) and decompression model (right).

Regarding the compression steps shown in Figure 1 (left), the input image is first loaded. Then, the
loaded image is divided into blocks with 8*8 pixels. Forward 2D DCT function is applied on each
block to obtain the DCT coefficients which are the 64 basis functions of the 8*8 pixel blocks from the
original image. An example of 8*8 image luminance block, with a range of 8-bit values f(i, j) [0, 255],
is illustrated in Figure 2 (left) [12].

Figure 2. An 8*8 block f(i, j) from an image (left) and its corresponding DCT coefficients F(u, v)
(right).
78
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

As shown in Figure 2 (right), except the DC and the first few AC components representing low spatial
frequencies, most of the DCT coefficients F(u, v) have small magnitudes. This is because the pixel
values in this block contain few high-spatial-frequency changes. We can see that the most information
is accurately described by the first few components of the DCT coefficients. Therefore, the remaining
components can be coarsely quantized, or even set to zero, with little signal distortion.
In the third step, quantization is applied on the DCT coefficients in order to minimize the number of
output values to a high smaller set and discard the less important components. This example uses the
luminance quantization table shown in Figure 3. The quantization step leads to high numbers of zeros
repeated as shown in Figure 4. Therefore, in the last step of encoding, Huffman encoding can be
applied to achieve better compression.

Figure 3. The luminance quantization table.

Figure 4. Quantization on DCT coefficients. Figure 5. De-quantization results.

In our experiments, however, we use a uniform scalar quantizer with a scalar factor q = 20. This
means that the DCT coefficients are divided by 20, which produces similar results to the results shown
in Figure 4. See subsection “4.5 Experimental Setup”. Contrariwise, for the decompression model
steps shown in Figure 1 (right), the compressed image file is loaded and then Huffman decoding is
applied. After that, dequantization is carried out to retain the DCT coefficients as shown in Figure 5.
In the last step, inverse discrete cosine transformation (IDCT) is applied to reconstruct an
approximated image of the original image.
Figure 6 shows the reconstructed 8*8 image block after applying IDCT. The figure shows that the
reconstructed block is in close approximation of the original block, which generally cannot be visually
noticed by the human eye in the image. To illustrate the quality of DCT compression, the error e =
original 8*8 block (Figure 2, left) – reconstructed 8*8 block (Figure 6) is shown in Figure 7.

Figure 6. The reconstructed 8*8 image block. Figure 7. Error of DCT compression.
 79
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

4.3 Compression/Decompression Model (DWT-H)

Similar to DCT-H, Figure 8 portrays the framework used to implement the
compression/decompression model for Discrete Wavelet Transformation combined with Huffman
coding (DWT-H). The objective of the wavelet transform is to decompose the input signal into
components some of which can be thresholded away. Moreover, the original image can be
approximately reconstructed using these components. Two-level 2D DWT is applied on the input
image using the MATLAB function “dwt2(input, ‘haar’)” to obtain the most significant components
that occur in the upper left side of the image (LL). Figure 9 shows the two-level DWT of 256*256
image. LL band holds the approximated version of the original image and represents the general trend
of pixel values of the input image. The other components can be efficiently encoded (after
quantization) or even discarded, because they have little significant information.

Figure 8. DWT-H, compression model (left) and decompression model (right).

After one level of 2D DWT is complete, the transformed image contains four subbands; LL, HL, LH
and HH, standing for low-low, high-low and so on, as Figure 9 shows. The LL subband can be further
decomposed to yield yet another level of decomposition (two-level DWT). This process can be
continued until the desired number of decomposition levels is reached or the LL component only has a
single element (left). Figure 10 shows an example of 16*16 gray image block (up) and its one-level
2D DWT (down). The two-level 2D DWT is also shown in Figure 11. We can see that, except the LL
components, the other components have small values which can be coarsely quantized and coded with
little signal distortion. In the third step of the compression model of DWT-H shown in Figure 8 (left),
quantization is applied to the DWT components in order to minimize the number of output values.
Finally, in the last step, Huffman encoding is applied for compression purpose. Contrariwise, for the
decompression steps shown in Figure 8 (right), Huffman decoding is applied on the compressed
image. Then, dequantization is carried out to retain the decoded DWT components. In the last step,
inverse wavelet transformation (IDWT) is applied to reconstruct the image.

Figure 9. Two-level DWT transformation.

4.4 Evaluation Metrics

To evaluate the performance of both combined compression techniques; i.e., DCT-H and DWT-H,
80
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

five performance metrics have been used; Compression Ratio (CR), Mean Square Error (MSE), Peak
Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) as well as compression
time and decompression time. They are described as follows:

Figure 10. A 16*16 block from a gray image (up) and its one-level 2D DWT (down).

Figure 11. Two-level 2D DWT of the 16*16 block shown in Figure 10.

4.4.1 Compression Ratio

The Compression Ratio (CR) is defined as the ratio of the number of bytes of the original image to
that of the compressed image. It can also be described as the ratio of the size of the original image to
the size of the compressed image. The formula of CR is given in Equation () [19]. The higher the ratio
is, the better is the compression technique.
(7)

4.4.2 Peak Signal-to-Noise Ratio (PSNR)

PSNR is a measure factor for the quality of the compressed image to the original image. It determines
 81
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

the error between the original and the reconstructed image. A high PSNR value means a high-quality
image. The objective of this factor is to measure partially the human visual response to image
quality[20]. The mathematical expression of PSNR is given in Equation () [18].
(8)

where:

m,n = image size > 0, I0 = original value and Ir = compressed value.

The Mean Square Error (MSE) [19] is the cumulative squared error between the original image and
the compressed image. A low value of MSE means a higher value of PSNR. A high PSNR provides a
better image quality after the reconstruction of the image.

4.4.3 Structural Similarity Index Measure (SSIM)

SSIM is a well-known quality metric used to measure the structural distortion between two images. It
was developed by Wang et al. [21] and is recently used by many researchers [22], [23], [24] for image
quality assessment. As opposed to SSIM, some studies revealed that PSNR performs badly in
discriminating structural content in images, since various types of degradations applied to the same
image can yield the same value of PSNR [25]. Generally, the value of PSNR can be predicted from
SSIM and vice-versa [26]. SSIM is designed to compare between the original and the distorted images
(x, y) using three factors: luminance, contrast and structural factors. It is defined as:
(9)

where, , and

where are the local means, standard deviations and cross-covariance for
images x and y.
The first term l(x, y) is the luminance comparison function, which measures the closeness of the two
images’ mean luminance ( ). The second term c(x, y) is the contrast comparison function,
which measures the closeness of the contrast of the two images. Here, the contrast is measured by the
standard deviation . The third term s(x, y) is the structure comparison function, which
measures the correlation coefficient between the two images x and y. Note that is the covariance
between x and y. The positive values of the SSIM index are in the range [0-1]. A value of 0 means no
correlation between images and 1 means x = y. The parameters α, β and γ indicate the importance of
the three components, which are set to 1. C1, C2 and C3 are constants and C3 = C2/2. So, the SSIM
simplifies to:
(10)

4.5 Experimental Setup

We have used MATLAB to conduct the experiments on a CPU @ 2.50GHz with an Intel(R) Core

(TM) i5-2450M processor. There are different images used for experimentation. We choose as
examples five different BMP gray images to conduct our experiments. BMP (BitMap) is standard file
format for Microsoft Windows, which can be stored uncompressed. The five gray images have the
same size of 256*256 = 65536 bytes (66 KB). The size of 256×256 pixels is chosen to ease
implementing DCT and WDT, where the image is split into 8x8 blocks, which are then transformed
into the frequency plane using Fast Discrete Cosine Transform (FDCT). For DWT, the 2D input
image of size 256 x 256 is subjected to two-level decomposition using discrete Haar wavelet transform
function. The scalar factor (q) for uniform quantization is set to 10, which affects the quantization
steps for Huffman coding. Quantization is the main source of losing information. So, we choose a
82
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

small value for q to obtain a small MSE. For DCT, a block image size of 8×8 is used with a scalar
factor q = 20. There are no precise rules for selecting the scalar factor q values. The trade-off between
quality and CR can be controlled by the scalar factor q which will define the size of the frequency
components [18]. Different scalar values (10 and 20) were chosen for DWT and DCT during the
evaluation process to have suitable compression ratio and MSE. To determine the execution time for
compression and decompression processes, we use tic and toc MATLAB functions together in order to
measure the amount of time MATLAB takes and display the time in seconds.

5. RESULTS AND DISCUSSION

5.1 Results
Figure 12 shows the test images and their corresponding resulting compressed images using DWT-H
and DCT-H. The experimental results with the comparison factors have been arranged in Table 1.
Table 1 shows the results of the five comparison factors obtained by applying DCT and Huffman
coding. It shows the compression ratio (CR), PSNR, MSE, SSIM, compression time and
decompression time. Table 2 shows the same five factors for DWT combined with Huffman coding
(DWT-H). Regarding the performance metrics, the results show that DWT-H outperforms DCT-H in
that DWT-H produced mostly higher CR, higher PSNR, higher SSIM and lower MSE than DCT-H, as
graphically shown in Figure 13. Performance metric values obtained are as shown in Table 1 and 2.
In contrast, for the compression/decompression time, the results show that DCT-H is less time-
consuming than DWT-H. The compression/decompression time values obtained are shown graphically
in Figure 14.

Table 1. The comparison factor results of DCT-H.

Image name CR PSNR MSE SSIM Decode Time (s) Code Time (s)
Camera man 1.72 36.04 16.3 0.575 0.12 1.14
Rice 2.56 38.04 10.28 0.772 0.12 0.2
MRI 3.37 39.37 7.57 0.898 0.11 0.15
Eye 2.66 36.7 13.9 0.734 0.14 0.16
Tiger 1.5 35.6 17.9 0.822 0.12 0.14
Average results 2.36 37.15 13.19 0.76 0.122 0.358

Table 2. The comparison factor results of DWT-H.

Image name CR PSNR MSE SSIM Decode Time Code Time
Camera man 1.96 40.8 5.4 0.712 2.44 2.7
Rice 1.97 41.7 4.4 0.856 2.4 2.7
MRI 8.24 49.6 0.72 0.993 0.65 0.7
Eye 2.1 40.1 6.4 0.854 2.27 2.64
tiger 1.6 40.5 5.8 0.851 2.9 3.17
Average results 3.17 42.5 4.54 0.85 2.13 2.38

5.2 DISCUSSION
In this study and due to our comparison study, we applied DCT and DWT algorithms in combination
with Huffman algorithm. For the DCT transform, the image is split into 8x8 blocks which are then
transformed into the frequency plane using Fast Discrete Cosine Transform (FDCT). After the
transform, most of information is concentrated to a few low-frequency components. These
components are then quantized in order to reduce the number of bits needed to represent the image.
This quantization step will lower the quality of the image by reducing the precision of the components.
Regarding the DWT algorithm, as a result of applying DWT on the input image, it is divided into four
non-overlapping multi-resolution sub-bands; LL, LH, HL and HH. The sub-band LL represents the
coarse-scale DWT coefficients while the sub-bands LH, HL and HH represent the fine-scale DWT
coefficients. Then, sub-band quantization and coding are used to reduce the number of bits needed to
represent the image. This step will lower the quality of the image but not like DCT, because in DCT,
the information is concentrated on fewer components that can be affected more sensitively by
 83
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

Figure 12. The five test images (Camera man, Rice, MRI, Eye, Tiger) and the corresponding
compressed images using DCT-H and DWT-H.

Figure 13. DCT-H and DWT-H: Compression Ratio, MSE, PSNR and SSIM.
84
"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.

Figure 14: Compression and decompression time of DCT-H and DWT-H.

quantization than in DWT. Moreover, DWT avoids blocking artifacts that can happen by dividing the
input image into blocks as in DCT [9].
The results illustrate that DWT combined with Huffman (DWH-H) coding give mostly higher
compression ratio, higher PSNR and higher SSIM than DCT combined with Huffman coding (DCT-
H). At the same time, we found that compression and decompression time of DCT-H was less than in
DWT-H. Our appreciation for the appearance of these results is based on the extent of similarity in
output values, when DCT or DWT algorithms are applied on the test images. This means that when
DCT or DWT output coefficients are mostly very small, then the quantization does not cause big loss
in the information. The quantization is used to reduce the number of distinct output values to a much
smaller set to efficiently represent the source output using any coding techniques such as Huffman
coding [12]. Using a large scalar value for quantization gives high compression ratio and low-quality
image reconstruction. So, in our experiments, we used a constant small scalar value (q) for
quantization (q=10 for DWT, q=20 for DCT) in order to obtain small MSE and high-quality
reconstruction. According to our observations, we conclude that using DWT with Huffman algorithm
achieves better compression ratio, PSNR and SSIM than using DCT with Huffman algorithm. Hence,
performance of DWT combined with Huffman coding is comparatively higher than that of DCT
combined with Huffman coding. Our results are consistent with [8], where DWT algorithm performed
much better than DCT algorithm in terms of Compression Ratio and Peak Signal-to-Noise Ratio
(PNSR). On the other hand, DCT-H is less time-consuming than WDT-H.

6. CONCLUSION
In this paper, we performed a comparative study based on DCT and DWT in combination with
Huffman coding in terms of Compression Ratio (CR), Mean Square Error (MSE), Peak Signal-to-
Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) as well as compression and
decompression time. Our experiments were conducted on five different BMP file images; Camera
man, Rice, MRI, Eye and Tiger. The experimental results showed that using DWT with Huffman
algorithm (DWT-H) achieves a comparable compression ratio with higher PSNR and less MSE than
DCT with Huffman algorithm (DCT-H). At the same time, we found out that DWT-H is time-
consuming in terms of compression and decompression. The average CR metric value of the five
images was 2.36 for DCT-H and 3.17 for DWT-H. Regarding MSE, PSNR and SSIM, the average
result values of DCT-H were: MSE = 13.19, PSNR = 37.15 and SSIM = 0.76. WDT-H has the average
result values of MSE = 4.54, PSNR = 42.5 and SSIM = 0.85. On the other hand, DCT-H outperforms
DWT-H in terms of compression and decompression execution times. DCT-H has an average
execution time of 0.358s for compression and an average execution time 0.122s for decompression,
while WDT-H has 2.38s compression time and 2.13s decompression time. In future works, we will
extend our experimental study to other image compression algorithms and use different coding
techniques.
 85
Jordanian Journal of Computers and Information Technology (JJCIT), Vol. 05, No. 02, August 2019.

REFERENCES
[1] M. Gupta and A. K. Garg, "Analysis of Image Compression Algorithm Using DCT," Int. J. Eng. Res.
Appl., vol. 2, no. 1, pp. 515–521, 2012.
[2] T. Sheltami, M. Musaddiq and E. Shakshuki, "Data Compression Techniques in Wireless Sensor
Networks," Futur. Gener. Comput. Syst., vol. 64, pp. 151–162, 2016.
[3] J. Uthayakumar, T. Vengattaraman and P. Dhavachelvan, "A Survey on Data Compression Techniques:
From the Perspective of Data Quality, Coding Schemes, Data Type and Applications," J. King Saud
Univ. Inf. Sci., 2018.
[4] A. Mr, S. Subramanian and S. Mr, "Comparing PSNR of Different Image Transforms (DCT, DFT,
DWT, DHT, DTT)," 2018.
[5] Nashar Luthfi Sugara, T. W. Purboyo and A. L. Prasasti, "Comparative Analysis of Image Compression
Using Huffman and DCT Method on JPG Image," J. Eng. Appl. Sci., vol. 13, pp. 4447-4452, 2018.
[6] G. Suseela and Y. A. V. Phamila, "Energy Efficient Image Coding Techniques for Low Power Sensor
Nodes: A Review," Ain Shams Eng. J., vol. 9, no. 4, pp. 2961–2972, 2018.
[7] W. Z. Wahba and A. Y. A. Maghari, "Lossless Image Compression Techniques: Comparative Study,"
Int. Res. J. Eng. Technol., vol. 3, no. 2, pp. 1–9, 2016.
[8] A. H. M. J. I. Barbhuiya, T. A. Laskar and K. Hemachandran, "An Approach for Color Image
Compression of JPEG and PNG Images Using DCT and DWT," Proc. of the International Conference on
Computational Intelligence and Communication Networks (CICN), pp. 129–133, 2014.
[9] N. Saroya and P. Kaur, "Analysis of Image Compression Algorithm Using DCT and DWT Transforms,"
Int. J. Adv. Res. Comput. Sci. Softw. Eng., vol. 4, no. 2, 2014.
[10] T. Kumar and R. Kumar, "Medical Image Compression Using Hybrid Techniques of DWT, DCT and
Huffman Coding," Int. J. Innov. Res. Electr. Electron. Instrum. Control Eng., vol. 3, no. 2, pp. 54–60,
2015.
[11] G. K. Wallace, "The JPEG Still Picture Compression Standard," IEEE Trans. Consum. Electron., vol. 38,
no. 1, pp. xviii–xxxiv, 1992.
[12] Ze-Nian Li, Drew, Jiangchuan Liu and Mark S. Drew, Fundamentals of Multimedia, 2nd Ed., Springer
International Publishing, 2014.
[13] C. Wang, R. Xiong, H. He, Y. Zhang and W. Shen, "Comparison of Decomposition Levels for Wavelet
Transform-based Energy Management in a Plug-in Hybrid Electric Vehicle," Journal of Cleaner
Production, vol. 210, pp. 1085–1097, 2019.
[14] A. J. Maan, "Analysis and Comparison of Algorithms for Lossless Data Compression," Int. J. Inf.
Comput. Technol., vol. 3, no. 3, pp. 139–146, 2013.
[15] A. M. G. Hnesh and H. Demirel, "DWT-DCT-SVD Based Hybrid Lossy Image Compression
Technique," Proc. of the International Conference on Image Processing, Applications and Systems
(IPAS), pp. 1–5, 2016.
[16] R. Kher and Y. Patel, "Medical Image Compression Framework Based on Compressive Sensing, DCT
and DWT," Biology, Engineering and Medicine, vol. 2, no. 2, pp. 1–4, 2017.
[17] S. Sharma and S. Kaur, "Image Compression Using Hybrid of DWT, DCT and Huffman Coding," Int. J.
Sci. Emerg. Technol. with Latest Trends, vol. 5, no. 1, pp. 19–23, 2013.
[18] O. Ghorbel, "DCT & DWT Image Compression Algorithms in Wireless Sensor Networks: Comparative
Study and Performance Analysis," Int. J. Wirel. Mob. Networks, 2013.
[19] A. Katharotiya, S. Patel and M. Goyani, "Comparative Analysis between DCT and DWT Techniques of
Image Compression," J. Inf. Eng. Appl., vol. 1, no. 2, pp. 9–17, 2011.
[20] R. Monika, S. Dhanalakshmi and S. Sreejith, "Coefficient Random Permutation Based Compressed
Sensing for Medical Image Compression," Advances in Electronics, Communication and Computing,
Springer, pp. 529–536, 2018.
[21] Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli et al., "Image Quality Assessment: From Error
Visibility to Structural Similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600–612,
2004.
 ‫‪86‬‬
‫‪"A Comparative Study of DCT and DWT Image Compression Techniques Combined with Huffman Coding , A. Maghari.‬‬

‫]‪[22‬‬ ‫‪R. Reisenhofer, S. Bosse, G. Kutyniok and T. Wiegand, "A Haar Wavelet-based Perceptual Similarity‬‬
‫‪Index for Image Quality Assessment," Signal Processing: Image Communication, vol. 61, pp. 33–43,‬‬
‫‪2018.‬‬
‫]‪[23‬‬ ‫‪K. Ma, Z. Duanmu, H. Yeganeh and Z. Wang, "Multi-exposure Image Fusion by Optimizing a Structural‬‬
‫‪Similarity Index," IEEE Transactions on Comput. Imaging, vol. 4, no. 1, pp. 60–72, 2018.‬‬
‫]‪[24‬‬ ‫‪S. Pistonesi, J. Martinez, S. M. Ojeda and R. Vallejos, "Structural Similarity Metrics for Quality Image‬‬
‫‪Fusion Assessment: Algorithms," Image Process. Line, vol. 8, pp. 345–368, 2018.‬‬
‫]‪[25‬‬ ‫‪Z. Wang and A. C. Bovik, "Mean Squared Error: Love It or Leave It? A New Look at Signal Fidelity‬‬
‫‪Measures," IEEE Signal Processing Magazine, vol. 26, no. 1, pp. 98–117, 2009.‬‬
‫]‪[26‬‬ ‫‪A. Hore and D. Ziou, "Image Quality Metrics: PSNR vs. SSIM," Proc. of the 20th International‬‬
‫‪Conference on Pattern Recognition, pp. 2366–2369, 2010.‬‬

‫ملخص البحث‪:‬‬
‫لقددددد سددددد ت نيدددددط نص رقلتددددد ل يدددددون نلادددددر لط دددددب نل ت ددددد ل ق ددددد ن ددددد نلددددد ط ددددد‬
‫حتدددددبن ً ر ب لتددددد ً ً تددددد ن ً يددددد يؤ قددددد ي لتدددددؤر دددددلج نلل دددددر نل ددددد ل ادددددر ي لتدددددؤ نل دددددرج‬
‫إلدددددم رلددددد ى نل ددددد ي دددددم رقلتددددد ل اد لدددددؤ لط دددددب نل ت ددددد ل ر جل ددددد ي ددددد ن ط ددددد ر ادددددى ددددد‬
‫نلر قدددددددؤ قددددددد ص ج نيدددددددؤ ق دددددددؤ دددددددت رن تددددددد ل رقلتطدددددددى ‪ DWT DCT‬ددددددد نيدددددددط نص‬
‫ددددددؤ يرن دددددد ددددددى ددددددد‬ ‫ر تددددددب را دددددد ر نل ق ددددددؤ اددددددى دددددد نل نيددددددؤ لتددددددؤ ي ددددددم‬
‫نلضددددددون ‪ )CR‬طريدددددددن ددددددد نل ددددددد ‪ ) MSE‬أي ددددددم ددددددد ن سددددددد إلدددددددم نلضددددددد ت‬
‫نلضون‪/‬إ نلؤ نلضونر‬ ‫‪ )PSNR‬قت س ي نلطش ه نل لتر ‪)SSIM‬‬
‫ر تلددددد نللطددددد‬ ‫أج ددددد نلط ددددد ص ي دددددم دددددت ردددددر رلط دددددى إلدددددم ددددد ل ج دددددؤ نلطددددد‬
‫أ نلط تدددددددب يدددددددط نص رقلتدددددددؤ ‪ DWT‬ضددددددد ى ست ددددددده يدددددددط نص رقلتدددددددؤ ‪ DCT‬ددددددد حتدددددددث‬
‫ددددددد نلضددددددون ‪ ) CR‬طقدددددد ص ي تدددددده ات دددددد طد ددددددى طريددددددن دددددد نل دددددد أي ددددددم دددددد‬
‫قتدددددد س ي دددددد نلطشدددددد ه نل لتددددددر ر ً دددددد طدددددد نلاددددددر نل ددددددت‬ ‫ن سدددددد إلددددددم نلضدددددد ت‬
‫نل حرردددددؤ ات ددددد دددددي دددددد نلضدددددون ادددددى نل دددددد ‪36‬ر‪17 2‬ر‪ 3‬ي دددددم نلط رتددددد ل ددددد ددددد‬
‫رقلتددددددددؤ ‪ DCT‬رقلتددددددددؤ ‪ DWT‬ر يددددددددذ ي ددددددددم نلددددددددي ًدددددددد ددددددددد نللطدددددددد لطقلتددددددددؤ ‪DCT‬‬
‫ً ددددددددددددددددددددددد ى ‪19=MSE‬ر‪13‬؛ ‪15 =PSNR‬ر‪37‬؛ ‪76 =SSIM‬ر‪ 0‬ر ي دددددددددددددددددددددددم نل ددددددددددددددددددددددد‬
‫نآل دددددددددد ًدددددددددد ددددددددددد نللطدددددددددد لطقلتددددددددددؤ ‪ DWT‬ي ددددددددددم نللحددددددددددر نآلرددددددددددى ‪54 =MSE‬ر‪4‬؛‬
‫‪5 =PSNR‬ر‪42‬؛ ‪85 =SSIM‬ر‪0‬ر‬
‫ددددد رط دددددر رقلتدددددؤ ‪ DCT‬ي دددددم رقلتدددددؤ ‪ DWT‬ددددد حتدددددث ددددد رل تددددد نلضدددددون‬ ‫ددددد ج ددددد‬
‫إ نلددددددؤ نلضددددددون حتددددددث ًدددددد ددددددد نللطدددددد لطقلتددددددؤ ‪358 DCT‬ر‪ 0‬ث تددددددؤ لددددددب نلضددددددون‬
‫ددددد إ نلدددددؤ نلضدددددون‬ ‫ددددد نلضدددددون‬ ‫‪122‬ر‪ 0‬ث تدددددؤ لدددددب إ نلدددددؤ نلضدددددون ؛ ادددددى حدددددت ًددددد‬
‫‪38‬ر‪ 2‬ث تؤ ‪13‬ر‪ 2‬ث تؤ ي م نلط رت لطقلتؤ ‪DWT‬ر‬
‫‪This article is an open access article distributed under the terms and conditions of the Creative‬‬
‫‪Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).‬‬

‫‪View publication stats‬‬