JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN

ELECTRICAL ENGINEERING

A REVIEW ON IMAGE COMPRESSION

1
MR.CHANDRESH K PARMAR, 2PROF.KRUTI PANCHOLI

1,2
Department of Electronics and Communication
L.J. Institute of Engineering and Technology, GTU,
Ahmedabad, Gujarat, India

[email protected], [email protected]
ABSTRACT : This Paper gives review of different types of Images and the different techniques for Image Compression.
Based on this Review we recommended general method for Image Compression. Image Compression is the technique of
reducing the image size without degrading the quality of the image. Various types of images and different compression
techniques are discussed here. Image Compression is the solution associated with transmission and storage of large amount
of information for digital Image. Transmission of Images includes different applications like broadcasting of Television,
remote sensing via satellite and other long distance Communication while Image storage is required for medical images,
satellite images, documents and pictures. Image compression deals with these types of applications. .

KEYWORDS: Image, Image compression technique, Huffman Coding, Run Length Encoding, Compression,
Decompression, Run Length Encoding, Transform Coding, LZW, Loss less and lossy image compression

1. INTRODUCTION
Image is a 2 Dimensional signal represented
by Digital system. Normally Image taken from the
camera is in the analog form. However for
processing, transmitting and storage, images are
converted in to digital form. A Digital Image is
basically 2- Dimensional array of pixels [1]. Basically
compressing of image is different than compressing
digital data. General purpose Data compression
algorithm can be used for Image compression but the Table 1 Different Uncompressed Images and its
result is less than optimal. Different types of images storage space
are used in remote sensing, bio medical and video B. Principle of Compression
processing techniques which require compression for Digital image is basically array of various pixel
transmission and storage. [2].Compression is values.[1] In the digital image Pixels of neighborhood
achieved by removing redundant or extra bits from are correlated and so that this pixels contain
the image. redundant bits. By using the compression algorithms
A. Need of Compression redundant bits are removed from the image so that
Uncompressed images can occupy a large size image size is reduced and the image is
amount of memory in RAM and in storage media, compressed.
and they can take more time to transfer from one Image compression Have two main Components:
device to another. Table 1 below shows the redundancy reduction and irrelevant data reduction
comparative size from normal text to high redundancy reduction is achieved by removing extra
compressed image. bits or repeated bits. While in irrelevant reduction the
Examples given in Table1clearly shows smallest or less important information is omitted,
need for sufficient storage space and more bandwidth which will not received by receiver. There are three
because long transmission time is required for types of redundancies.
uncompressed image. So there is only one solution is Coding redundancy is present when less number of
to compress the image. code words required instead of larger symbol. Inter

ISSN: 0975 – 6736| NOV 12 TO OCT 13 | VOLUME – 02, ISSUE - 02 Page 281
 JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN
ELECTRICAL ENGINEERING
pixel redundancy results from correlated pixels of an In scalar Quantizer quantization is performed on each
image. In psycho visual redundancy data is ignored coefficient while in vector quantization it can be
by the normal visual system. Image compression is performed on groups.
applied to reduce the number of bits which represent An entropy encoder compressed the quantized values
the image. and improves the compression. The reverse Process
1.1 PERFROMANCE PARAMETERS Decoder, De quantizer and inverse mapper is
There are two performance parameters are used to obtained to reconstruct the image and it is called
measure the performance of the image compression decompression.
algorithms. One is PSNR (peak signal to noise ratio) Organization of Paper
and second is Mean square error (MSE). PSNR is the In this paper, we present a review of various types of
measurement of the peak error between the images is discussed in section II and its compression
compressed image and original image. The higher the techniques in section III. In section IV the general
guidelines are mention to compress the image.
PSNR contains better quality of image.
2 TYPES OF IMAGES
To compute the PSNR first of all MSE (mean square A. TIFF
error) is computed. The TIFF (Tagged Image File Format) is a
Mean Square Error (MSE) is the cumulative flexible format which can be used for lossless or
difference between the compressed image and lossy Compression [4].In practice, TIFF is used as a
original image. Small amount of MSE reduce the lossless image storage format in which image
error and improves image quality. compression is not used. For web transmission TIFF
files are not used because TIFF files require large
size.
1.1
B. GIF
In the previous equation, M and N are the number of
Graphics Interchange Format (GIF) is useful for
rows and columns in the input images.
images that have less than 256 colors, grayscale.GIF
The PSNR is computed from following equation
is limited to an 8 bit or 256 colors. so that it can be
used to store simple graphics ,logos and cartoon
1.2 style images. It uses loss less compression.
C. RAW
RAW file format includes images directly taken
from Digital cameras. These formats normally use
loss less or lossy compression method and produce
smaller size Images like TIFF. The Disadvantage of
RAW Image is that they are not standardized image
and it will be different for different manufactures. So
these images require manufacture's software to view
the images.
D. PNG
The PNG (portable Network Graphics) file
Fig 1 General Compression Decompression format supports 8 bit, 24 bit, 48 bit true color with
As shown in Fig.1, First of all the image is taken and without alpha channel. Lossless PNG format is
from the image dataset. The mapper converts the best compare to lossy JPEG. Typically, an image in a
input image into inter pixel coefficients. PNG file can be 10% to 30% more compressed than
Transformation for the mapper may be DCT, wavelet in a GIF format [5].PNG format have smaller size and
or Curvelet transform. Each has its own advantages more colors compare to others.
and disadvantages. E. JPEG
Second stage is the quantizer which simply Joint Photographic Expert Group (JPEG) is a
reduces the number of bits needed to store the lossy compression technique to store 24 bit
transformed coefficients. It is many to one mapping photographic images. It is widely accepted in
in which larger values are quantized into smaller multimedia and imaging industries. JPEG is 24 bit
value. It is a lossy process and it is the main source of color format so it have millions of colors and more
compression in an encoder. Quantization reduced the superior compare to others[6].it is used for
number of bits so it results some kind of information VGA(video graphics Array) display.JPEG have
loss. Quantizer can be scalar or vector quantization.

ISSN: 0975 – 6736| NOV 12 TO OCT 13 | VOLUME – 02, ISSUE - 02 Page 282
 JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN
ELECTRICAL ENGINEERING
lossy compression and it support 8 bit gray scale RLE is the simplest image compression technique in
image and 24 bit color images. which sequence of identical symbols are replaced by
F. JPEG2000 a pair containing the symbol and the length at which
JPEG 2000 is a compression standard for the number is repeated.[8]. it is widely accepted
lossless and lossy storage.JPEG2000 improves the compression technique in the fax standard.
JPEG format. it is nearly same as JPEG. b. Statistical Coding
G. Exif The following techniques are included. 1. Huffman
The Exif (Exchangeable Image File Format) is Encoding, 2. Arithmetic Encoding 3. LZE Encoding.
similar to JFIF format with TIFF extension. it is [9][10]
used to record and exchange of images with image 1. Huffman Encoding
metadata between the digital camera and editing Huffman coding can reduce the file size by 10% to
and viewing software. 50% by removing the irrelevant information. In this
H. WEBP technique smaller bit code is given to the pixel values
WEBP is a new image format that use lossy which occur frequently and the higher bit code for
image compression. It was designed by Google to repeated pixel value. In order to encode images
reduce image file size to increase the speed when • First of all image is divided in to 8X8 blocks
web page is loading. It is based on VP8s infra- • Then each block is coded with particular
symbols
frame coding.
• Huffman code is applied to the each block
I. BMP
• Encoding all the blocks
The Bitmap (BMP) file format deal with 2. Arithmetic Encoding
graphic file related to Microsoft windows OS. Arithmetic encoding was introduced by Rissanen
Normally these files are uncompressed so they are in which the last symbol is encoded and decoded
large. These files are used in basic windows first.[11]Arithmetic encoding is based on
programming. [7] BMP images are binary following principle.
files.BMP file does not support true colors. • The symbol alphabet should not infinite.
J. NETPBM • All possible symbol sequence of give length
NetPbm format contain three family formats: should not infinite
the PPM (portable Pixel Map), the PGM (portable • The number of real number in the interval
Gray Map) and the Portable bit map. [8] These files [0,1] can assign a unique subinterval for any
are pure ASCII files or raw binary files. given input sequence of symbols.
3. LZW coding
3. COMPRESSION ALGORITHM
It is dictionary based coding, in the static
There are Two types of compression
dictionary coding the dictionary is fixed
algorithm: Lossless and Lossy. In the loss less
during the encoding and decoding while in
compression the compressed image is totally replica
dynamic dictionary coding the dictionary is
of the original input image, there is not any amount of
updated when new word is introduced.
loss present in the image. While in Lossy
4. Area coding
compression the compressed image is not same as the
It is an enhanced version of the RLE. Area
input image, there is some amount of loss is present
coding is highly effective and it can give
in the image.
better compression ratio but it has certain
A. Lossless compression Techniques
limitation that it can be applied to non linear
In lossless compression scheme reconstructed image
transformation
is same to the input image. Lossless image
B. Lossy Compression Techniques
compression techniques first convert the images in
Lossy compression technique provides
to the image pixels. Then processing is done on each
higher compression ratio compare to lossless
single pixel. The First step includes prediction of
compression. In this method, the compressed
next image pixel value from the neighborhood
image is not same as the original image; there is
pixels. In the second stage the difference between
some amount of information loss in the image.
the predicted value and the actual intensity of the
Lossy compression scheme is shown in fig. 2
next pixel is coded using different encoding
methods.
Different Encoding and Decoding Methods for Loss
less compression are discussed below.
a. RLE (Run Length Encoding)

ISSN: 0975 – 6736| NOV 12 TO OCT 13 | VOLUME – 02, ISSUE - 02 Page 283
 JOURNAL OF INFORMATION, KNOWLEDGE AND RESEARCH IN
ELECTRICAL ENGINEERING
4. CONCLUSION
Hence all basic image compression techniques have
been discussed. To conclude all the image
compression techniques are useful in their related
areas and every day new compression technique is
developing which gives better compression ratio.
This review paper gives clear idea about basic
compression techniques and image types. Based on
Fig 2 Lossy Compression Techniques review of different types of images and its
Compression algorithms we conclude that the
a. Transform Coding
compression algorithm depends on the three factors:
Transform Coding algorithm usually starts by
quality of image, amount of compression and speed
partitioning the original image into small blocks of of compression.
smaller size. Then for each block related transform 5. REFERENCES
coefficients are obtained based on their transform, [1] Subramanya A. “Image Compression
DCT, wavelet and Curvelet are the example of the Technique,”potentials IEEE, Vol. 20, issue 1,
transform coding. The resulting coefficients are then pp19-23, Feb-March 2001.
computed by quantization techniques and then the [2] Woods, R. C. 2008. Digital Image processing.
output of the quantizer is used for symbol encoding New Delhi: Pearson Pentice Hall, Third Edition,
technique to produce the output. At the decoder the Low price edition, Pages 1-904.
reverse process is obtained and image is [3] http://en.wikipedia.org/wiki/Image_file_formats
[4] Keshab K. Parhi, Takao Nishitan; “Digital
reconstructed. Signal processing for multimedia systems”, ISBN
b. Block Truncation coding 0-8247-1924
In this the image is divided into non overlapping [5] “Understanding Image
blocks of pixels. Then the quantizer is used to find Types”http://www.contentdm.com/USC/tutorial/i
mage-filetypes.pdf.1997-
mean of the pixel values of the all the non 2005,DiMeMa,Inc,Unpublished.
overlapping blocks. After that thresholding is done so [6] http://www.find-images.com
that the image pixels above the threshold values are [7] John Miano; “Compressed image file formats:
set to zero or one. Then for each segment in the JPEG,PNG, GIL, XBM, BMP”, Edition-2,
January-2000, page 23.
bitmap the related reconstruction value is obtained. [8] Majid Rabbani, Paul W.Jones; “Digital Image
Larger block size gives greater compression ratio but Compression Techniques”. Edition-4, 1991.page
it reduces the quality of an image. 51.
c. Sub –band Coding [9] Ronald G. Driggers; “Encylopedia of optical
engineering”, Volume 2, Edition 1,2003.
The sub band coding split the frequency bands of a
[10] Ioannis Pitas; “Digital image processing
signal and then each sub band is coded by encoder algorithms and applications.”, ISBN 0-471-
and bit rate which is related to that particular 37739-2
band.[12]. SBC is generally used in speech coding [11] A.S. Ragab, Abdalla S.A. Mohmed, M.S. Hamid,
“Efficiency of Analytical Transforms for Image
and image coding. At the decoder the sub band
Compression” 15th National Radio Science
signals are decoded and un sampled and passed Conference, Feb.24-26, 1998, Cairo- Egypt.
through a synthesis filters. Then all the sub band [12] Rafael C. Gonzalez, Richard Eugene; “Digital
coefficients are properly summed up to yield the image processing”, Edition 3, 2008, page
466.Alan Conrad Bovik; “Handbook of image
compressed image.
and video processing”, Edition 2 1005, page 673
d. Vector quantization [13] Deke McClelland, Galen Fott; “Photoshop
Vector quantization (VQ) technique is the extension elements3fordummies”Edition 1, 2002, page 126.
of Scalar quantization in multiple dimensions. This [14] Keshab K. Parhi, Takao Nishitan; “Digital Signal
technique develops a dictionary of fixed-size vectors processing for multimedia systems”, ISBN 0-
8247-1924-7, United States of America, page 22.
which are called code vectors. A given image again [15] Majid Rabbani, Paul W. Jones; “Digital
partitioned into non-overlapping blocks called image imagecompression techniques”; ISBN 0-8194—
vectors. Then for each image vector, the closest 0648-1,Washington, page 129.
[16] Vasudev Bhaskaran, Konstantinos Konstantinides
matching vector in the dictionary is determined and
“Image and video compression standards:
its index in the dictionary is used as the encoding of algorithms and architectures”. Edition 2, 1997,
the original image vector [13]. Because of its fast page 61.
lookup capabilities at the decoder side, Vector [17] Sonal, D. K. 2007. A study of various image
compression techniques. COIT, RIMT-IET. Hisar
Quantization-based coding schemes are normally
used in multimedia applications.

ISSN: 0975 – 6736| NOV 12 TO OCT 13 | VOLUME – 02, ISSUE - 02 Page 284