Squashed JPEG Image Compression via Sparse Matrix

International Journal of Computer Science & Information Technology (IJCSIT) Vol 9, No 3, June 2017
DOI:10.5121/ijcsit.2017.93012 139
SQUASHED JPEG IMAGE COMPRESSION VIA SPARSE
MATRIX
Shaista Qadir
Department of Computer Science, King Khalid University, Abha, Saudi Arabia
ABSTRACT
To store and transmit digital images in least memory space and bandwidth image compression is needed.
Image compression refers to the process of minimizing the image size by removing redundant data bits in a
manner that quality of an image should not be degrade. Hence image compression reduces quantity of the
image size without reducing its quality. In this paper it is being attempted to enhance the basic JPEG
compression by reducing image size. The proposed technique is about amendment of the conventional run
length coding for JPEG (Joint Photographic Experts Group) image compression by using the concept of
sparse matrix. In this algorithm, the redundant data has been completely eliminated and hence leaving the
quality of an image unaltered. The JPEG standard document specifies three steps: Discrete cosine
transform, Quantization followed by Entropy coding. The proposed work aims at the enhancement of the
third step which is Entropy coding.
Keywords
Entropy coding, JPEG image compression, Compaction using sparse, optimized/reduced run length
coding.
1. INTRODUCTION
Image compression is the removal of redundant data bits of digital images to reduce the actual
image size. It is the process of encoding information in fewer bits than their original
representation. It is simply a technique of applying compression on digital images to reduce the
size of actual image in order to transfer it easily in least possible memory size. Digital images
commonly contain lots of redundant data, these images need to be compressed to remove
redundancy and minimize the storage space and transport bandwidth. Instead of keeping track of
runs of redundant values as in conventional RLC, the proposed technique keeps track of the exact
location of the non-zero element in the zig-zag matrix and the value itself (concept of sparse
matrix). Further the technique keeps track of only first element in the sub-string if the non-zero
elements are stored at consecutive location and rest of the elements are assumed to be in
continues locations. As the non-zero elements are lest in the quantized matrix so we keep track of
the VALUE and its LOCATION using the zig-zag sequence used to read DTC coefficients.
The proposed modification in the JPEG image compression algorithm has been tested on various
JPEG images under consideration in matlab. Results have proved the efficiency of the proposed
algorithm for all the images used for testing.
2. RELATED WORK
Numbers of researches have been carried out to work upon the image compression and most of
these are using the concept of conventional run length coding scheme and some have Optimized
the other blocks of the compression technique [1]. Work on international standards for image
compression started in the late 1970s with the CCITT (currently ITU-T) need to standardize

140
binary image compression algorithms for Group 3facsimile communications [2],[3],[4]. some
authors have modified the entropy encoding part by modifying the run length coding for space
research program of IST [5] , Other technique [6] had modified version of image
compression/decompression algorithm using block optimization and byte compression method
(BOBC). After BOBC, it is followed by run-length encoding and its block is optimized by
varying block size. The Experimental results show that compression ratio of this algorithm is
better than the previous BOBC algorithm and JPEG compression techniques. Image quality
(PSNR) is almost the same or better as compared to that of the above mentioned compression
techniques.
Although the proposed algorithm is no exception but aims towards the better or even equal
compression rates(in worst case) as ever offered by the other algorithms by further modifying the
conventional [7] and optimized run length coding [8].
3. JPEG IMAGE COMPRESSION
JPEG Compression is an image compression algorithm, developed by Joint Photographic Expert
group which is used to curtail the file size of photographic images. Fig 1 shows the main
procedures for all encoding processes based on the DCT.
Figure 1. DCT-based encoder simplified diagram
The standard JPEG specifies the following three steps:
◦ Discrete cosine transform
◦ Quantization
◦ Entropy coding
3.1. Discrete cosine transform
Discrete Cosine Transform (DCT) exploits cosine functions, it transform a signal from spatial
representation into frequency domain [9].In the encoding process the input component’s samples
are grouped into 8×8 blocks, and each block is transformed by the DCT into a set of 64 values
referred to as DCT coefficients. One of these values is referred to as the DC coefficient and the
other 63 as the AC coefficients. DCT actually transforms image data from temporal to spatial
domain. With JPEG image compression fourier-based DCT transformation is employed which
AIMS at reduction of correlation between the pixels [10], [11]. The 8x8 2-Dimentional DCT is
F(u,v)= C u C v ∑ ∑ f x, y cos cos
……………(1)
DCT Quantizer
Compressed
image
Entropy
coder

141
For u = 0,1,……7 & v = 0,1,….7
Where C(k) = √
for k = 0
1 otherwise
*
3.2. Quantization
Each of the 64 coefficients is then quantized by simply dividing each Component in the
frequency domain by a constant for that component as shown in quantization Matrix Q(x,y) (Fig
2), and then rounding to the nearest integer. The image reformed later at the receivers end can’t
bring back and hence given the name loosy.
16 11 10 16 24 40 51 61
12 12 14 19 26 58 60 55
14 13 16 24 40 57 69 56
14 17 22 29 51 87 80 62
18 22 37 56 68 109 103 77
24 35 55 64 81 104 113 92
49 64 78 87 103 121 120 101
72 92 95 98 112 100 103 99
Figure 2. Quantization Matrix Q(X,Y)
3.3. Entropy coding
The DC coefficient and the 63 AC Coefficients are prepared for entropy encoding. The previous
quantized DC coefficient is used to predict the current quantized DC coefficient, and the
difference is encoded. The 63 quantized AC coefficients undergo no such differential encoding,
but are converted into a one-dimensional zig-zag sequence Fig 3. The quantized coefficients are
then passed to any of the entropy encoding procedures for image compression such as run length
coding, arithmetic coding or Huffman coding.
Figure 3. Zig-Zag sequence.
0 1 5 6 14 15 27 28
2 4 7 13 16 26 29 42
3 8 12 17 25 30 41 43
9 11 18 24 31 40 44 53
10 19 23 32 39 45 52 54
20 22 33 38 46 51 55 60
21 34 37 47 50 56 59 61
35 36 48 49 57 58 62 63

142
4. RUN LENGTH CODING
Run length coding is a lossless data compression technique in which runs of data are stored as a
single data value and count rather than as original run. This coding technique is very useful with
the data representation which contains numerous redundant runs. For such redundant data by
employing run length coding we can represent an image in very fewer bits. It is not useful with
files that don't have many runs as it could greatly increase the file size.
Run length is the number of consecutive zero-valued AC coefficients in the zig-zag sequence
present before the non-zero AC coefficient.[8] This method counts the number of repeated zeros
which is represented as RUN and appends the non-zero coefficient represented as LEVEL
following the sequence of zeros. When the last (63rd) AC coefficient is encountered, a special
sequence of (0,0) means End of Block is appended. When a sequence of non-zero coefficients is
countered it adds redundancy in the encoded data, as for the occurrence of consecutive non-zero
sequence the value of RUN is zero for most of the time. So the Conventional Run Length Coding
scheme encodes the redundant data, when it was meant to compress the original one.
Y
N
Y
N
Figure 4. Flow diagram for original run length encoding[3]
Run=0
coeffNo=0
CoeffNo++
Read Value at I/P=A
CoeffNo
== 64
A==0
Send(Run,Level=A)
Run++
Append EOB

143
According to the original run length coding algorithm the output (32 digit sequence) of the 8x8
image block in Fig.5 using Conventional Run Length Coding would be:
Figure 5. 8x8 Image block after Quantization Phase
Output Fig 5 using conventional RLC:
[(0,-33)(0,21)(0,-3)(0,-2)(0,-3)(0,-4)(0,-3)(1,2)(0,1)(1,1)
(1,-2)(0,-1)(0,-1)(3,-2)(11,1)(0,0)].
{32 digits after compression}
Figure 6. 8x8 image block after Quantization Phase
[(0,15)(0,9)(0,-1)(0,6)(0,-5)(0,8)(0,-1)(0,5)(0,7)(2,3)
(1,12)(0,-2)(0,4)(7,8)(6,6)(6,1)(1,7)(0,0)].
102 -33 -3 -4 -2 -1 0 0
21 -2 -3 0 -1 0 0 0
-3 0 1 0 0 0 0 0
2 0 0 0 0 0 0 0
1 0 0 1 0 0 0 0
-2 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
72 15 -5 8 12 -2 0 0
9 6 -1 0 4 0 0 0
-1 5 3 0 0 0 0 0
7 0 0 8 6 7 0 0
0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0

144
Figure 7. 8x8 image block after Quantization Phase
[(0,-13)(0,1)(0,8)(0,7)(1,10)(0,-6)(0,15)(0,2)(0,17)(9,-2)
(0,1)(15,-5)(0,1)(0,-2)(0,-8)(0,7)(0,18)(0,0)].
5. PROPOSED COMPACT CODING
The Flow Diagram for the proposed compact coding is shown in Fig 8. Instead of keeping track
of runs and Levels as in conventional run length coding the proposed coding represents the
strings of LOCATIONS (Loc as in Fig 8) and the VALUES (AcCoff As in Fig 8) of non-zero Ac
coefficients present in zig-zag matrix. This technique will help in further compression of image
size or it may yield equal sized result in worst case. The proposed technique is a slightly modified
run length coding. As conventional run length coding keeps track of both runs of zeros and ac
coefficients, but the proposed technique will track ac coefficient and its location only. If ac
coefficients are placed at consecutive locations in Quantized matrix, there is no need to store
locations for all the consecutive coefficients in that particular sub-string, only the location of first
coefficient need to be stored rest can be determined. In that case the proposed technique can help
in more compaction.
This modification allows removing the extra parameter from the run length coded messages that
utilized an extra memory space. This modification has been made after studying many image
samples and making the following observations,
• AC coefficients are placed at consecutive locations i.e., one after another.
• The coefficients are stored in sparse matrix where non-zero entries are very fewer.
• There is no need to keep track of non-zero entries.
Applying proposed compact coding to the same matrices (Figures 5,6,7) we can get far much
better results as compared to conventional run length coding, as shown below:
Output of Fig 5 using proposed RLC :
[(1+,,-33-> 21 -> -3 -> -2 -> -3 -> - 4 -> -3 ) (9+,, 2 -> 1) (12++++,, 1) (14++++, , -2 -> -1 -> -1) (20++++,,-2 ) (32++++,
,1)(0)]
{22 digits used after compression}
98 -13 0 10 0 0 0 0
1 7 -6 0 0 0 0 0
8 15 0 0 0 0 0 0
2 0 0 0 0 0 0 18
17 0 0 0 0 0 7 0
-2 0 0 0 0 -8 0 0
1 0 0 0 -2 0 0 0
0 0 -5 1 0 0 0 0

145
Output of Fig 6 using proposed RLC:
[(1+,,15-> 9 -> -1 -> 6 -> -5 -> 8 -> -1 -> 5 -> 7 ) (12++++,, 3) (14++++,, 12 -> -2 -> 4) (24++++,, 8 ) (3+, , 6) (38++++,,1 )
(40++++, ,7)(0)]
Output of Fig 7 using proposed RLC:
[(1+,,-13-> 1 -> 8 -> 7) (6+,, 10 -> -6 -> 15-> 2 -> 17) (20++++,,-2 ->1) (47++++, ,-5 -> 1 -> -2 -> -8 -> 7 -> 18
)(0)]
Figure 8. Flow diagram for proposed compact coding.
6. METHOD
In output (proposed compact coding) for Fig 5 LOCATION in the sub-strings is represented as an
element with arrow head and Ac coefficient is represented as an element without any arrow head.
Starting from left to right as shown in the output for the sub-string
(4++,,-33-> 21 -> -3 -> -2 -> -3 -> - 4 -> -3 )
At location 1+, in the zig-zag sequence matrix Ac coefficient -33 is stored. As you can see Next
element 21 in the sub-string is not any location but Ac coefficient again which simply means the

146
Ac coefficient 21 is stored in the next consecutive location (there is no 0 element in between -33
and 21) and likewise -3 is at the next consecutive location followed by -2 then -3 , -4 and -3 .
As it is clear sub string ends at Ac coefficient -3 which can be determined to be at LOCATION7+,.
In the next sub-string from the output of fig 5 i.e,
(5++,, 2 -> 1)
At LOCATION 9+, Ac coefficient 2 is stored and at the very next LOCATION ( 10++++, th
LOCATION) there is Ac coefficient 1 stored . As we don’t need to mention consecutive locations
in output so if the next element in the sub string is an element without arrowhead it’s understood
the element is an Ac Coefficient and not LOCATION.
Likewise in the next sub-string from the same output Fig 5
(46+++++,, 1)
At LOCATION 12++++, Ac coefficient 1 is stored.
The next sub-string in the output is
(47+++++, , -2 -> -1 -> -1)
At LOCATIONS 14++++,, 15++++, and 16++++, Ac coefficients -2,-1 and -1 are stored respectively but the sub
string must show the first LOCATION only as the other 2 LOCATIONS (15++++, and 16++++,) are
consecutive so no need to mention in the string, hence reducing the number of digits in the
output.
We will continue in the same manner in the next sub-strings
(69+++++,,-2) & (:6+++++, ,1)
At LOCATION 20++++, Ac coefficient -2 and at 32++++, Ac coefficient 1 is stored.
The sub-string (0) represents the end of the string.
In between the sub-strings there are Runs of 0’s present
(4++,,-33-> 21 -> -3 -> -2 -> -3 -> - 4 -> -3 ) Runs of one 0 present at LOCATION 8 (5++,, 2 -> 1) Runs
of one 0 present at LOCATION 11 (46+++++,, 1) Runs of one 0 present at LOCATION 13 (47+++++, , -2 -> -1
-> -1) Runs of three 0’s present at LOCATIONS 17, 18 & 19 (69+++++, -> -2) Runs of eleven 0’s
present from LOCATIONS 21 to 31 (:6+++++, ->1) Runs of thirty-one 0’s present from LOCATION 33
to 64 (0)].
7. RESULTS
Proposed algorithm when implemented turned out to generate better compressed image size as
compared to conventional RLC. The proposed algorithm has been tested and used upon various
images using matlab. The result of different image blocks under considration after compression
is shown in Table1. And Fig 9 Shows the Chart of results of different image blocks under
consideration.
;<<=>=?@>A =
BCDEFGHIHJKLMBCDEFNOHPKL
BCDEQGHIHJKL
…….…………… (2)

International Journal of Computer Science & Infor
S. No Algorithm Used
1 Conventional
RLC
2 Proposed RLC
3 Efficiency %
Table 1. Results
Figure 9. Chart of results of different image blocks under considration
8. CONCLUSION
In this work, we have proposed an amendment to the conventional run length coding. The
proposed technique is based on encoding runs
Based on the experimental results, it has been demonstrated that the proposed technique
outperformed the conventional run length coding. Further the proposed technique has resulted in
significant increase in the test data compression ratio for all considered test cases and improving
the compression ratio from 31% to 39% for sample images. Thus the proposed technique has the
advantage of enormously reducing image size. This is evident by the reduction in t
number of encoded runs.
AUTHORS
Shaista Qadir is Lecturer at King Khalid University , Abha
and MCA from Jamia Humdard University, N
National Institute of Electronics &
interests include image processing, big
REFERENCES
[1] C.Taskin and S.K. Sarikoz. :
and GIS Data Processing and Other Applications.
179, 2008
[2] A.M.Raid ,W.M.Khedr , M. A. El
0
5
10
15
20
25
30
35
40
Algorithm Used Digits used in Encoding Sequence
Fig 5 Fig 6 Fig 7
Conventional 32 36 36
Proposed RLC 22 25 22
Efficiency % 31.25 30.55 38.89
esults of different image blocks under consideration
9. Chart of results of different image blocks under considration
proposed technique is based on encoding runs of non-zero values only as rest of the runs are 0s.
e in the test data compression ratio for all considered test cases and improving
advantage of enormously reducing image size. This is evident by the reduction in t
King Khalid University , Abha, Kingdom Of Saudi Arabia
umdard University, New Delhi. She received her PGDCA from
National Institute of Electronics & Information Technology (NIELIT) .Her research
interests include image processing, big data, cloud computing and Data Warehouse.
. : An Overview of Image Compression Approaches. Remote Sensing
and GIS Data Processing and Other Applications. IEEE Conference Publications , Page(s): 174
M.Khedr , M. A. El-dosuky and WesamAhmed. : Jpeg Image Compression
Conventional RLC
Proposed RLC
mation Technology (IJCSIT) Vol 9, No 3, June 2017
147
Digits used in Encoding Sequence
zero values only as rest of the runs are 0s.
e in the test data compression ratio for all considered test cases and improving
advantage of enormously reducing image size. This is evident by the reduction in the total
Remote Sensing
Page(s): 174 –
dosuky and WesamAhmed. : Jpeg Image Compression

148
Using Discrete Cosine Transform. - A Survey International Journal of Computer Science &
Engineering Survey (IJCSES) Vol.5, No.2, April 2014
[3] Walaa M. Abd-Elhafiez. : New Approach for Color Image Compression. International Journal
ofComputer Science and Telecommunications. Volume 3, Issue 4, April 2012
[4] Walaa M. Abd-Elhafiez, WajebGharibi. : Color Image Compression Algorithm Based on the
DCT Blocks. IJCSE Volume-4, Issue-4 , Page no. 34-38, Apr-2016
[5] M.B Akhtar, , A.M Qureshi, , Qamar-ul-Islam . :Optimized run length coding for jpeg image
compression used in space research program of IST. International Conference on Computer
Networks and Information Technology, IEEE Conference Publications, Page(s): 81 – 85, 2011
[6] A. Banerjee and A. Halder. : An Efficient Dynamic Image Compression Algorithm based on Block
Optimization,Byte Compression and Run-Length Encoding along Y-axis. , published in (ICCSIT),
3rd IEEE International Conference on (Volume:8 ), 2010.
[7] A. Gupta, M.C. Srivastava, S.D. Pandey and V. Bhandari, . :Modified Runlength Coding for
Improved JPEG Performance. IEEE Conference Publications, Page(s): 235 – 237, 2007.
[8] A. Singh and V P Singh. . : An Enhanced Run Length Coding for JPEG Image Compression. Volume
72 - Number 20, IJCA Journal, 2013
[9] S.P. Bagal and V.B. Raskar . : JPEG Image Compression by Using DCT. Research Paper,
pages(34-38), Volume-4 , Issue-4 , 2016.
[10] Nasir Ahmad. The DCT – an algorithm that impacts the world of digital audio and video.
QUANTOM – Research and scholarship at the university of Mexico.
[11] A.K Jain –. : Fundamentals of digital image processing. , prentice hall , 1989, ISBN 0-13-
336165-9.

Squashed JPEG Image Compression via Sparse Matrix

More Related Content

What's hot (17)

Similar to Squashed JPEG Image Compression via Sparse Matrix (20)

More from AIRCC Publishing Corporation (20)

Recently uploaded (20)

Squashed JPEG Image Compression via Sparse Matrix