220 The International Arab Journal of Information Technology, Vol. 12, No.

3, May 2015

An Efficient ROI Encoding Based on LSK and

Fractal Image Compression
TMP Rajkumar1 and Mrityunjaya Latte2
1
Research Scholar, Anjuman Engineering College, India
2
Principal, JSS Academy of Technical Education, India

Abstract: Telemedicine is one of the emerging fields in medicine which is characterized by transmitting medical data and
images between different users. The medical images which are transmitted over the internet require huge bandwidth. Even
images of single patient are found to be very huge in size due to resolution factor and number of images per diagnosis. So,
there is an immense need for efficient compression techniques that can be used to compress these medical images. In medical
images, only some of the regions are considered to be more important than the others (e.g., tumor in brain Magnetic
Resonance Imaging (MRI)). This paper reviews the application of ROI coding in the field of telemedicine. The image coding is
done using Wavelet Transform (WT) based on Listless Speck (LSK). The Region of Interest (ROI) is obtained from user
interaction and coded with the user given resolution to get high Compression Ratio (CR). In our proposed method, instead of
decompressing all the blocks, we decompress only the similar blocks based on the index valued stored on the stack. Thus, our
proposed method efficiently compresses the medical image. The performance measure can be analyzed by using Peak Signal
to Noise Ratio (PSNR). The execution time of the proposed method will be reduced when compare to the other existing
methods. The experimental result shows that the application of ROI coding using LSK brings about high compression rate and
quality ROI.

Keywords: Image compression, ROI, LSK, fractal image compression, MRI images, iterated functions systems.

Received January 28, 2012; accepted August 26, 2012; published online June 26, 2014

1. Introduction with a error tolerance and in such system lossy

compression where more suitable. System where
Visual communication is becoming increasingly accuracy is prime factor lossy compression schemes
important with applications in several areas such as cannot be used. To achieve higher compression with
multimedia, communication, transmission and storage lower error modifier version of compression such as
of remote sensing images, education and business shape based compression is developed [28].
documents and medical images etc., [18]. In many The compression of medical images has a great
situations multiple, large images require processing in demand. The medical community has been very
a short period of time [36]. Compression is useful reluctant to adopt lossy algorithms in clinical practice.
because it helps reduce the consumption of expensive However, the diagnostic data produced by hospitals
resources, such as hard disk space or transmission has geometrically increased and a compression
bandwidth [38]. Image compression is achieved by
technique is needed that results with greater data
reducing redundancy between neighboring pixels but,
reductions and hence transmission speed. In these
preserving features such as edges and contours of the
original image [29]. Increase in the use of color images cases, a lossy compression method that preserves the
in the continuous expansion of multimedia applications diagnostic information is needed [35]. Medical images
has increased the demand for efficient techniques that used at medical facilities are now commonly
can store and transmit visual information. This demand digitalized due to corresponding advances in
has made image compression a vital factor and has information technology. Computed Tomography (CT)
increased the need for efficient algorithms that can or Magnetic Resonance Image (MRI) generates
result in high Compression Ratio (CR) with minimum digitalized signals by its own and diagnostic images
loss [1]. There are various methods of compressing from legacy devices can be digitalized by film scanner
still images and every method has three basic steps: and such [20]. Lossless compression with progressive
Transformation, quantization and encoding [19, 27]. transmission is playing a key role in telemedicine
Image compression may be lossy or lossless. applications [24, 25].
Lossless compression is preferred for archival purposes The lossy image compression techniques, which
and often for medical imaging, technical drawings, clip reproduce completely acceptable decoded images on
art, or comics. This is because lossy compression scenes such as houses or landscapes, do not fulfill the
methods, especially when used at low bit rates, strictest quality requirements needed in medical
introduce compression artifacts. Various practical applications. There is always the possibility that a
applications demands for high date rate compatibility
An Efficient ROI Encoding Based on LSK and Fractal Image Compression 221

vague detail might give a reason to suspect some image blocks can be calculated using Euclidean
critical changes in a patient’s condition. For this distance. The rest of the paper is organized as follows:
reason, the lossy techniques, which tend to give high Section 2 describes some of the recent related works.
CRs, such as 1:10-1:30, are not acceptable in medical Section 3 briefs the fractal image compression process.
image compression [21]. Image compression is The proposed part is detailed in section 4.
required to minimize the storage space and reduction Experimental results and analysis of the proposed
of transmission cost. Medical images like MRI and CT methodology are discussed in section 5. Finally,
are special images require lossless compression as a concluding remarks are provided in section 6.
minor loss can cause adverse effects. Prediction is one
of the techniques to achieve high compression. It 2. Related Works
means to estimate current data from already known Numerous researches have been proposed by
data [26]. Image communication systems for medical researchers for the medical image compression
images have bandwidth and image size constraints that process. In this section, a brief review of some
result in time-consuming transmission of important contributions from the existing literature is
uncompressed raw image data. Thus, image presented.
compression is a key factor to improve transmission Ganguly et al. [13] deals with the various aspects
speed and storage, but it risks losing relevant medical and types of medical imaging. With the growth of
information. It exploits common characteristics of computers and image technology, medical imaging has
most images that are the neighboring picture elements greatly influenced the medical field. The diagnosis of a
or pixels are highly correlated [1]. It means a typical health problem was highly dependent on the quality
still image contains a large amount of spatial and the credibility of the image analysis.
redundancy in plain areas where adjacent pixels have El-Rube et al. [11] proposed a contour let-based
almost the same values [15]. Medical imaging compression scheme for medical endoscope images.
modalities include: The proposed algorithm was compared with two well
1. CT. known transform coding algorithms; the Discrete
2. MRI. Cosine Transforms (DCT) and the Wavelet Transform
3. Ultrasonography (US). (WT).
4. X Radiographs, etc. Tamilarasi and Palanisamy [32] proposed a wavelet
based contour let image compression algorithm.
These modalities provide flexible means for viewing Recent reports on natural image compression have
anatomical cross sections and physiological states. shown superior performance of contour let transform, a
Medical images are mostly gray scale images, with new extension to the WT in two dimensions using
diagnostically important region in the middle of the Laplacian Pyramid (LP) and directional filter banks. In
image and background of the image is usually uniform the diagnosis of medical images, the significant part
dark gray [30]. The improved compression Region Of Interest (ROI) was separated out from the
performance will be accomplished by making use of rest of the image using fuzzy C-means algorithm and
clinically relevant regions as defined by physicians. then to the resultant image optimized contour let
Images taken of patients will be aligned to pre-stored transform was applied to enhance the visual quality.
image models stored in an atlas. The atlas will contain Bhat et al. [7] proposed a scheme allows achieving
models of typical classes of images. If we are trying to a CR up to approximately 40:1 with reasonable image
compress a chest x-ray image, then it will be matched quality.
with a pre-stored chest x-ray model that is stored in the Yang et al. [39] developed an information hiding
atlas [8, 41]. methodology that includes the RSA encryption
High quality compression on cardiac MR images algorithm and a DCT based hiding technique. With
using wavelet-based methods has been compared with that system, any medical image that would be
standard JPEG. Since, the last one is an unrestricted electronically transferred (i.e., emailed, faxed, etc.,)
algorithms designed for true-color realistic images, its would have the patient’s information hidden and
performance compressing gray level images can be embedded in the image outside of the ROI.
improved with alternate algorithms specifically Sumalatha et al. [31] proposed a lossless
designed for this purpose [34]. Using image compression approach based on 3D integer wavelet
compression techniques, such as JPEG, can no doubt transform, 3D SPIHT algorithm of MR images. The
reduce the physical size of image but the image quality use of lifting scheme allows to generate truly lossless
is unavoidable to degrade [1]. To obtain improved integer to integer wavelet transforms. The main
image quality, median filtering is applied as post- objective of that work was to reject the noisy
processing [22]. background and reconstructs the image portion
In our proposed method, the given medical image losslessly.
can be compressed using LSK encoder and to avoid Bhardwaj and Ali [6] attempted to describe the
compression on similar blocks of the image using algorithm for image compression using MFHWT and
fractal image compression. The self similarity of the
222 The International Arab Journal of Information Technology, Vol. 12, No. 3, May 2015

showed better results than those obtained by using any sufficiently close to the original image and to store the
other method on an average. It includes a number of parameters of W instead of the intensity values.
examples of different images to validate the utility and
significance of algorithm’s performance. 4. An Efficient Technique for Image
Gaudeau and Moureaux [14] presented a new lossy
coding scheme based on 3D WT and lattice vector Compression using LSK and Fractal
quantization for volumetric medical images. The main Image Compression Method
contribution of that work was the design of a new Compressing the medical images effectively by
codebook enclosing a multidimensional dead zone employing, Listless Speck (LSK) and fractal image
during the quantization step, which enables to better compression techniques. The primary intension of our
account correlations between neighbor vowels. research is to compress the medical image efficiently.
Furthermore, they presented an efficient rate-distortion So initially, the medical image is divided into number
model to simplify the bit allocation procedure for our of equal sized blocks. In a medical image, more
intra-band scheme. Their algorithm has been evaluated number of blocks is similar. In order to, avoid
on several CT and MRI volumes. compression on all the similar blocks; the fractal image
compression technique can be used. So, this fractal
3. Fractal Image Compression image compression technique will identify all the
The mathematical process called Fractal encoding is similar blocks in the image [3]. Then, store all the
utilized to encode a given image into a set of similar blocks and its indexes in the separate stack.
mathematical data that illustrates the fractal properties Compress the image by employing LSK method. The
of the image [16]. Fractal encoding is based on the fact entire process of the proposed method is described in
that all objects consists of information in the form of the following.
related, repeating patterns called an attractor. An image Let us considered the set of medical images D{I1,I2,
is converted into fractal code mostly by fractal …, IN}. Here I be the image of size M×N. Then, divide
encoding. The huge number of iterations required to the image into number of P×P non overlapping blocks.
determine the fractal patterns in an image makes the This can be represented as:
encoding process to have extreme computation [9, 33].
I={Ib1, Ib2, …, Ibpb} (1)
Either Iterated Function Systems (IFS) or by
Partitioned Iterated Function Systems (PIFS) are used Where, Nb represents the total number of blocks in the
to achieve FIC [37]. image. After the block division, the similar blocks in
the image can be identified by employing the concept
3.1. Iterated Functions Systems of fractal image compression scheme.
Here, we briefly present the underlying mathematical 4.1. Fractal Image Compression
principles of fractal image compression based on a
theory of IFS [4, 12]. An IFS consists of a collection of The property of self-similarity of fractal objects is used
contractive affine transformations {wi: M →M| 1≤ I ≤ m}, by fractal compression and fractal encoding [5, 17].
where wi maps a metric space M to itself. A transform w Some of the blocks obtained by dividing the color
is said to be contractive when there is a constant image into several 8×8 blocks are similar.
number s(0≤ s≤ 1) and d(w(µ1), w(µ2))≤ sd(µ1, µ2) hold for any Our proposed method, calculates the similarity of
points µ1, µ2 ∈M, where d(µ1, µ2) denotes the metric the ath block for the fractal image compression by
between µ1 and µ2.The collection of transformations comparing the distance measure of the ath block and its
n neighboring blocks. The following equation is used
defines a mapW (S ) = Umi=1w i (S ), where S ⊂ M . Note that, the
to calculate the distance measure:
map W is applied not to the set of points in M but to the
2
set of sub-sets in M. Let W be a contractive map on a Sd = (
∑ Iba - Ibb ) (2)
compact metric space of images with the bounded
intensity and then the contractive mapping fixed-point Here, Iba and Ibb; where b={1, 2, …, n} represent the
theorem asserts that there is one special image xw, current block and blocks adjacent to the current block,
called the attractor, with the following properties: respectively. This is shown in the following Figure 1.
1. W(xw)= xw.
2. xW = limn→∞W n (S 0 ) , which is independent of the choice
of an initial image S0.
Fractal image compression relies on this theorem.
Suppose the attractor of a contractive map W is the
original image to be compressed. From property 2, the Figure 1. Range blocks and domain blocks.
original image is obtained from any initial image by
applying W. The goal of fractal image compression is The flag value is set according to the threshold Dtsh,
to find the contractive map whose attractor is after the distance measure is calculated. The calculated
An Efficient ROI Encoding Based on LSK and Fractal Image Compression 223

distance Sd is compared with the threshold Dtsh as is used to store these coefficients. For ease, array val
follows: and array sign are used to refer to the magnitude part
and the sign part respectively. An array, named mark,
I Fb = 1; if S d < D tsh of length I, each coefficient consisting of 2 bits
 b
b
IF = (3)
I F = 0; otherwise represents the state table. A one-to-one correspondence
exists between val and mark.
Here, b and F represent the image block and the flag For LSK, the pre-computed maximum descendent
value of each block of the image, respectively. Both ath magnitude array, d max, has length I/14 and the
and bth blocks are said to be similar if the bth block maximum grand descendant magnitude array, g max,
yields a flag value of 1, when it is compared with the has length I/16. These arrays can be eliminated at the
ath block. Otherwise, they are said to be dissimilar. This expense of repeated searching over significant trees for
is illustrated in Figure 2: significant coefficients at the encoder.

4.2.2. State Table Markers

In order to, keep track of the set partitions, the
following markers are kept in the 2 bit per coefficient
state table mark. Each element of mark indicates
something about the corresponding element in the val
Figure 2. Flags assigned to each domain blocks. image array, if it signifies a block. Each marker and its
meaning are listed below:
So, we store the indices by identifying the blocks
• MIP: The pixel is unimportant or untested for this
similar to the ath block. In fractal image compression,
bit-plane.
ath block is the range block and the analogous similar
• MNP: The pixel is newly important, so it will not be
blocks are domain blocks. Instead of all the similar
refined for this bit-plane.
domain blocks, we use only the range block once the
• MSP: The pixel is important and will be refined in
indices of range block and its corresponding domain
this bit plane.
blocks are accumulated. The time and memory
• MS2: The block is of size 2×2, i.e., 4 elements to be
complexity is decreased by this.
skipped.
After identifying the similar blocks in an image
using Fractal image compression technique the similar
blocks and its corresponding indexes are stored in the
separate stack for decompression process. These
similar blocks are then compressed by using LSK
encoder.

4.2. Listless Speck Figure 3. State array diagram of LSK.

LSK [10] represents Listless set partitioning embedded In Figure 3, MD stands for marker D. The MD is the
block. Same block partitioning rules used by Speck are first child pixel of each descendent set and MG is the
also used by LSK. By placing markers at all initial pixel of the first grandchild in a set consisting of all
pixels for all the sub-bands, partitioning of set ‘I’ into grand descendants of its grandparent pixel, but not
three ‘S’ sets is modified in the proposed algorithm including the grand parent or the children of the grand
[29]. Though the proposed LSK algorithm employs parent. MN is the first grandchild of MD set. MN3 is a
some minimum number of bits to specify the first great grandchild of MD or MG. Lower levels of
insignificant sub-band blocks in the initial passes, the the set are labeled with MN2 markers.
required computational time and complexity are
similar to that of speck [7]. 4.2.3. Initialization
This paper describes a new image/video codec
called LSK that uses the set partitioning rules of speck Floating point transform coefficients are quantized to
and does an explicit breadth first search without using integers by performing a dyadic sub band transform on
lists. State information is kept in a fixed size array that the image, with L= 5 levels. After that, the transform is
corresponds to the array of coefficient values; with read into the linear array val. The entire two-
about four bits per coefficient to enable fast scanning dimensional transform matrix is mapped into one-
of the bit planes. dimensional array and it can be found for each (r, c) by
bit interleaving and the corresponding coefficients can
4.2.1. Storage be moved. val is scanned to find the most important
non zero bit plane, B.
In the DC band, the number of coefficients is In LSK, this procedure can be included by
determined as: employing additional pointer in the proposed
Idc=RdcCdc, where, Rdc=R2-L, Cdc=C2-L, L is the number algorithm, to skip all the elements after it. Each
of sub-band decomposition levels. A vector of length I coefficient is marked with MIP and first element of
224 The International Arab Journal of Information Technology, Vol. 12, No. 3, May 2015

every full sized block with MS* markers to initialize algorithms. Figure 4 illustrates the entire compression
the state table. process.
m ark [ i ] = M IP (4)
Original Fractal
Partitioning
m ark [ i + j ] = M S 2 (5) Medical Image Encoding

Where, j=0, 1, 2…, e and elog2(m/2L). During

Compressed
initialization, some full sized blocks are marked. To Image LSK Encoding
avoid computing the above indices, a small pre-
computed constant look up table can be used. Figure 4. Block diagram of the compression process.

4.2.4. Encoder Algorithm 4.3. Decompression Process

The main encoder algorithm performed for each bit- The decompression process is quite simple. We first
plane, b, starting with B and decremented to 0, or until decode the fractal parameters and repeatedly transform
a bit budget is met. There are three passes in the an arbitrary image by fractals.
algorithm. They are: A fractal approximation of the original image is
1. Insignificant Pixel Pass. produced by this process. Finally, we get the
2. Insignificant Set Pass. decompressed image by transforming the addition of
3. Refinement Pass. For each Bit-plane, the the fractal approximation and the difference image by
Significance Level is given by. s=2b. LSK decoding. Figure 5 illustrates the entire
decompression process.
In binary form, s has a single bit set, at a bit position b,
the test to determine whether the coefficient val [i ] is Compressed Fractal
LSK Decoding
newly significant is d=val[i] and s. The coefficient is Image Decoding
newly significant if and only if its bth bit is set. If d is
non zero, val [i] significant. Start with Insignificant
Decompressed
pixel pass, the new significant elements are identified Image
and marked as the MNP. Then, move to next element.
This can be described as if mark[i]==MIP then the Figure 5. Block diagram of the decompression process.
elements in the state array are marked as MNP i.e.
output(d = S b [val ( i )]) (6) 5. Experimental Results
In this section, we illustrate the effectiveness of the
mark[i]=MNP (7) proposed image compression scheme by means of the
Then, start the Insignificant Set pass, the block results obtained from the experimentation. The
significance can be checked in the state table mark. If proposed method was implemented in MATLAB (Mat
the block is significant means it is split up into 4 lab 7.10) and the proposed compression scheme was
children. Then, check the block significance of the evaluated using medical images. The test images used
children. If the grand descendent block is significant in the experiments include: Knee MRI and Brian MRI
we can further divide them into four. And check for the images [10]. The sample output obtained from the
significance once again. This can be described by the proposed method is as follows:
following pseudo code: Initially, after input the image from the database,
select the region of interest in the input image. This
For each j in n
can be represented in Figures 6 and 7. Both knee and
mark [i+j*22*(cnt-MS2-1) to i+j*22*(cnt-MS2-1)+ cnt-MS2-1]=MS2 brain MRI images are used for compression process.
End for
This figure also shows the selected region from the
In the refinement pass, set mark[i]=MSP and image. Then, block partitioning process is performed.
i=i+22*cnt-ms2. The same procedure followed by the The selected region from the input image is divided
encoder is also followed by the decoder but with a few into blocks of same size. This can be described in
low level changes. Figure 8.
The main reason for the increase in memory Figures 9 and 10 represent the decompression
requirement is because of this use of the state table process of the ROI selected portion of the input image.
markers. LSK uses linear indexing technique and The original image compression without ROI selection
utilizes coefficients stored in a linear array for the is displayed in Figures11and12.
coding process unlike SPIHT, Speck and other existing
algorithms which operate on a two dimensional array
of coefficients. In addition to, simplifying the parent-
child relationship in Hierarchical Trees, skipping of
child trees has also been made easy by such use of
linear array [39, 40]. Therefore, the complexity of the a) Original image.
algorithm is less compared to other existing
An Efficient ROI Encoding Based on LSK and Fractal Image Compression 225

b) Region of interest selected.

a) ROI selected part of brain MRI images.

c) ROI selected parts of the original images.

Figure 6.The sample output obtained from the ROI selection b) Decompressed images.
process in Knee MRI images. Figure 10. The sample output obtained from the proposed
compression process.

a) Original image.
a) Original Knee MRI images.

b) Region of interest selected. b) Decompressed images.

Figure 11. The sample output obtained from the compression

process without ROI selection.

c) ROI selected parts of the original images.

Figure 7. The sample output obtained from the ROI selection

process in brian MRI images. a) Original Brian MRI images

b) Decompressed images
a) Knee MRI image
Figure 12. The sample output obtained from the compression
process without ROI selection.

Comparative Analysis: In the comparative ananlysis

part, the proposed method is compared with the
b) Brian MRI image existing image compression process in terms of Peak
Figure 8. The sample output obtained from the block partitioning Signal to Noise Ratio (PSNR) measures.
process. PSNR: To evaluate the performance of the motion
estimation technique in the proposed system, the PSNR
based on the Mean Square Error (MSE) is used as a
quality measure and its value can be determined using
the following equation:

a) ROI selected part of knee MRI images.  (255) 2 

PSNR = 10log  MSE  dB (8)
 
1
( ( ) ( ))
2
MSE = ∑ fˆ x, y - f x, y (9)
MN
b) Decompressed images.
Here, MN is the total number of pixels in the image.
Figure 9. The sample output obtained from the proposed
fˆ (x, y ) is the decompressed image and f(x, y) is the
compression process.
original image. Initially, we compare the PSNR values
226 The International Arab Journal of Information Technology, Vol. 12, No. 3, May 2015

for the proposed compression scheme with ROI with EZW [2] SOFM [7] and MFHWT [6]. This can be
selection and without ROI selection. For our proposed described in Figure 15.
method, we use both Knee and Brian MRI images. The
Figures 13 and 14 show the comparison of the 50
proposed method with ROI selection and without ROI

PSNR (in dB)

selection.

42.5 [

42 0
Tow Modified RIO Propose SOFM MFHWT
41.5

41
Figure 15. PSNR comparison.

Knee mage Knee MRI image CR: Is the ratio between the number of bits required
with ROI without ROI
storing the image before compression I and the number
Figure 13. PSNR comparison of the proposed method with and of bits required to store the image after compression O.
without ROI selection process in Knee MRI images.
CR=I/O (10)
Here, I is the size of the original image and O is the
size of the decompressed image. The following table
describes the CR for the proposed method. Here, we
observe that the proposed method with ROI selection
process is highly compressed when compared to the
Brian Image Brian Image
proposed method without ROI selection process.
with ROI without ROI
Table 3. The CR comparison of the proposed method.
Figure 14. PSNR comparison of the proposed method with and
without ROI selection process in Brian MRI images. CR
Methods
Knee MRI Brian MRI
From the Figures 13 and 14, we observe that the Proposed Method with ROI Selection 2.80:1 3.435:1
proposed method gives high PSNR values in ROI Proposed Method without ROI Selection 3.014:1 4.88:1
selection method. These details can be described in
Table 1. Figure 16 represents the CR of the both knee and
brian MRI images in proposed method. From this
Table 1. PSNR comparisons of proposed method.
results, we conclude that our proposed method with
PSNR (in dB) ROI selection process gives high PSNR values and
Methods
Knee MRI Brian MRI
efficient CR when compared to the other existing
Proposed Method with ROI Election 42.0396 39.79625
Proposed Method without ROI Election 41.5789 36.3733
methods.

Then, we compared the proposed method with other

existing methods. The Table 2 represents the PSNR
values of the proposed method and the other existing
methods.
Table 2. The PSNR comparison between proposed and existing
methods. Knee MRI Brian MRI
Methods PSNR (in dB) Figure 16. CR comparison.
Two Component Compression Method 37.91
modified EZW 32.08
ROI with EZW 36.1 6. Conclusions
Proposed Method 40.91
SOFM 27.7
The application of ROI coding in the field of
MFHWT 27.1348 telemedicine has been presented in this paper. An
efficient compression scheme with fractal encoding has
The image quality of the decompressed image is been proposed in this article. The proposed encoding
better when compared to the existing methods. So that, scheme made use of the renowned wavelet based
the proposed methodology obtained better PSNR image encoding scheme LSK. The ROI coding begins
values. In our proposed method the image quality and with the selection of ROI by the user. By employing
CR can be improved based on the LSK encoder and the fractal encoding, we identify the similar blocks in
similar block identification process. the given input image. By using LSK encoder, the
From Table 2, we observed that the proposed different ROIs are encoded. The final encoded data is
method has the better PSNR values when compared to formed by integrating all the encoded ROI image data,
the existing methods such as two component and the similar, dissimilar blocks with its indexes. In
compression method [35], Modified EZW [32], ROI
An Efficient ROI Encoding Based on LSK and Fractal Image Compression 227

the decoder, each ROI is decoded using the encoded [12] Fisher Y., Fractal Image Compression, Springer,
data and the information available in the header. Using New York, 1995.
the co-ordinate details in the header the decoded ROI [13] Ganguly D., Chakraborty S., and Kim T., “A
image data are placed in their corresponding results. Cognitive Study on Medical Imaging,”
The experimental results show that the proposed International Journal of Bio-Science and Bio-
encoding scheme using LSK gives high CR and better Technology, vol. 2, no. 3, pp. 1-18, 2010.
PSNR values. [14] Gaudeau Y. and Moureaux J., “Lossy
Compression of Volumetric Medical Images with
References 3D Dead-zone Lattice Vector Quantization,”
Annals of Telecommunications, vol. 64, no. 5-6,
[1] Annadurai S. and Sundaresan M., “Wavelet pp. 359-367, 2010.
Based Color Image Compression using Vector [15] Ghrare S., Ali M., Jumari K., and Ismail M., “An
Quantization and Morphology,” in Proceedings Efficient Low Complexity Lossless Coding
of International Conference on Advances in Algorithm for Medical Images,” American
Computing, Communication and Control, New Journal of Applied Sciences, vol. 6, no. 8, pp.
York, USA pp. 391-396, 2009. 1502-1508, 2009.
[2] Babu D. and Alamelu N., “Wavelet Based [16] Gray R. and Neuhoff D., “Quantization,” IEEE
Medical Image Compression using ROI EZW,” Transaction Information Theory, vol. 44, no. 6,
International Journal of Recent Trends in 1998.
Engineering, vol. 1, no. 3, pp. 97-100, 2009. [17] Jacquin A., “Fractal Image Coding: A Review,”
[3] Balraju M., Govardhan A., Chandra N., Basha in Proceedings of IEEE, pp. 1451-1465, 1993.
S., and Srinivas P., “Loss Controllable Image [18] Kharate G, Ghatol A., and Rege P., “Image
Compression for Color Images using Block Compression using Wavelet Packet Tree,”
Based Binary Plane Technique,” International ICGST-GVIP Journal, vol. 5, no. 7, pp. 41-43,
Journal of Engineering Studies, vol. 1, no. 1, pp. 2005.
59-70, 2009. [19] Kharate G. and Patil V., “Color Image
[4] Barnsley M., Fractals Everywhere, Second Compression Based On Wavelet Packet Best
Edition, Elsevier, 1993. Tree,” International Journal of Computer
[5] Barnsley M., Ervin V., Hardin D., and Lancaster Science Issues, vol. 7, no. 2, pp. 31-35, 2010.
J., “Solution of an Inverse Problem for Fractals [20] Kil S., Lee J., Shen D., Ryu J., Lee E., Min
and Other Sets,” National Academy of Sciences H., and Hong S., “Lossless Medical Image
of the United States of America, vol. 83, no. 7, Compression using Redundancy Analysis,”
pp. 1975-1977, 1986. International Journal of Computer Science and
[6] Bhardwaj A. and Ali R., “Image Compression 50 Network Security, vol. 6, no. 1A, pp. 50-56,
using Modified Fast Haar Wavelet Transform,” 2006.
World Applied Sciences Journal, World Applied [21] Kivijarvia J., Ojala T., Kaukoranta T., Kuba A.,
Sciences Journal, World Applied Sciences Nyulb L., and Nevalainen O., “A Comparison of
Journal, vol. 7, no. 5, pp. 647-653, 2009. Lossless Compression Methods for Medical
[7] Bhat G., Baba A., Khan E., “Efficient Image Images,” Computerized Medical Imaging and
Compression Technique using Self Organizing Graphics, vol. 22, no. 4, pp 323-39, 1998.
Feature Maps,” International Journal of [22] Koh C. and Mukherjee J., “New Efficient
Engineering Science and Technology, vol. 2, no. Methods of Image Compression in Digital
12, pp. 7609-7615, 2010. Cameras with Color Filter Array,” IEEE
[8] Bhavani S., “Performance Evaluation of Transactions on Consumer Electronics, vol. 49,
Adaptive Mesh Based 3D MRI Compression no. 4, pp. 1448-1456, 2003.
using Wavelet Coding Schemes,” International [23] Lam K., Lau W., and Li Z., “The Effects on
Journal of Engineering Science and Technology, Image Classification using Image Compression
vol. 2, no 6, pp. 2354-2358, 2010. Technique,” available at: http: // www. isprs. org/
[9] Bovik Z. and Alan C., “A Universal Image proceedings/ XXXIII/ congress/ part7/ 744_
Quality Index,” IEEE Signal Processing Letters, XXXIII-part7.pdf, last visited 2000.
vol. 9, no. 3, pp. 81-84, 2002. [24] Liu D., Sun X., and Wu F., “Edge-Based in
[10] Brian MRI images from Painting and Texture Synthesis for Image
http://brainweb.bic.mni.mcgill.ca/brainweb/. Compression,” in Proceedings of IEEE
[11] El-Rube I., Fayed H., and El-Nahas S., “Medical International Conference on Multimedia and
Endoscopic Image Coding: A Comparative Expo, Beijing, pp. 1443-1446, 2007.
Study,” Canadian Journal on Image Processing [25] Palanisamy G. and Samukutti A., “Medical
and Computer Vision, vol. 1, no. 2, pp. 16-21, Image Compression using a Novel Embedded Set
2010. Partitioning Significant and Zero Block Coding,”
228 The International Arab Journal of Information Technology, Vol. 12, No. 3, May 2015

the International Arab Journal of Information Transactions on Image Processing, vol. 13, no. 4,
Technology, vol. 5, no. 2, pp 132-139, 2008. pp. 600-612, 2004.
[26] Ramesh S. and Shanmugam A., “Medical Image [38] Wu F. and Sun X., “Image Compression by
Compression using Wavelet Decomposition for Visual Pattern Vector Quantization,” in
Prediction Method,” International Journal of Proceedings of Data Compression Conference,
Computer Science and Information Security, vol. Snowbird, pp. 123-131, 2008.
7, no. 1, pp. 262-265, 2010. [39] Yang M., Trifas M., Chen L., Song L., Buenos-
[27] Ruchika M. and Singh A., “Compression of Aires D., and Elston J., “Secure Patient
Medical Images using Wavelet Transforms,” Information and Privacy in Medical Imaging,”
International Journal of Soft Computing and Journal of Systemics, Cybernetics and
Engineering, vol. 2, no. 2, pp. 339-343, 2012. Informatics, vol. 8, no. 3, pp 63-66, 2011.
[28] Satyanarayana B., Govardhan A., and Murthy H., [40] Zeybek E. and Nait-Ali A., “Improvement of
“A Novel Method of Shape Based Image JPEG2000 Lossy Compression Performances
Compression using Spectral Curvature Scaling,” using Preliminary Non-linear Filtering,”
International Journal of Advanced Engineering International Journal of Information and
and Application, vol. 1, pp. 203-206, 2010. Communication Engineering, vol. 4, no. 1, pp.
[29] Sousa C., Cavalcante A., Guilhon D., and Barros 24-29, 2008.
A., “Image Compression by Redundancy [41] Zukoski M., Boult T., and Iyriboz T., “A novel
Reduction,” in Proceedings of the 7th Approach for Medical Image Compression,”
International Conference on Independent International Journal Bioinformatics Research
Component Analysis and Signal Separation, and Applications, vol. 2, no. 1, pp. 89-103, 2006.
Springer Berlin Heidelberg, pp. 422-429, 2007.
[30] Sudha V. and Sudhakar R., “Two Dimensional TMP Rajkumar received his BE
Medical Image Compression Techniques-A degree in electronics engineering
Survey,” ICGST-GVIP Journal, vol. 11, no. 1, from SSIT Tumkur and MTech in
pp. 9-20, 2011. industrial electronics from
[31] Sumalatha R. and Subramanyam M., “Region Karnataka Regional Engineering
Based Coding of 3D Magnetic Resonance Images college (currently known as NITK)
for Telemedicine Applications,” International Surthkal. Presently he is perusing his
Journal of Computer Applications, vol. 5, no.12, PhD from JNTU Hyderabad and working as faculty in
pp. 1-3, 2010. E and C Dept Anjuman Engineering College Bhatkal.
[32] Tamilarasi M. and Palanisamy V., “An Efficient He is Fellow member of Institution of Engineers (FIE)
Embedded Coding For Medical Image India, and also Fellow member of Electronics and
Compression using Contourlet Transform,” Telecommunication Engineers (FIETE).
European Journal of Scientific Research, vol. 49,
no. 3, pp. 442-454, 2011. Mrityunjaya Latte received his BE
[33] Thakur N. and Kakde O., “Color Image degree in electrical engineering and
Compression with Modified Fractal Coding on ME from SDM College of
Spiral Architecture,” Journal Of Multimedia, vol. Engineering and Technology.
2, no. 4, pp. 55-66, 2007. Dharwad, Karnataka India. He was
[34] Vaquero J., Vilar R., Santos A., and Pozo F., awarded the PhD degree in 2004 for
“Cardiac MR Imaging Compression: Comparison his work in the area of digital Signal
between Wavelet Based and JPEG Methods,” in Processing. Presently he is working as Principal, JSS
Proceedings of Computers in Cardiology, Academy of Technical Education, Bangalore. His
Vienna, Austria, pp. 657-660, 1995. research interests include coding, image processing
[35] Vidhya K. and Shenbagadevi S., “A Two and multiresolution transforms. He received a “Best
Component Medical Image Compression PAPER” award for his paper in a National Conference
Technique,” International Journal of Recent NCSSS 2002 held at Coimbatore India MES.
Trends in Engineering, vol. 1, no. 1, pp. 591-
593, 2009
[36] Walmsley N., Skodras A., and Curtis K., “A Fast
Picture Compression Technique,” IEEE
Transactions on Consumer Electronics, vol. 40,
no. 1, pp. 11-19, 2002.
[37] Wang Z., Bovik A., Sheikh H., and Simoncelli
E., “Image Quality Assessment: From Error
Visibility to Structural Similarity,” IEEE