Adaptive-Quality Image Compression Algorithm

The International Journal of Multimedia & Its Applications (IJMA) Vol.8, No.3/4, August 2016
DOI : 10.5121/ijma.2016.8402 15
ADAPTIVE-QUALITY IMAGE COMPRESSION
ALGORITHM
Abdel Rahman Alzoubaidi1
, Tamer Al-Sous2
and Hussein Al-Bahadili3
1
Department of Computer Engineering, Al Balqa Applied University, Salt, Jordan
2
Faculty of Information Technology, Middle-East University, Amman, Jordan
3
Faculty of Information Technology, University of Petra, Amman, Jordan
ABSTRACT
This paper presents the description and performance evaluation of a new adaptive-quality image
compression (AQIC) algorithm. The compression ratio (C) and Peak Signal to Noise Ratio (PSNR)
achieved by the new algorithm are evaluated through a number of experiments, in which a number of
widely-used images of large and small sizes are compressed. In all experiments, C and PNSR achieved by
the new algorithm are compared against those achieved by the PNG lossless compression image formats,
JPEG lossy compression image format, and ZIP and WinRAR lossless compression tools. For all
experiments the new algorithm provides C≈1.6, which is higher than those achieved by PNG, ZIP, and
WinRAR, and lower than JPEG; and a PNSR of more than 30 dB, which is better than that achieved by
JPEG.
KEYWORDS
Image compression; lossless data compression; lossy data compression; HCDC algorithm; image quality,
compression ratio; PSNR.
1. INTRODUCTION
Data compression algorithms are developed to reduce the size of data so it requires less disk
space for storage and less bandwidth when transmitted over data communication channels [1]. In
wireless devices, data compression reduces the amount of accumulated errors and devices power
consumption due to the reduction in the amount of the exchanged data [2, 3]. There are two
fundamentally different styles of data compression can be recognized depending on the fidelity of
the decompressed data, these are: lossless and lossy. In lossless data compression, an exact copy
of the original data is reproduced after decompression; therefore, it is used whenever it is
important to have an identical copy of the original data. Examples of lossless compression
applications are the popular ZIP and WinRAR, and also lossless compression is used as a post-
processing component within lossy compression applications [4, 5].
On lossy data compression an approximate copy of the original data set is reproduced after
decompression; therefore, it can be used whenever it is not necessary to reproduce an exact copy
of the original data, such as in some image and video compression applications. Because some
information is discarded, it may achieve higher data compression ratios, depending on the type of
compressed data set, and the amount of variations that are allowed to be introduced in the

16
decompressed data set. In image compression, most lossy data compression algorithms
approximate the colors values regardless whether they are common colors or uncommon colors
within the image color set. It is clear that any approximation to the common colors could
significantly affect the decompressed image quality; therefore, in order to enhance the quality of
the compressed imageit is very important to develop new algorithms that maintain highest
possible image compression ratio while reserving the original image qualities [6].
This paper presents a detailed description of a new algorithm that scans the image data to identify
the most common colors (MCCs) and the least common colors (LCCs) within the image, and
develops a shortest possible equivalent binary code for each color that ensure exact retrieval of
MCC colors and minimal approximation to LCCs. This practically introduces minimum effect on
the compressed image quality, and the overall effect will depend on the image colors frequencies.
Therefore, we refer to this algorithm as adaptive-quality image compression (AQIC) algorithm.
In the new algorithm, the frequencies of the 8-bit colors of the image are determined and the
colors are sorted from the MCC to the LCC. The colors are then approximated by eliminating the
Least-significant-Bit (LSB) if the color set has more than 128 colors, and very slightly different
colors are merged together to have not more than 32 colors in the image approximated color set.
The colors are then re-sorted and split into two groups. The first one includes an optimized
number of colors from the MCCs, and called the most common group (MCG); while the second
one includes the remaining colors, which are usually the LCCs, and called the least common
group (LCG). Then, a list of color equivalent binary codes is derived to ensure that most of the
colors in the MCG can be exactly retrieved, and only colors in the LCG will be slightly affected
introducing minimal effect on the image quality.
The performance of the AQIC algorithm is evaluated through a number of experiments, in which
the algorithm was used to compress standard images of large and small sizes. The performance of
the new algorithm is compared against the performance of a number of compressed image
formats (PNG and JPEG) and a number of lossless compression tools (ZIP and WinRAR).
The paper is divided into seven sections. This section provides an introduction to the main theme
of the paper. The rest of the paper is organized as follows: Section 2 reviews some of the most
recent and related work. A description of the AQIC algorithm is given in Section 3. Section 4
describes the decompression procedure of the algorithm. Section 5 defines the parameters that are
used in evaluating the performance of the new algorithm, while Section 6 presents, compares, and
discusses the experimental results. Finally, in Section 7, based on the obtained results conclusions
are drawn and a number of recommendations for future work are pointed-out.
2. LITERATURE REVIEW
This section reviews some of the most researches on image compression. A comprehensive
review can be found in [7]. A novel bit-level lossless data compression algorithm based on the
error correcting Hamming codes, namely, the Hamming Codes based Data Compression (HCDC)
algorithm was developed by Al-Bahadili [8]. The HCDC algorithm has demonstrated an excellent
performance and used by many researchers for text compression [9-10] and audio data
compression applications [11-12].
Douaket. al. [13] developed a lossy image compression algorithm dedicated to color still images.
They applied the Discrete Cosine Transform (DCT) followed by an iterative phase to guarantee a

17
desired image quality. Then, to achieve the best possible compression ratio, they applied adaptive
scanning providing for each (n, n) DCT block a matching (n×n) vector containing the maximum
possible run of 0s at its end. Afterwards, they applied a systematic lossless encoder. Rahman et.
al. [14] examined the relationship between image enhancement and data compression methods
with special emphasis on image enhancement and lossy JPEG image compression. They also
looked at the impact of compression on recovering original data from enhanced image.
Singh and Kumar [15] developed an Image Dependent Color Space Transform (ID-CST),
exploiting the inter-channel redundancy optimally, which is very much suitable compression for
large class of images. The comparative performance evaluated and a significant improvement was
observed, objectively as well as subjectively over other quantifiable methods.
Telagarapu [16] analyzed the performance of a hybrid data compression scheme that uses the
DCT and Wavelet transform. They concluded that selecting proper threshold method provides
better PSNR. A novel Context-based Binary Wavelet Transform Coding (CBWTC) approach
developed in [17], which shows that the average coding of the CBWTC is superior to that of the
state-of-the-art grayscale image coders, and always outperforms the JBIG2 algorithm and other
BWT-based binary coding technique for a set of test images with different characteristics and
resolutions.
Ameer and Basir [18] described a simple plane fitting image compression scheme. The scheme
can achieve a compression ratio of >60, while maintaining acceptable image quality. The
compression ratio is further improved by optimizing its predicted model parameters to 100. The
improvement in the compression ratio came at the expense of moderate to small quality
degradations.
Li et. al. [19] improved the performance of the Vector Quantization (VQ) image compression and
achieved a relatively high compression ratio. To obtain better reconstructed images, they
developed an approach called the Transformed Vector Quantization (TVQ). A comparison of
reconstructed image quality is made between TVQ, VQ, and standard JPEG approach. Hu and
Chang [20] developed a lossless image-compression scheme, which is a two-stage scheme. The
scheme reduces the cost for Huffman coding table while achieving high compression ratio. The
scheme provides a good means for lossless image compression.
3. THE AQIC ALGORITHM
The AQIC algorithm is an adaptive-quality data compression algorithm especially develops for
standstill image compression, where the quality of the compressed image depends on the range
and frequency of colors within the original image. This section describes in details the
compression procedure of the AQIC algorithm.
In this algorithm, the data of the original image is read one byte or character at a time; where each
byte represents color value (Vi) between 0-255. Afterwards, the size of the colors set or the
number of different colors in the image (Nco) using 8-bit color depth is found. If Nco>128, then the
Least-Significant-Bit (LSB) of the color value is discarded and the remaining color bits are
shifted one place to the right, which in turn, converts the 8-bit color to 7-bit color producing Vi
between 0-127. Subsequently, the size of the colors set is changed and the new value of Nco must
be determined using the 7-bit color depth (Nco≤128).

18
The algorithm then computes the color frequency (Fi) (i=1 to Nco) by dividing the counts of color
i by the sum of counts of all colors (N) (i.e., Fi=Oi/N); where N can be calculated by:
= ∑ (1)
The colors frequencies are then sorted from the MCC to the LCC. Starting from the first MCC,
the algorithm adds counts of colors that have either their 1st, 2nd, or 3rd bit differs from the first
MCC. The new counts of the first MCC (O1) can be calculated as: O1=O1i+O1st+O2nd+O3rd, where
O1i is the initial counts for color 1; and O1st, O2nd, O3rd are counts of colors that have either their
1st, 2nd, or 3rd bits differs from that for O1.
For example, assume the color value of the first MCC is 35 (0100011), then the counts of color
value 34 (0100010), 33 (0100001), and 39 (0100111) are added to the counts of the color value
35 and the colors 34, 33, and 39 are discarded and replaced by color value 35 during the
decompression phase. So that if the initial counts of color 35 is 10, 34 is 8, 33 is 6, and 39 is 4,
then the new counts for color 35 is 28. The colors 34, 33, and 39 are called the merged colors.
The colors are then shifted up to fill the gap left-out by the merged color(s), and Ncois reduced by
the number of merged colors (e), which is varied between 0 and 3 as it is explained in Table 1
(Nco=Nco-e).
Table 1. Associate colors.
e Explanation
0 No merged color is found
1 Only one merged color is found (33 or 34 or 39)
2 Two merged colors are found (33 & 34 or 33 & 39 or 34 & 39)
3 Three merged colors are found (33 & 34 & 39)
The merging procedure continues until all possible colors are merged. By the end of this
procedure, it can be approved that the value of Nco is always less than or equal to 32 (Nco≤32).
The new colors list is then sorted in descending according to their new frequencies, and the new
Nco and the sorted color list should be stored in the compressed image header.
The next step is the main core and contribution of the AQIC algorithm, which is called the
encoding procedure, in which each 8-bit or 7-bit color value is replaced with the shortest possible
binary representation to ensure maximum compression ratio. Although, there are many algorithms
that have been developed and used to derive optimum colors equivalent codes (binary
representation), such as Huffman, adaptive Huffman, Shannon-Fano, HCDC], etc. [7-10. Here,
we develop a different, adaptive and very efficient coding procedure to ensure maximum possible
compression ratio.
In AQIC encoding procedure, the sorted colors are divided into two groups. The first group
comprises a carefully selected and optimized number of colors from the MCCs and it is called the
most common group (MCG); while the second group comprises the remaining colors, which are
usually the LCCs, and called the least common group (LCG). In order to maximize the
compression ratio, the number of colors in the MCG (GM) should be equal to 2m-1
, where m is the
number of bits required to represent the colors in MCG. The number of colors in LCG (GL) is
calculated as GL=Nco-GM, and the number of bits required to represent the colors in LCG (n) is
calculated as n=1+⌈ln(Nco-GM)/ln(2)⌉. It can also be easily realized that for a maximum
compression m should always be less than or equal to n (m≤n).

19
The colors in each group is encoded, i.e., converted to m- or n-bit binary sequence equivalent to
its sequence number in the group starting from 0 to GM-1 or GL-1 for MCG and LCG,
respectively. This is similar to the adaptive coding used in [9] but for each group separately. In
order to distinguish the MCG colors from the LCG colors during the decompression process, the
MCG colors are preceded by 0, while the LCG colors are preceded by 1, except when GM=1,
where the color is replaced by 1-bit only (“0”) or when GL=1, where the color is replaced by 1-bit
only (“1”).
Based on the above discussion and to simplify the encoding procedure, we calculate all possible
combinations of GM and GL and consequently m and n for each Nco, and the results obtained are
summarized in Table 2.
Table 2. Associate colors.
Nco GM (m) GL (n) GM (m) GL (n) GM (m) GL (n) GM (m) GL (n) GM (m) GL (n)
1 1 (1) 0(0)
2 1 (1) 1 (1)
3 1 (1) 2(2)
4 1 (1) 3(3) 2(2) 2(2)
5 1 (1) 4(3) 2(2) 3(3)
6 1 (1) 5(4) 2(2) 4(3)
7 1 (1) 6(4) 2(2) 5(4) 4(3) 3(3)
8 1 (1) 7(4) 2(2) 6(4) 4(3) 4(3)
9 1 (1) 8(4) 2(2) 7(4) 4(3) 5(4)
10 1 (1) 9(5) 2(2) 8(4) 4(3) 6(4)
11 1 (1) 10(5) 2(2) 9(5) 4(3) 7(4)
12 1 (1) 11(5) 2(2) 10(5) 4(3) 8(4)
13 1 (1) 12(5) 2(2) 11(5) 4(3) 9(5) 8(4) 5(4)
14 1 (1) 13(5) 2(2) 12(5) 4(3) 10(5) 8(4) 6(4)
15 1 (1) 14(5) 2(2) 13(5) 4(3) 11(5) 8(4) 7(4
16 1 (1) 15(5) 2(2) 14(5) 4(3) 12(5) 8(4) 8(4)
17 1 (1) 16(5) 2(2) 15(5) 4(3) 13(5) 8(4) 9(5)
18 1 (1) 17(6) 2(2) 16(5) 4(3) 14(5) 8(4) 10 (5)
19 1 (1) 18(6) 2(2) 17(6) 4(3) 15(5) 8(4) 11(5)
20 1 (1) 19(6) 2(2) 18(6) 4(3) 16(5) 8(4) 12(5)
21 1 (1) 20(6) 2(2) 19(6) 4(3) 17(6) 8(4) 13(5)
22 1 (1) 21(6) 2(2) 20(6) 4(3) 18(6) 8(4) 14(5)
23 1 (1) 22(6) 2(2) 21(6) 4(3) 19(6) 8(4) 15(5)
24 1 (1) 23(6) 2(2) 22(6) 4(3) 20(6) 8(4) 16(5)
25 1 (1) 24(6) 2(2) 23(6) 4(3) 21(6) 8(4) 17(6)
26 1 (1) 25(6) 2(2) 24(6) 4(3) 22(6) 8(4) 18(6)
27 1 (1) 26(6) 2(2) 25(6) 4(3) 23(6) 8(4) 19(6)
28 1 (1) 27(6) 2(2) 26(6) 4(3) 24(6) 8(4) 20(6)
29 1 (1) 28(6) 2(2) 27(6) 4(3) 25(6) 8(4) 21(6)
30 1 (1) 29(6) 2(2) 28(6) 4(3) 26(6) 8(4) 22(6)
31 1 (1) 30(6) 2(2) 29(6) 4(3) 27(6) 8(4) 23(6)
32 1 (1) 31(6) 2(2) 30(6) 4(3) 28(6) 8(4) 24(6) 16(5) 16(5)
It can be seen from Table 2 that when Nco>3 there are more than one possible combination for GM
and GL and subsequently m and n. For example, for Nco=12, the possible combination for GM and
GL are 1 and 11, 2 and 10, and 4 and 8. Since, we have the values of m, n, GM, GL, and all values
of Oi (i=1 to Nco); the length of the compressed binary sequence (Sb) for each combination can be
calculated using the following general equation:

20
= ∑ + ∑ (2)
The combination of GM and GL (m and n) that will be used to encode the colors is the one that
provides minimum Sb, which subsequently provides the maximum compression ratio. The
optimum combination will depends on the individual color count (Oi) (i=1 to Nco). In this way, we
can be sure that the optimum coding rate is always selected.
After selecting the optimum combination for GM and GL (m and n), the algorithm starts
constructing the colors equivalent binary codes depending on the selected m and n. For example,
Table 3 presents the equivalent binary codes of the sorted color list of 12 colors (Nco=12) for the
three possible combinations for m and n mentioned above.
The algorithm, then start construction the compressed binary sequence by marching through the
image starting from the first color value up to the last one and replacing each color with its
equivalent binary code.
Table 3. Sample of colors equivalent binary codes.
i
Color Equivalent Binary Code
m=1, n=11 m=2, n=10 m=4, n=8
1 0 00 000
2 10000 01 001
3 10001 10000 010
4 10010 10001 011
5 10011 10010 1000
6 10100 10011 1001
7 10101 10100 1010
8 10110 10101 1011
9 10111 10110 1100
10 11000 10111 1101
11 11001 11000 1110
12 11010 11001 1111
After the construction of the compressed binary sequence, the algorithm enters the last stage,
which includes: conversion of the compressed binary sequence to 8-bit character string,
construction of the compressed image header (which should include all information necessary for
the decompression process), appending the compressed string to the image header, creation of the
compressed image file, and save the combined image header and compressed string into the
compressed image file. In this case, the size of the compressed image file is calculated as:
Sc=H+⌈Sb/8⌉, where Sc is the size of the compressed image in Bytes, H is the length of the
compressed image header, and Sb is the length of the compressed binary sequence. Figure 1
outlines the compression procedure of the AQIC algorithm.
Read image data
Calculate the number of colors (Nco)
If (Nco>128)
Convert 8-bit color to 7-bit color
Re-calculate the number of colors (Nco)
End If
If (Nco=1)
Set m=1 and n=0
Construct the colors equivalent binary codes by setting the color equivalent binary code to “0”
Else If (Nco=2)
Calculate the occurrence (Oi) or counts of each color
Calculate the sum of counts of all colors (N)

21
Calculate the colors frequencies (Fi=Oi/N)
Sort the colors according to Fi from the MCC to the LCC
Set m=1 and n=1
Construct the colors equivalent binary codes by setting the MCC to “0” and the LCC to “1”
Else
Calculate the occurrence (Oi) or counts of each color
Calculate the sum of counts of all colors (N)
Calculate the colors frequencies (Fi=Oi/N)
Sort the colors according to Fi from the MCC to the LCC
Perform merging procedure
Re-calculate new number of colors (Nco)
Re-calculate the new colors frequencies (Fi=Oi/N)
Re-sort the colors according to the Fi from the MCC to the LCC
Calculate the optimum GM and GL (m and n)
Construct the colors equivalent binary codes
End If
Read image data and replace each color value with its equivalent m or n bit binary code.
Convert the compressed binary sequence to 8-bit character string.
Construct the compressed image header containing all information required during decompression.
Append the compressed image string to the image header.
Create a compressed image file.
Save the combined image header and the compressed image string into the compressed image file.
Figure 1.The compression procedure of the AQIC algorithm.
4. DECOMPRESSION PROCEDURE OF THE AQIC ALGORITHM
The decompression procedure of the AQIC algorithm is very simple and straightforward and it is
accomplished much faster than the compression process, therefore, the algorithm behaves as an
asymmetric compression algorithm due to the difference between the compression and the
decompression processing times.
The decompression algorithm of the AQIC algorithm can be divided into two main phases. In the
first phase, the algorithm reads in the header data and prepares the list of the colors values. In
particular, it reads the values of GM and GL and computes m, n, and Nco. Then it reads the sorted
colors values from Vi (i=1 to Nco).
Finally, in this phase, the algorithm constructs the list of the colors equivalent binary codes. In the
second phase, which is the core of the decompression procedure, the algorithm reads in the
compressed image data and converts every compressed color binary code to its equivalent
uncompressed color value to re-construct the decompressed image. Figure 2 outlines the
procedure for the decompression procedure of the AQIC algorithm.
Read-in the compressed image data
Extract the compressed image header and get values of GM and GL.
Calculate m, n and Nco=GM+GL.
Read-in the colors values (Vi for i=1 to Nco)
Extract the compressed image string.
Convert the compressed image string to binary sequence.
If (Nco=1) Then
Construct colors equivalent binary codes (MCC=“0”).
Do
Read-in 1-bit at a time.
Find the associate color from the above list and append it to the uncompressed image data.

22
Loop until end of image compressed binary sequence.
Else If (Nco=2)
Construct the colors equivalent binary codes (MCC=“0” and LCC=“1”).
Do
Read-in 1-bit at a time.
Else
Construct the colors equivalent binary codes.
Do
Read in 1-bit (B)
If (B=”0”) Then
Read in (m-1)-bit.
Else
Reads in (n-1)-bit.
End If
End If
Figure 2.Procedure for the decompression procedure of the AQIC algorithm.
5. PERFORMANCE MEASURES
The performance of the AQIC algorithms is evaluated in terms of two parameters, namely:
compression ratio (C) and Peak Signal to Noise Ratio (PSNR). C represents the ratio between the
sizes of the original image file (So) and the size of the compressed image file (Sc). It can be
calculated as [1]:
C = So/Sc (3)
The MSE is the cumulative squared error between the compressed and the original images, and it
can be calculated by [1, 7]:
=

∙
∑ ∑ , !" − $ , !"%&
'( (4)
Where I(x,y) is the original image, K(x,y) is the approximated version (which is actually the
decompressed image) and X and Y are the dimensions of the images. A lower value for MSE
means lesser error.
The PSNR is a measure of the peak error, which is most commonly used as a measure of quality
of reconstruction of lossy compression images, and it is usually expressed in terms of the
logarithmic decibel (dB) scale as follows [1, 7]:
) * = 20 ∙ -./ 0 1
23 4
√267
8 (5)
Where, MAXI is the maximum possible pixel value of the image. When the pixels are represented
using 8-bit per pixel, this is 255. More generally, when samples are represented using linear Pulse
Code Modulation (PCM) with p-bit per pixel, MAXI is 2p
−1. For 24-bit image, the definition of
PSNR is the same except MSE is the sum over all squared value differences divided by image size

and by 3. Alternately, for color images the image is converted to a different color space and
is reported against each channel of that color space.
A higher value of PSNR is good because it means that the ratio of signal
the signal is the original image, and the noise is the error in reconstruction. So, if a compression
scheme having a lower MSE (higher
for the PSNR in lossy image and video compression are between 30 and 50 dB, where higher is
better. Acceptable values for wireless transmission quality loss are considered to be
to 25 dB. In lossless compression, the two images are identical, and thus
PSNR is undefined [1].
6. EXPERIMENTAL RESULTS AND
A number of experiments are performed to evaluate the performance of the AQIC
These experiments use the algorithm to compress a set of widely
small sizes. In particular, we selected five images, the dimensions and sizes of these images are
given in Table 4 and the images are shown in Figure 3
Table 4.
# Image
Large images
Dimensions
(Pixel)
1 Flowers 500x362
2 CornField 512x480
3 AirPlane 512x512
4 Monarch 768x512
5 Girl 720x576
Flowers.BMP
Monarch.BMP
International Journal of Multimedia & Its Applications (IJMA) Vol.8, No.3/4, August 2016
and by 3. Alternately, for color images the image is converted to a different color space and
inst each channel of that color space.
is good because it means that the ratio of signal-to-noise is higher. Here,
(higher PSNR), it can be recognized as a better one. Typical values
in lossy image and video compression are between 30 and 50 dB, where higher is
better. Acceptable values for wireless transmission quality loss are considered to be
to 25 dB. In lossless compression, the two images are identical, and thus MSE is zero. In this case
ESULTS AND DISCUSSIONS
A number of experiments are performed to evaluate the performance of the AQIC
These experiments use the algorithm to compress a set of widely-used test images of large and
given in Table 4 and the images are shown in Figure 3.
Table 4. Dimensions and Sizes of test images.
Large images Small images
Size (Byte)
Dimensions
(Pixel)
Size (Byte)
543,054 239x240 172,854
737,334 256x240 184,374
786,486 320x213 204,534
1,179,702 300x240 216,054
1,244,214 320x231 221,814
CornField.BMP AirPlane.bmp
Monarch.BMP Girl.BMP
Figure 3. Test images.
23
and by 3. Alternately, for color images the image is converted to a different color space and PSNR
noise is higher. Here,
), it can be recognized as a better one. Typical values
in lossy image and video compression are between 30 and 50 dB, where higher is
better. Acceptable values for wireless transmission quality loss are considered to be about 20 dB
is zero. In this case
A number of experiments are performed to evaluate the performance of the AQIC algorithm.
used test images of large and
Size (Byte)
172,854
184,374
204,534
216,054
221,814
AirPlane.bmp

24
The C and PNSR of AQIC and a number of standard lossless and lossy compressed image
formats, and lossless compression tools are listed in Tables 5 and 6. It can be seen from Table 5
that AQIC achieves a C of ≈1.6. This is because the compressed colors span over two groups,
each of 16 colors because the images selected have a continuous color distribution; so that each
uncompressed 8-bit color is expressed with only 5-bit (1-bit identifying the group number and 4-
bit identifying the sequence of the color within the group). The variation from 1.6 is due to the
effect of the adding the compressed file header to the compressed image.
AQIC achieves higher C than that achieved by the lossless PNG and less than that achieved by
the lossy JPEG. This is at the cost of some reduction/improvement in the image quality, where
AQIC provides better image quality than that produced by the JPEG format and of course less
image quality in comparison with PNG. It is also clear from Table 5 that AQIC compression ratio
is higher than ZIP and competitive to WinRAR.
Table 5. Comparing C for various images.
Image PNG JPEG ZIP WinRAR AQIC
Large images
Flowers 1.089 5.949 1.226 1.603 1.600
Corn
Field
1.126 7.139 1.268 1.698 1.600
Air Plane 1.229 8.748 1.400 1.889 1.600
Monarch 1.247 9.063 1.451 2.229 1.600
Girl 1.195 8.603 1.351 2.310 1.600
Small images
Flowers 1.247 7.011 1.372 1.846 1.599
Corn
Field
1.118 5.838 1.246 1.523 1.599
Air Plane 1.064 6.701 1.186 1.774 1.599
Monarch 1.066 6.425 1.180 1.783 1.599
Girl 1.071 5.256 1.191 1.564 1.599
Table 6.Comparing PNSR for various images.
Large images
Image PNG JPEG ZIP WinRAR AQIC
Flowers ∞ 28.03 ∞ ∞ 33.26
Corn Field ∞ 30.14 ∞ ∞ 31.90
Air Plane ∞ 29.96 ∞ ∞ 32.76
Monarch ∞ 35.50 ∞ ∞ 32.60
Girl ∞ 33.13 ∞ ∞ 33.06
Small images
Flowers ∞ 30.41 ∞ ∞ 32.80
Corn Field ∞ 29.12 ∞ ∞ 32.58
Air Plane ∞ 26.74 ∞ ∞ 33.04
Monarch ∞ 30.67 ∞ ∞ 32.79
Girl ∞ 31.26 ∞ ∞ 32.30
For many images, AQIC provides standards compressed image quality and also achieves higher
image quality than JPEG, where PSNR is always above 30 dB, where higher is better; while JPEG
for some cases has <30 dB PNSR values. The PNG, ZIP and WinRAR are lossless compression;

25
therefore, they present undefined PNSR (∞). AQIC almost provides the same performance in
terms of C and PNSR for both large-size and small-size images compression, while all other
compression formats or tools provide variable and unpredicted performance
7. CONCLUSIONS
The main conclusions of this paper are: for images of continuous color distribution, the AQIC
algorithm can achieve a compression ratio of ≈1.6 as explained above. The compression ratio
achieved by AQIC is higher than that achieved by the lossless PNG and less than that achieved by
the lossy JPEG. This is at the cost of some reduction/improvement in the decompressed image
quality. In particular, the new algorithm provides higher C than ZIP and competitive performance
to WinRAR for almost all standard test images. In terms of quality, the new algorithm provides
better image quality than that produced by JPEG and less image quality in comparison with PNG.
The main recommendations for future work are to perform further investigations to cover a wider
range of image sizes and of various colors frequencies, and evaluate the compression ratio after
applying lossless compression to images compressed with the new algorithm.
REFERENCES
[1] Sayood, K. (2012). Introduction to data compression (4th Ed.). Morgan Kaufmann.
[2] Kolo, J. G., Shanmugam, S. A., Lim, D. W. G., Ang, L. M., and Seng, K. P. (2012). An adaptive lossless data
compression scheme for wireless sensor networks. Journal of Sensors, Vol. 2012, Article ID 539638, 20 pages.
doi:10.1155/2012/539638.
[3] Xu, R., Li, Z., Wang, C., & Ni, P. (2003). Impact of data compression on energy consumption of wireless-
networked handheld devices. Proceedings of the 23rd International Conference on Distributed Computing
Systems (ICDCS '03), 302-311.
[4] Kung, W.-Y., Kim, C.-S., &. Kuo C.-C. J. (2005). Packet video transmission over wireless channels with
adaptive channel rate allocation. Journal of Visual Communication and Image Representation, Vol. 16, Issue 4-5,
475-498.
[5] Rueda, L. G., &Oommen, B. J. (2006). A Fast and Efficient Nearly-Optimal Adaptive Fano Coding Scheme.
[6] Brittain, N. J., & El-Sakka, M. R. (2007). Grayscale true two-dimensional dictionary-based image compression.
Journal of Visual Communication and Image Representation, 35–44.
[7] Alsous, T. (2013). Developing a high-performance adjustable-quality data compression scheme for multimedia
messaging. M.Sc Thesis. Middle East University, Faculty of Information Technology, Amman-Jordan.
[8] Al-Bahadili, H. (2008). A novel lossless data compression scheme based on the error correcting Hamming codes.
Computers & Mathematics with Applications, Vol. 56, Issue 1, 143–150.
[9] Al-Bahadili, H., & Rababa’a, A. (2010). A bit-level text compression scheme based on the HCDC algorithm.
International Journal of Computers and Applications (IJCA), Vol. 32, Issue 3.
[10] Al-Bahadili, H., & Al-Saab, S. (2011). Development of a novel compressed index-query Web search engine
model. International Journal of Information Technology and Web Engineering (IJITWE), Vol. 6, No. 3, 39-56.
[11] Al-Zboun, F., Al-Bahadili, H., Abu Zitar, R., &Amro, I. (2011). Hamming correction code based compression
for speech linear prediction reflection coefficients. International Journal of Mobile &Adhoc Network, Vol. 1,
Issue 2, pp. 228-233.
[12] Amro, I., Abu Zitar, R., & Al-Bahadili, H. (2011). Speech compression exploiting linear prediction coefficients
codebook and hamming correction code algorithm. International Journal of Speech Technology, Vol. 14, No. 2,
65-76.
[13] Douak, F., Benzid, R., &Benoudjit, N. (2011). Color image compression algorithm based on the DCT transform
combined to an adaptive block scanning. AEU - International Journal of Electronics and Communications, Vol.
65, Issue 1, 16–26.
[14] Rahman, Z., Jobson, D. J., &Woodell, G. A. (2011). Investigating the relationship between image enhancement
and Image compression in the context of the multi-scale retinex. Journal of Visual Communication and Image
Representation, Vol. 22, Issue 3, 237–250.
[15] Singh, S. K., & Kumar, S. (2011). Novel adaptive color space transform and application to image compression.
Journal of Signal Processing: Image Communication, Vol. 26, Issue 10, 662–672.

[16] Telagarapu, P., Naveen, V. J., Prasanthi, A. L., &Santhi, G. V. (2011). Image Compression Using DCT and
Wavelet Transformations. International Journal of Signal Proces
Vol. 4, No. 3, 61-74.
[17] Pan, H., Jin, L. Z., Yuan, X. H., Xia, S. Y., & Xia, L. Z. (2010). Context
using binary wavelet transform. Journal of Image and Vision Computing, Vol. 28
[18] Ameer, S., &Basir, O. (2009). Image compression using plane fitting with inter
Image and Vision Computing, Vol. 27, Issue 4, 385
[19] Li, R. Y., Kim, J., & Al-Shamakhi, N. (2002). Image compression using tr
of Image and Vision Computing, Vol. 20, Issue 1, 37
[20] Hu, Y. C., & Chang, C. C. (2000). A new lossless compression scheme based on Huffman coding scheme for
image compression. Journal of Signal Processing: Image Com
AUTHORS
Abdel Rahman Alzoubaidi is Associate Professor at
Systems Engineering and Computer Center Director at Al
University.He studied Automations a
University of Iasi and then left for Mu’tah University where he worked as teaching
assistant.He obtained his MPhil and DPhil in data Communications and Computer
Networks in 1996.He subsequently joined the Department of Computer Engineering
at Mu’tah University as faculty where he became Associate Professor in 2004 and
served as Director of Computer Center from 1996 to 2007. He was appointed
President's Assistant for ICT.His research in
systems, Cloud Computing, eLearning,
consultant for the Ministry of Education
Jordan. He has given numerous invited talks and tutorials, and is a consultant to companies
Internet technologies.
Tamer Sous (tamer_sous@hotmail.com
from Zaytoonah Private University, Amman, Jordan in 2006. He received his M.Sc in
Computer Science for the Middle East University, Amman, Jordan in 2013. His
research interests include image processing and compression, multimedia
applications, computer architecture, wire and wireless computer networks, distributed
and cloud computing, software developm
programming.
Hussein Al-Bahadili (hbahadili@uop.edu.jo
of Petra. He received his PhD and M.Sc degrees from University of London (Queen
Mary College) in 1991 and 1988. He has published many papers in different fields of
science and engineering in numerous leading scholarly and practitioner journals, and
presented at leading world-level scholarly conferences. He published four novel
algorithms in Computer Networks, Data Compression, Network Security, and Web
Search Engine. He supervised more than thirty PhD and M.Sc theses. He edits a book
titled Simulation in Computer Network Design and Modeling: Use and Analysis,
which is published by IGI-Global. He has published more than ten chapters in prestigious books in
Information and Communication Technology. He is also a reviewer for a number of books. His research
interests include computer networks design and architecture, routing protocols opti
distributed computing, cryptography and network security, data compression, software and Web
engineering.
Telagarapu, P., Naveen, V. J., Prasanthi, A. L., &Santhi, G. V. (2011). Image Compression Using DCT and
Wavelet Transformations. International Journal of Signal Processing, Image Processing and Pattern Recognition,
Pan, H., Jin, L. Z., Yuan, X. H., Xia, S. Y., & Xia, L. Z. (2010). Context-based embedded image compression
using binary wavelet transform. Journal of Image and Vision Computing, Vol. 28, Issue 6, 991–1002.
Ameer, S., &Basir, O. (2009). Image compression using plane fitting with inter-block prediction. Journal of
Image and Vision Computing, Vol. 27, Issue 4, 385–390.
Shamakhi, N. (2002). Image compression using transformed vector quantization. Journal
of Image and Vision Computing, Vol. 20, Issue 1, 37–45.
Hu, Y. C., & Chang, C. C. (2000). A new lossless compression scheme based on Huffman coding scheme for
image compression. Journal of Signal Processing: Image Communication, Vol. 16, Issue 4, 367-372.
is Associate Professor at Department of Computer
Systems Engineering and Computer Center Director at Al-Balqa Applied
studied Automations and Computer Engineering at the Technical
niversity of Iasi and then left for Mu’tah University where he worked as teaching
e obtained his MPhil and DPhil in data Communications and Computer
Networks in 1996.He subsequently joined the Department of Computer Engineering
ity as faculty where he became Associate Professor in 2004 and
served as Director of Computer Center from 1996 to 2007. He was appointed
President's Assistant for ICT.His research interests center on improving computing
systems, Cloud Computing, eLearning, Cyber Security, Wireless and Ad-hoc networks.He worked as
consultant for the Ministry of Education. He designed and implemented many innovative ICT projects in
Jordan. He has given numerous invited talks and tutorials, and is a consultant to companies
tamer_sous@hotmail.com) holds a B.Sc degree in Computer Science
from Zaytoonah Private University, Amman, Jordan in 2006. He received his M.Sc in
for the Middle East University, Amman, Jordan in 2013. His
research interests include image processing and compression, multimedia
applications, computer architecture, wire and wireless computer networks, distributed
and cloud computing, software development, and Web and mobile applications
hbahadili@uop.edu.jo) is an associate professor at University
Petra. He received his PhD and M.Sc degrees from University of London (Queen
Mary College) in 1991 and 1988. He has published many papers in different fields of
science and engineering in numerous leading scholarly and practitioner journals, and
level scholarly conferences. He published four novel
thms in Computer Networks, Data Compression, Network Security, and Web
Search Engine. He supervised more than thirty PhD and M.Sc theses. He edits a book
titled Simulation in Computer Network Design and Modeling: Use and Analysis,
Global. He has published more than ten chapters in prestigious books in
interests include computer networks design and architecture, routing protocols optimizations, parallel and
26
Telagarapu, P., Naveen, V. J., Prasanthi, A. L., &Santhi, G. V. (2011). Image Compression Using DCT and
sing, Image Processing and Pattern Recognition,
based embedded image compression
1002.
block prediction. Journal of
ansformed vector quantization. Journal
Hu, Y. C., & Chang, C. C. (2000). A new lossless compression scheme based on Huffman coding scheme for
372.
hoc networks.He worked as ICT
He designed and implemented many innovative ICT projects in
Jordan. He has given numerous invited talks and tutorials, and is a consultant to companies involved in
Global. He has published more than ten chapters in prestigious books in
mizations, parallel and

Adaptive-Quality Image Compression Algorithm

More Related Content

What's hot (15)

Viewers also liked (20)

Similar to Adaptive-Quality Image Compression Algorithm (20)

Recently uploaded (20)

Adaptive-Quality Image Compression Algorithm