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

net/publication/261461640

Lossy medical image compression using Huffman coding and singular value
decomposition

Conference Paper · April 2013

DOI: 10.1109/SIU.2013.6531592

CITATIONS READS

43 533

3 authors, including:

Gholamreza Anbarjafari Hasan Demirel

3S Holding Eastern Mediterranean University
222 PUBLICATIONS 6,094 CITATIONS 197 PUBLICATIONS 6,692 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Gholamreza Anbarjafari on 23 February 2017.

The user has requested enhancement of the downloaded file.

 Huffman Kodlama ve Tel<il Deger Ayrl�lml
l<ullanaral< Yitimli Tlbbi imge Slkl�tlrma
Lossy Medical Image Compression Using
Huffman Coding and Singular Value Decomposition
Awwal Mohammed Rufai, Gholamreza Anbarjafari Hasan Demirel
Elektrik ve Elektronik Mühendisligi Bölümü Elektrik ve Elektronik Mühendisligi Bölümü
Uluslararasl Klbns Üniversitesi Dogu Akdeniz Üniversitesi
Lefko�a, Kuzey Klbns Türk Cumhuriyeti Gazimagusa, Kuzey Klbns Türk Cumhuriyeti
(20111890, sjafari}@ciu.edu.tr Hasan.demirel@emu.edu.tr

Özetr;e-TIbbi imgeler «;ok büyük oranda veri kullamlarak proposed compression technique over the aforementioned
üretildiginden bellek maliyeti ve iletim süresi apsmdan kazamm compression technique.
eide edilebilmesi i«;in verimli imge slkl�tJrma tekniklerinin
kullamlmasml gerektirmektedir. Bu «;ah�mada Huffman Keywords -Lossy medical image compression, singular value
kodlama ve tekil deger ayn�lml (SVD) kullanarak yitimli tlbbi decomposition (SVD), Huffman coding.
imge slkl�tlrma teknigi önerilmektedir. Önerilen teknik
kapsammda giri� imgesi ilk olarak SVD kullamlarak I. INTRODUCTION
ayn�tlnlmakta ve sonrasmda imge matrisinin kertesi, daha With the enormous amount of digital images generated by
dü�ük olan özdegerler ve bu degerlere denk gelen askl ve
medical applications such as computed tomography (CT),
hizalama matris satJrlan silinerek, dü�ürülmektedir. Daha sonra
magnetic resonance Imaging (MRI), positron emission
geri «;atIlan yitimli görüntü Huffman kodlamasl kullanarak
tomography (PET), single photon emISSIOn computed
yeniden slkl�tJnlmaktadlr. Toplam slkl�tJrma oram SVD ve
tomography (SPECT), and X-Ray there is a need to employ
Huffman kodlama ile eide edilen slkl�tlrma oranlarmm
efficient image compression techniques in order to save on cost
«;arpllmasl ile eIde edilmektedir. Önerilen tekni in performensl �
. of storage and time of transmission of these images [1-4].
«;e�itli tlbbi görüntüler üzerinde test edilmi�tir. Onerilen yöntem
sonucunda eIde edilen sonu«;lar, konvansiyonel Huffman Image compression has been center of interests for
kodlamasl ve JPEG2000 sonu«;lan ile kar�Ila�tlfllml�tJr. Nicel ve
researcher over half a century [5, 6]. Image compression deals
görsei sonu«;lar önerilen slkl�tJrma tekniginin kar�Ila�tJnlan
with reducing the number of bits needed to represent an image
konvansiyonel yöntemlere göre daha üstün bir performans
by removing redundant data. Psychovisual redundancy takes
ortaya koydugunu göstermektedir.
advantage of the fact that the human eyes ignore some data [7],
coding redundancy uses code-words to represent the statistics
Anahtar Kelimeier - Yitimli tlbbi imge Slklfitlrma; Tekil Deger
Ayrtfi1ntl (TDA); Huffman kodlama.
of the original data while inter-pixel redundancy explores the
fact that some pixels in an image have the same or almost the
same value [8]. Image compression is broadly c1assified into
two categories, namely lossless and lossy, depending on
Abstract-Medical images generate enormous amounts of whether the original image can be recovered from the
data and therefore, efficient image compression techniques need
compressed image [8]. In lossless techniques, the original
to be employed in order to save on cost and time of storage and
image can be recovered perfectly from the compressed image.
transmission respectively. In this research work, we propose a
In lossy teclmiques, the original image cannot be recovered
new lossy compression technique by using singular value
from the compressed image as some noise is deliberately
decomposition (SVD) followed by Huffman coding. In the
proposed technique firstly the image is decomposed by using
introduced to encode the image.
SVD and then the rank is being reduced by ignoring some of the JPEG2000 is a high performance image compression
lower singular values as weil as rows of hanger and aligner technique developed by the Joint Photographic Experts Group
matrices. Then the reconstructed lossy image is being compressed
commitlee. JPEG 2000 is based on the discrete wavelet
again by using Huffman coding. The compression ratio is
transform and also uses 'tiling' which refers to the partitioning
obtained by multiplication of the compression ratio achieved by
of the original image into rectangular non-overlapping blocks
using SVD with the compression ratio achieved by using
(tiles), which are compressed independently, as though they
Huffman coding. The proposed technique is tested on several
were entirely distinct images [9, 10]. Tiling reduces memory
medical images. The obtained results were also compared with
those of conventional Huffman coding and JPEG2000. The
requirements and also improves the compression performance.
quantitative and visual results are showing the superiority of the
However JPEG may not be suitable for all applications as it
involves many complicated procedures.
 Simpler algorithms like the Huffinan coding and the
As L has less row and column with respect to I thus some
singular value decomposition (SVD) can also be used for
column of U and rows of V need to be reduced in order to be
image compression [11-13]. The Huffinan algorithm, a lossless
able to conduct the matrix muItiplications for reconstructing
compression technique, was developed by David Huffman in
1950 [14, 15]. Huffinan coding is based on the frequency of the image, as shown in eq. (3).
occurrence of symbols and it uses shorter binary symbols to
represent code-words which appear frequently. Huffinan Umxm= UmXP[ Vmx ( m-p ) ] (3)

coding is widely used in many algorithms such as JPEG,

MPEG, and GZIP. The SVD on the other hand is another [
Vnxn= Vnxq Vnx ( n-q ) ]
simple algorithm which uses a smaller rank to approximate the Hence the reconstructed matrix can be obtained by
matrix representing an image [16, 17].
(4)
In this paper, a new medical image compression technique
is proposed in which it uses SVD and Huffman coding. The
Amxn =Umxp�pxq(Vnxqr
input medical image is fist compressed using the SVD Because the singular matrix has sorted singular values (in
technique which is actually a quantization technique. Then descending order) by using the physcovisual concept, ignoring
Huffinan coding is carried out to further compression of the low singular value will not significantly reduce the visual
image. The SVD and Huffman image compression techniques quality of the image.
are discussed in the next section. In our proposed technique,
B. HujJman Coding
the SVD and Huffman methods were combined to make a
lossy image compression technique. The proposed image The Huffman coding is a lossless image compression
compression technique has been tested on several medical technique. lt uses 'variable length coding', which means that
images and has been compared with the JPEG2000 symbols in the data are converted to binary symbols based on
compression technique. The quantitative experimental resuIts how often each symbol occurs. The character with the highest
based on PSNR show that the proposed technique overcomes probability is represented by a shorter binary symbol. The
the state-of-the-art techniques. This paper is organized as coding and decoding follows a certain pattern using trees built
follows: In the second section, we discuss the overview of the from bottom up. The two characters with the least probability
methodologies used in our proposed image compression are first used as leaves to make a binary tree. The tree replaces
technique in which SVD and Huffinan coding are studied in the two symbols and is given a probability equal to the sum of
more details. Then in the third section the proposed hybrid probabilities of its leaves. This procedure is repeated until a
lossy image compression is introduced and discussed in details. tree is formed with all the input symbols as its leaves. The
The experimental resuIts and discussions are placed in section paths from the root to the tree are then labeled with Os and ls.
four followed by conclusion in the last section. A binary code-word is then given to each input symbol [15, 18]
for compression.

TTT. THE PROPOSED LOSSY IMAGE COMPRESSTON

11. OVERVLEW OF SVD AND HUFFMAN CODlNG TECHNIQUE
A. Singular Value Decomposition As it was mentioned in the previous section, the SVD based
Animage is actually a matrix of numbers whose elements compression is lossy due to the nature of the process. However
are the intensity value of corresponding pixels of the image. the qualitative loss is not psychovisually noticeable up to some
Singular value decomposition is used in order to decompose a point. The proposed compression technique is benefiting from
given matrix into three matrices known as, U, I, and V in cascading SVD based compression followed by Huffman based
which U and V are orthogonal and I is a diagonal matrix compression. The compression ratio of SVD based
containing the sorted singular values of the input matrix in compression technique is obtained by calculating the required
descending order [13]. Eq. (1) is showing the size of the U, I, number of bits to represent U, 2:, and V over the required
and V matrices for a given mxn input matrix. number of bit to represent U
mx p, L pxq , and V
qxn . Then
(1) the compression ratio is obtained by multiplication of the
compression ratio of the former technique by that of the latter
The number of non-zero elements on the diagonal of I one. First the image is being compressed by using SVD as
determines the rank of the input matrix. Compression is done shown in eq. (4). The reconstructed image by using eq. (4) is
by using a smaller rank I obtained by eliminating small compressed by using Huffman coding. The block diagram of
singular values (a;) to approximate the original matrix. the proposed compression technique is shown in Fig. 1.
Mathematically it can be describe as folIows:
In the next section the experimental resuIts are reported.
2: 0 (2) The results are clearly shows the superiority of the proposed
and q :s; n
mxn =
["'f.PXq ]
0
.. p :s; m
lossy image compression technique over JPEG2000.
.
 SYD eompressiol1 teehnique

Original Image
Compressed Bit
Stream

Compressed Bit Loss)'

Stream
Uneompressed
Image

Fig t: Block diagram of the proposed medical image compression technique

IV. EXPERIMENTAL RESUL TS AND DlSCUSSION mr 2331 38.98 35.07

mr 2371 65.55 39.49
As it was mentioned in the introduction, the proposed lossy
image compression was tested on various medical images [19]. Table 3 : PSNR va ues In dB for 60 1 eompreSSlOn
Table 1 is showing the quantitative comparison between the Image! Technique Proposed Technique JPEG2000
proposed techniques and JPEG2000 by use of PSNR for er 4507 63.79 42.97
compression ratio of 20: I. In order to see the performance of er siem 14 02 48.41 38.71
ct 3071 45.62 37.44
the proposed image compression technique with different
mr 2331 34.66 32.76
compression ratios Table 2 and 3 are with compression ratio of
mr 2371 56.34 37.30
40: 1 and 80: 1 respectively are prepared. All the images used
are 512x512 with 8-bit grey-scale representation. Table 4: PSNR values in dB for 80: I compression
Image! Technique Proposed Technique JPEG2000
Table 1: PSNR va1ues In dB for 20 1 compresslOn
er 4507 59.25 41.79
Image! Technique Proposed Technique JPEG2000
er siem 14 02 45.6 37.37
er 4507 Inf 47.10
et 3071 42.07 36.25
er siem 14 02 inf 42.98
mr 2331 31.97 30.99
ct 3071 73.87 41.75
mr 2371 51.09 35.05
mr 2331 50.13 39.07
mr 2371 inf 42.51
Fig. 2 shows the comparison of the proposed medical
Table 2: PSNR va ues In dB for 40 1 compreSSlOn image compression technique with JPEG2000 by showing
Image! Technique Proposed Technique JPEG2000 different PSNR value at different compression ratios. Fig. 3
er 4507 74.91 44.55 shows the visual comparisons of the cr_2371 image at 40:1
er siem 14 02 53.32 40.83 compresslOn.
ct 3071 51.61 39.37

gO
70 ... Proposed Technique
-JPEG2000
�60
01
� 50
40
3�·'c-0 --�40c------;'6";:-( ) --CgCCO-----;Cl0-;0c--;-1;;20--;-01 4(";:-) --;-;\60 3�0 40 60 COlll80pression100Ratio 120 140 \60
Compression Ratio
(a) (b)
80,----�-�-�-�-�---___, 60-
... Proposcd Techlliquc
5( -JPEG2000
�
01 4(
[:;:;
"'-
30
J�[o==4:0==60=Comprcssion
=80==10Ratio: 0=: 1�20::�::: �140::;::!
::: 160 2�0 40 60 Compressioll
80 100Ralio 120 140 160
(e) (d)
 ... Proposed TeClmiqUel -
"'JPEG2000

.0 jC':'
Compression Ratio
(e)
Fig. 2: Graph of PSNR and eompression ratio eomparing the proposed teehnique with JPEG2000 for (a) er_4507, (b) er_siem_14_02, (e) et_3071, (d) mr_2331
and (e) mr_2371 images.

w )
Fig. 3: Visual eomparison of the er_2371 image at 40: 1 eompression (a) Original er_2371 image (b) er_2371 image eompressed using the proposed teehnique (e)
er_2371 image eompressed using JPEG2000.

7. R. C. Gonzalez and R. E. Woods, Digital Image Processing,

V, CONCLUSION
Prentice Hall, 2nd Edition, 20m.
In this paper, we have proposed a new lossy medical image 8. S. Jayaraman, S. Esakkirajan, T. Veerakumar, Digital Image
compression technique by using SVD and Huffman co ding, st
Processing, Tata McGraw-Hill Education, 1 Edition, 2009.
The proposed technique was using SVD in order to neglect the 9. M. Charrier, D.S. Cruz, and M. Larsson, "JPEG2000, the next
low singular values, and then reconstruct a compressed image, millennium compression standard for still images", IEEE
This image was compressed again by using Huffman co ding. International Conference on Multimedia Computing and
The compression ratio was obtained by multiplication of the Systems, 1999, Vol. 1, pp. 131- 132.
SVD based compression ratio with the Huffman coding based 10. A. N. Skodras and T. Ebrahimi, "JPEG2000 image co ding
compression ratio. The results of the proposed technique were system theory and applications", IEEE International Symposium
also compared with JPEG2000 technique. The quantitative and on Circuits and Systems, 2006, pp. 3866 - 3869.

visual results were showing the superiority of the proposed 11. G. O. H. Katona and T. O. H. Nemetz, "Huffman codes and seIt:'
information", IEEE Transacüon on Information Theory, 1976,
compression technique over JPEG2000.
Vol. 22, pp. 337-340.
12. J. F. Yang and C. L. Lu, "Combined Techniques Of Singular
Value Decomposition and Vector Quantization For Image
VI. REFERENCES Coding", IEEE Transactions on Image Processing, 1995, Vo1.4,
1. F. Ritter, T. Boskamp, A. Homeyer, H. Laue, M. Schwier, M. pp.1141 - 1146.
F. Link, H. Peitgen, "Medical Image Analysis", IEEE Pulse, 13. M. Greenberg, Differential equations & Linear algebra, Prentice
2011, pp. 60- 70. Hall, 1sI Edition, 2001.
2. J. Tang, X. Jiao, "Research progress of digitized medical X-ray 14. H. Tanaka, "Data structure ofHutfman codes and its application
imaging system", International Conference on Computer to efticient coding and decoding", IEEE Transaction on
Science and Information Processing (CSIP), 2012, pp. 1074 - Information Theory, 1987, V(])l. 33, pp. 154- 156.
1077. 15. H. Tanaka, "Data structure ofHuffman codes and its application
3. Kumar, S.P. Singh, ; A. Mohan, A. Anand,; H.V. Singh, to efficient encoding and decoding (Corresp.)", IEEE
"Statistical modelling of wavelet coeftieients of CT scan Transaction on Information Theory, 1987, Vol. 33, pp. 154 -
image", International Conference on Emerging Trends in 156.
Electronic and Photonie Devices & Systems, 2009, pp. 162 - 16. D. KaIman, "A Singularly Valuable Decomposition", The
165. College Mahtematics Journal, 1998, Vo1.27, pp. 2 - 23.
4. R. Sivakumar, M.K. Gayathri, D. Nedumaran, "Speckle 17. L. Knochaert, B. De Backer, and D. De Zutter, "SVD
filtering of ultrasound B-Scan Images - a comparative study compression, unitary tmnsforms, and computational
between spatial and diffusion filters", IEEE Conference on Open complexity", IEEE Transaction on Signal Processing, 1999,
Systems (ICOS), 2010, pp. 80- 85. Vol. 47, pp. 2724-2729.
5. J. Wholey, "The coding of pictorial data", IRE Transaction on 18. A. J. Penrose, Extending Lossless Image Compression, William
Information Theory, 1961, Vol. 7, pp. 99-104. Gates Building, Cambridge, IUnited Kingdom, December 2001.
6. M. Boliek, "Beyond compression: a survey of functionality 19. Medical Image Data, Retrieved January 2013, from the World
derived from still image coding", IEEE Conference on Signals, Wide Web: http://sun.aei.pol:sl.pl/�rstaros/mednat/index.html
Systems and Computers, 2004, Vol. 2, pp. 1971-1974.

View publication stats