NOVATEUR PUBLICATIONS

International Journal of Research Publications in Engineering and Technology [IJRPET]

ISSN: 2454-7875
VOLUME 3, ISSUE 7, July-2017
COMPARATIVE STUDY OF TECHNIQUES FOR IMAGE COMPRESSION
SACHIN.R.PRASAD,
UG Student, BNMIT, Bangalore, India

SUNIL RAJ.H,
UG Student, BNMIT, Bangalore, India

AKSHATH.K
UG Student, BNMIT, Bangalore, India

PRADEEP.M.R
UG Student, BNMIT, Bangalore, India

PRIYA.R.SANKPAL
Assistant Professor, Dept. of telecommunication, BNMIT, Bangalore, India

ABSTRACT: Fractal image compression uses iterated function

Now a day’s image storage and transmission has system where in each pixel is compressed using suitable
become one of the biggest concerns due to shortage of coding technique. Compression happens iteratively. At
storage and delay in transmission. Therefore image standard compression ratios like 50:1, fractal compression
compression is very much needed for faster provides similar results as compared to existing methods
transmission and saving of space. The existing like JPEG. But at higher compression ratios, the fractal
technologies have a number of limitations like loss of image compression provides very good results compared
information, lower compression ratios, lower PSNR to the existing methods.
and high compression time. Fractal image
compression using quad tree decomposition and II. EXISTING SYSTEM:
Huffman coding has been designed and used on a set of In this survey, the aim is to identify different
standard images like Lena, cameraman and baboon. image compression techniques available. There are three
The results obtained yield good compression ratio and different image compression techniques identified. Those
PSNR. The method is also applied to set of fractal are JPEG, wavelet, VQ compression and fractal
images and satellite images. The results obtained are compression. JPEG is based on quantizing the DCT blocks
found to be better for fractals and satellite images in the image and then compressing the blocks using
compared to other existing methods. Huffman coding. Wavelet compression is based on
KEYWORDS: fractals, compression ratio, PSNR, converting each of the image components into wavelets
satellite images. and processing them. VQ compression deals with the
image displayed as a vector and quantizing each vector for
I. INTRODUCTION: encoding and decoding operation. Fractal compression
Images are pictorial representation of required deals with natural textures and images with repetitive
information. Internet is one of the major sources to store pattern. [13]
all the images. With the advancement of the information In this paper, the types of image compression
age the need for mass information storage and retrieval techniques are compared for their performance. The three
grows. The capacity of commercial storage devices, main techniques used are Huffman coding, discrete
however, has not kept pace with the proliferation of image wavelet transform (DWT) and fractal image compression.
data. Any real time application requires the information These techniques are simple to use and requires less
size to be less for efficient processing. Thus image memory to store the data. Huffman coding is used when
compression plays an important role in efficient use of there is a lot of redundancy in the input data. It is used to
time and space for data processing. reduce the redundancy of the data. DWT is used to enhance
Fractals are naturally occurring patterns that can the quality of the compressed image. The degradation in
be analyzed easily compared to regular images. the quality happens after compression. DWT enhances the
There are many image compression techniques quality using suitable wavelet substitutions. Fractal image
available using standard algorithms. However, there are compression involves encoding and decoding operation
limitations for these methods. and it increases the compression ratio after
decompression. By using the above algorithms the
112 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 7, July-2017
calculation of Peak signal to noise ratio (PSNR), Mean III. FRACTAL IMAGE COMPRESSION:
Square error (MSE) and compression ratio (CR) and Bits Veena Devi s. V and a. G. Ananth proposed a method
per pixel (BPP) of the compressed image by giving which is more secure than the existing techniques
512×512 input images and also the comparison of available. Fractal image compression using quad tree
performance analysis of the parameters with that above decomposition and Huffman coding provides better
algorithms is done. The result clearly explains that Fractal compression ratio than some of the available methods. The
algorithm provides better Compression ratio (CR) and time taken for compression is less thereby saving delay in
Peak Signal to noise ratio (PSNR). [12] transmission of digital images. The size of the
In this paper, the proposed method is an attempt decompressed image is almost 50% lesser than the
to reconstruct and image using inverse halftone and original image size for standard set of images like Lena,
Huffman coding. The importance of transmission of audio baboon, cameraman. The proposed system is further
and video data over certain devices using multimedia implemented for fractal images which have repetitive
application has increased. Biometry, content based image pattern. The values of PSNR and compression ratios
retrieval and CCTV footage require a lot of memory to obtained for fractals are high. The information loss in the
store the information. Thus inverse half tone and Huffman proposed system is very much less (almost negligible)
coding is very much useful to reduce the memory space when compared to the existing system. Proposed system
required to store these information. To achieve high yields in complete information retrieval as when compared
compression ratios, lossy halftone and lossless Huffman to other existing system when satellite images are used.
coding is combined and used. This technique can be used Satellite images also provide better decompression size
for low bit rate video transmission and mass image when the proposed method is used for compression.. The
storage. [6] compression and decompression yields to almost 80-90%
In this paper, the proposed method is an attempt reduction in the size of the original data, thus it can be
to use one of the basic methods for fractal image used for efficient transmission. In the proposed method,
compression. The whole idea is to determine the the image is first converted to blocks. Each block is further
probability of error for each image data and apply suitable applied with Huffman encoding algorithm and reduced to a
technique to reduce the mean square error (MSE) of the code word. Each code word is then processed using
image blocks. However, there is a limitation for fractal Huffman decoding to get back the original data blocks. All
image compression as the coding time in more compared the blocks are combined to get back the original
to the decoding time. This paper proposes a method to decompressed image.
reduce the coding time by classifying the blocks according
to an approximate error measure. It proposes the use of a WORKING:
preset block with can be used to time the range or domain
Input image Quad tree
blocks thereby reducing the comparing time. This process decomposition Huffman
is efficient for bit rate variations and computing time encoding
variations of the blocks. [9]
Block truncation coding (BTC) is one of the basic
and standard methods used for satellite image
compression. The traditional method involves computation
Decompre Huffman Compressed
of a high mean and the low mean to replace the original
ssed image decoding image
pixel values. Here the first and the second moments are
preserved and the bit rate is a constant value
approximately 2 bits per pixel. The disadvantage with this Fig.1: Block Diagram of Proposed System
method is the blocking attributes increases as the block Input data in the form of digital image is fed into
size increases. Many methods have been evolved in order the system. The input image is divided into a set of square
to improve the compression ratio and also to reduce the bit blocks using quad tree decomposition. The image is
rate. In this paper an improved block truncation coding divided into two types of blocks namely range and domain
algorithm along with an adaptive lossless compression blocks. The blocks are compared for matching of
scheme is proposed to improve on the compression ratio information. If the information in both the blocks is
and Peak signal to noise ratio. The computational matching, then the input pixels are mapped onto the
complexity is also further reduced and the blocking corresponding blocks.
artifacts which are inherent in the traditional BTC are also Each block is further applied with encoding. Here
minimized to a great extent. [11] Huffman encoding algorithm is used. Huffman encoding
assigns probabilities for each blocks and arranges the

113 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 7, July-2017
blocks in descending order. The last two probabilities are Table 1 shows the results obtained for these
then added and the resulting probability is places back in images before and after compression.
the table in its suitable position. The last two probabilities Table 1: results obtained for general images
that are added are assigned with a binary value of 0 or 1. Test Dimen Compr Compr Decompr PSNR %
input sion ession ession ession (DB) reduction
This process is repeated until the output reduced to a time ratio time in size
single code words with a string of 0’s and 1’s. (sec) (sec)
Lena 256*2 0.668 9.681 6.9759 25.24 48.6
At the decoding stage, the code word is fed into
56 1 6
the decoding block. Huffman decoding algorithm is used to 512*5 0.712 9.456 7.3408 24.97 48.6
get back the original blocks of data. The first bit is read, if it 12 7 1
1024* 0.693 9.512 7.0196 25.03 48.6
is 0 follow the bottom edge of the tree. If the first bit is 1 1024 5 2
then follow the top edge of the tree to get back the original
Came 256*2 0.60 13.78 4.7227 24.93 48.1
block. The blocks are combined later to get back the ra 56 25
original image in the decompressed form. man
512*5 0.566 13.54 4.7901 24.53 48.4
The quad tree decomposition uses affine
12 5 74
transformation where the image is tilted by an angle of 45 1024* 1.364 13.54 6.6683 24.63 48.4
degrees so that after block conversion the information is 1024 7 46
Babo 256*2 0.618 9.296 21.7766 26.29 47.9
properly mapped onto the blocks. Affine transform is used on 56 7 8
to prevent loss of information at the boundaries and edges 512*5 0.575 8.666 6.7806 25.83 49.3
of the image. Huffman coding algorithm uses Iterated 12 2 7
1024* 0.541 8.783 6.7620 25.92 49.3
Function System (IFS) in order to encode and decode the 1024 8 2
given data. The compression ratio is used to measure the Fig.3 shows some of the fractal images that are
ability of data compression by comparing the size of the used for analysis.
image being compressed to the size of the original image.
Peak signal to noise ratio (PSNR) is given by the formula,

PSNR (db) = 20 log ((signal power)/(noise power))

IV. EXPERIMENTAL RESULTS:

Fractal image compression has been applied to
different set of images. The images include standard
images like Lena, cameraman and baboon with dimensions
256*256, 512*512 and 1024*1024. The method is also
applied to satellite images of variable dimensions. The
method is also applied to a set of fractal images such as
Christmas tree, papaya leaf, fern leaf, cauliflower, fractal Fig.3: fractal images used for analysis
design and color pattern. Table 2 represents the results obtained for these
Fig.2 represents the output obtained for a set of fractal images.
standard images like Lena, cameraman and baboon. Table2: results obtained for fractal images
Test Dimen Compress Compre Decompress PSNR(D %size
input sion ion time ssion ion time B) reduce
(sec) ratio (sec) d
Fractal 512*5 1.79 5.8860 13.5584 14.716 44.72
design 12
Christm 512*5 0.7691 8.2917 9.9925 23.964 74.22
as 12
Tree
Color 512*5 4.73 7.66 60.18 25.48 42.25
Pattern 12
Papaya 512*5 4.99 8.96 50.64 24.26 90.85
Leaf 12
Fern 500*5 2.79 13.9 28.81 26.84 92.71
Leaf 00
Cauli 3276* 3.46 13.84 34.96 26.63 89.45
flower 3276

Fig.4 shows some of the satellite images for which

fractal image compression is applied.
Fig.2: output obtained for general images
114 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 7, July-2017

Fig.7: DCT compression output

Fig.4: satellite images used Table 5 shows the results obtained for DCT
Table 3 represents the results obtained for the compression.
satellite images used. Table5: DCT compression results
Method(Lena PSNR(db) Compression %compression
Table3: results obtained for satellite images image) ratio
Test Dime Compr Compr Decompres PSNR %size DCT 23.1452 1.0949 <10
input nsion ession ession sion (DB) reduced
time ratio time(sec)
Fig.8 shows the output obtained for satellite image
(sec) compression using BTC technique.
City 156* 0.860 6.598 10.4101 20.216 77.88
157
Crater 500* 0.867 7.528 9.0497 23.869 96.78
500
Fig.5 shows the output obtained for JPEG image
compression for Lena image.

Fig.8: BTC output

Fig.9 shows the output for same image using proposed
method.

Fig.9: proposed method output

It is evident from figures 8 and 9 that fractal image
Fig.5: JPEG compression output compression will not have any information loss when
Fig.6 shows the plot of number of DCT co-efficient compared to BTC technique. Also the gray levels are not
taken for compression versus the PSNR values obtained for reduced for processing rather it is maintained at the same
different co-efficient. levels.

V. COMPARISON OF METHODS:
Table 6 represents the comparison between the
existing methods and the proposed method.
Table 6: results comparison
Method used for PSNR Compression %compression
Lena image ratio
Fractal image 25.24 9.6816 48.61
compression (FIC)
JPEG compression 45.5 10.456 <45
DCT compression 23.1452 1.0949 <10
From the table 6, it is seen that the results
Fig.6: DCT co-efficient vs PSNR values obtained for FIC is far more superior when compared to
Table 4 represents the results obtained for JPEG the basic DCT technique. But the results obtained for JPEG
image compression for Lena image. image compression is higher than the FIC technique. Thus
Table4: JPEG compression results the performance of FIC technique lies between DCT and
Method(Lena PSNR(db) Compression %compression JPEG technique when it comes to Lena image.
image) ratio The main disadvantage with the JPEG image
JPEG 45.5 10.456 <45
compression is that the block size is limited to 64 (8*8).
Fig.7 shows the DCT compression output.
115 | P a g e
 NOVATEUR PUBLICATIONS
International Journal of Research Publications in Engineering and Technology [IJRPET]
ISSN: 2454-7875
VOLUME 3, ISSUE 7, July-2017
The blocks which exist above the 64 mark are not compared to BTC where the information at higher gray
processed as it is limited by DCT. But in fractal image levels is discarded.
compression the blocks of any size can be processed
without re-sizing the blocks.
Coming to satellite images, the compression ratio REFERENCES:
obtained for the image considered (8 *8 blocks) for BTC 1) Veena Devi S. V and A. G. Ananth “fractal image
technique (approx. 20) [11] is almost twice the compression using quad tree decomposition and
compression ratio obtained for FIC technique. The PSNR huffman coding”, SIPIJ, vol. 3.no.2, April 2012.
values obtained are also more when compared to the FIC 2) Fischer Y “Fractal image compression: theory and
technique. But the loss of information is very less in the FIC application”, New York, Springer- Verlag, 1995.
technique when compared to the BTC technique. Also the 3) Arnaud E. Jacquin “Fractal image coding”, Proceedings
gray levels in the image need not be modified in the FIC of IEEE, vol. 81.no.10, pp. 1451-1465, October 1993
technique. 4) Bohong Liu and Yung Yan “Improved fractal image
coding based on the Quad tree”, IEEE 3rd International
VI. CONCLUSION: Congress on image and signal processing, pp. 529-532,
Fractal image compression using quad tree 16-18 October 2010.
decomposition and Huffman coding has been applied to set 5) Dr. Muhammed Kamran, Amna Irshad Sipra and
of standard images like Lena, cameraman and baboon for Muhammd Nadeem “A novel domain optimization
various dimensions like 256*256, 512*512 and technique in fractal image compression”, IEEE
1024*1024. The compressed images gave almost no loss of proceedings on 8th world congress on intelligent
information as when compared to standard JPEG image control and automation, pp. 994-999, 7-9 july 2010.
compression algorithm. The method is also applied to 6) H. B. Kekre, Tanuja S. Sarode, Sanjay R. Sange “image
satellite images and natural textures and fractals reconstruction using fast inverse halftone and huffman
(repetitive pattern) and it is found that the results coding technique”, IJCA, vol. 27.no.6,pp. 34-40, august
obtained for these images are much efficient when 2011.
compared to standard DCT and JPEG compression. 7) Manoj Aggarwal and Ajai Narayan “Efficient Huffman
The results have been compared for standard Lena image decoding”, IEEE Trans, pp. 936-939, 2010.
of 256*256 dimensions with existing JPEG and DCT 8) Hui Yu, Li Li, Dan Liu, Hongyu Zhai, Xiaoming Dong
algorithms. It is evident from the results shown that JPEG “Based on quad tree fractal image compression
provides better compression ratios and PSNR values when improved algorithm for research”, IEEE Trans., pp. 1-3,
compared to the proposed method. But the compressed November 2010.
image has more information loss when compared to 9) Riccardo Dastasi, Michele Nappi and Daniel Riccio “A
proposed method. range or domain approximation error based approach
The higher frequency components of the image for fractal image compression”, IEEE Trans. On image
that are less sensitive to the human eye are discarded processing, vol. 15.no.1, pp. 89-97, 12 December 2005.
using DCT or JPEG technique. But in our method the higher 10) Veena Devi S. V and A. G. Ananth “Fractal image
frequency components are given the same priority as the compression of satellite imageries, IJCA, vol. 30.no.3,
lower frequency components there by reducing the loss of pp. 33-36, September 2011.
information. The quantization levels are reduced to a block 11) S. Chandravadana and N. Nithyanandam “Compression
size of 64 (8*8) in the JPEG technique. But in our method of satellite images using lossy and lossless coding Coding
the block size can be varied according to the user. Also our techniques”, IJET, vol. 6.no.1, February 2014.
method can be used in any dimension, size and format 12) R. Praisline Jasmi, B. Perumal and M. Pallikonda
(JPG,GIF,TIFF,PNG etc..) thus being more flexible for Rajashekaran “Comparison of image compression
processing when compared to the existing methods. techniques Using Huffman coding, DWT and fractal
In our project we have compared the satellite algorithm, ICCCI, January 2015.
image results with the existing block truncation coding 13) Sharanjeet Kaur “A review paper on Image compression
(BTC). Our method provides no information loss when Techniques”, IJCSE, vol. 5. Issue 1,February 2016.

116 | P a g e