International Journal of Innovative Research in Computer Science & Technology (IJIRCST)

ISSN: 2347-5552, Volume-6, Issue-2, March-2018

DOI: 10.21276/ijircst.2018.6.2.1

A Review of Image Compression Using Fractal

Image Compression with Neural Network
Shamina Khatun, Anas Iqbal

Abstract- Generally the fractal image compression is a I. INTRODUCTION

new process in the images compression. It is a block Information transmission plays a very important role in
based image compression technique, which detects and transmission of data related to particular events like video
decodes the existing similarities between different conferences, medical data transfer etc, where much image
regions in the image. The main disadvantage of FIC is data is to be transmitted. But we know that, the storage
that the encoding time is comparatively very high, w space of very big data can requires large memory. Hence to
here as the decoding time is very short. An artificial overcome this problem all the images are compressed and
intelligence technique like neural network is used to decompressed using several compression and
reduce the search space and encoding time for the MRI decompression procedures. Image Compression plays a
images with an algorithm called as “back propagation” very important role in applications like medical imaging,
neural network algorithm. Initially, MRI image is remote sensing, tele-video conferencing and satellite image
divided into ranges and domains of fixed size. The best application. There are mainly two types of image
matched domain is selected for each range block and its compression
range index and best matched domain index are  lossless image compression and
produced, which acts as input to the expert system and  lossy image compression.
which results reduced the sets of matched domain
blocks. The neural network is then trained with these
A. Lossless Image Compression
resultant values. This trained net is now used to
compress other MRI images which lead to a very less In this, type of compression, the original image can be
encoding time. During the decoding phase, the actually recovered from the compressed data .It is generally
transformation parameters are recursively applied to used for discrete data, such as text, computer generated
any random original image, which then converges to the data, web Data and some kind of image & video
fractal image after some changes. The simulation results information.
show that the performance of this Neural Network
based FIC is really. This paper shows the neural B. Lossy Image Compression
network based FIC which produces high development in In this, type of compression, the original image cannot be
encoding time without corrupting the image quality accurately recovered from the compressed data. Due to
when compared to normal FIC. some losses of information of data cannot be exactly
recovered from compressed data. In this, much higher
Keywords- Fractal Image Compression (FIC), Neural compression ratios are achieved in compare to lossless
Network (NN), Back Propagation Neural Network compression.
(BPNN), Range Blocks, Neural Network (NN), Encoding
Time (ET),Iterated Function System (IFS). II. IMAGE COMPRESSION
The purpose of compressing image reduction, image
irrelevance and redundancy to transmit mode data in
efficient storage. Compression is obtained by removing one
or more data from the three unnecessary bases :
(1) Encoding and redundancy present when the code words
Manuscript Received March 05, 2018 used.
Shamina Khatun, M.Tech. Scholar All Saints College of Technology, (2) Pixel redundancy, between drifts gives pixel
RGPV University, Bhopal Madhya Pradesh India. correlations of an image.
Anas Iqbal, Professor, Department of Electronics & Communication
Engineering, All Saints College of Technology, RGPV University, (3) Psycho-visual Redundancy, What Causes and Data is
Bhopal , Madhya Pradesh , India. Mobile-+91-7407115241 Ignored by Visual Human System.

Copyright © 2018. Innovative Research Publications. All Rights Reserve 9

 A Review of Image Compression Using Fractal Image Compression with Neural Network

III. IMAGE COMPRESSION APPLICATION called as “back propagation neural network algorithm” is
 Effective use of storage space. used to increase the speed of the encoding time. At the
initial stage MRI image is divided into ranges and domains
 Reduce the transmission time of an image to be sent
of fixed size and twice size of the range size of the domain.
over the Internet or downloaded from websites.
A search is done to select the best matched domain for each
 Image Archiving: Standard Data.
range block and then its range index and best matched
 Image Transmission: Web Data. domain index are produced. This range block index is given
 Multimedia Application: Desktop Editing as a input to the expert system which results reduced set of
The main aim of the image compression is to represent the more or less matched domain blocks. Now the search is
image data in as few bits as possible and at the same time done only in resulted set of domain blocks and it produces
should maintain the high quality of the image necessary for set of Fractal codes as output.
an application. Image compression has two great purposes:
reduction of time for image transmission and reduction of A. Image compression and coding method based on
storage area. Image data redundancy is a key for image Neural Network
compression. Fractal image Compression (FIC) process is A new technique has been adopted in this while initializing
applied on images of different modalities like MRI, CT the load between input and unknown layer adopted as a
SCAN, X-rays, Ultrasound etc. This paper shows the neural substitute of randomizing the original weight. In this, the
network based Fractal image Compression (FIC) which Back Propagation Algorithm is based on multi-layered
produces high expansion in terms of encoding time without feed-forward net and is the most versatile algorithm in
corrupting the image quality compared to the normal FIC which the first layer receives the input and it produces the
which excludes exhaustive search. output. This process is repeat again and again and many
layers are produces its output which will act as a input for
A. Fractal Image compression the upcoming layer. After giving the input data to the input
Fractal image compression lies in the category of lossy layer, information passes through the network from the
compression technique, which was invented by Michael input layer to the output layer generally called as “forward
Barnsley in 1988 and which is based on iterated function propagation”.
system through which better reconstructed image with good
quality is obtained than the other image compressed V. LITERATURE SURVEY
methods. Fractal Image Compression (FIC) is a block based The fractal Image Compression was firstly proposed in a
image compression, which detects and decodes the
paper by Jacquin. After Jacquin paper many more paper
similarities between different regions of the images. The comes on this topic. Here the some paper describes the
two major benefits of changing images to fractal data. The design of fractal image compression. Some of the papers
first one is that the fractal data occupies less size of are successful to give high compression ratio and some of
memory than the amount of memory used to store the these gave less encoding time.
original data. The second benefit is that since the data is Geoffrey Fellows et al [1] proposed that these temporary
mathematical, the image can easily be scaled up or down a folders help forensic analysts to provide evidence for the
size without changing the detail of the image.
files that have been viewed from the WinRAR archive.
Such evidence will be very useful when user denies that the
B. Image compression and coding method based on files were not accessed by him. WinRAR software
fractal dimension furnishes encryption along with compression.
The fractal dimension is the most important parameter to Wang et al [2] proposed a method in which related
describe the fractal attribute structure. weights of confined features are added in both spatial and
The segmentation of images are of two types domain block descriptor domain and this method is useful to retrieve
and range block. Here, the domain block is the maximum images from database which contains large number of
and range block is minimum in size. For the first, use a images. With very small computational overhead power of
fractal image encoding method based on iterative algorithm individual local features can be improved. Experimental
system is use the Discrete Cosine Transform (DCT) results from benchmark tests show this method has greater
compression method and coding. Through the classification performance with less computation and retrieval accuracy.
of image block, the index image block of the original image Stephen et al [3] made a comparative study of lossless
is displayed, and then up to the current data compression and lossy radiography image compression and presented the
and IFS JPEG code the code. This type of image-based challenges of using lossy methods for medical records by
segmentation method based on visual characteristics has a explaining fundamental concepts of radiologic imaging and
high compression ratio. then made a detailed study of current compression
technique in the field of medical imaging and discussed
IV. ARTIFICIAL NEURAL NETWORK regulatory policies and legal questions facing the use of
An artificial intelligence technique like neural network is compression in this field.
used to reduce the search space and encoding time for the Khafaji et al [4] proposed an image compression scheme
Magnetic Resonance Image (MRI) images. A algorithm based on locally changing image characteristic. Derived

Copyright © 2018. Innovative Research Publications. All Rights Reserve 10

 International Journal of Innovative Research in Computer Science & Technology (IJIRCST)
ISSN: 2347-5552, Volume-6, Issue-2, March-2018
DOI:10.21276/ijircst.2018.6.2.1

from this information an image is divided into blocks and image after fewer than ten iterations. Initially, the input
then using polynomial approximation the image is MRI image is partitioned into fixed blocks of size are called
decomposed into less compressed data. Finally performance Range Blocks (RB) and the block process operation is
rate is improved by Huffman encoding. executed to reduce the recently formed image by averaging
Mohammed S. Mahaboob Ismail.B Basha et al [5] said the intensities of four neighboring pixels and the resulted
improved image fractal image compression (IFIC) method three vectors of encoding phase are considered now. A
for color images made for the variable block size. The single block is extracted from the reduced image first. Then
image here is divided into blocks considering the maximum its corresponding best matched domain block index, scaling
and minimum range. Here, the fixed block size range of 4 x parameter and its offset values are retrieved from the
4 iterations and other existing methods. This card reaches a vectors. The pixel values of the reduced domain block are
compression ratio of up to 20 with a high peak signal-to- then placed in the location in the range obtained by the
noise ratio (PSNR) of 30 dB. orientation information after scaling and offsetting. All the
range blocks constitute a single iteration. The decompressed
M.K. Revathy Jayamohan et al [6] show that the
image will be constructed after one to ten iterations.
computational complexity of fractal image compression is
mainly due to a large number of comparisons needed to find
VIII. CONCLUSION
a corresponding block corresponding domains
corresponding to block ranges within the image. This article This Review paper represents the Concept of Fractal Image
discusses the multi-way computational efficiency of tree compression using Neural Network. All compression
research has been investigated for domain information techniques are useful and each day is developing a new
storage. Cabin domains will be listed in a B ++ tree ordered compression technique that provides a better compression
in one or more selected local features of each domain. ratio. Image compression is one of the most important
applications in the field of digital image processing. Due to
VII. WORKING OF IMAGE COMPRESSION BASED the limited storage capacity and bandwidth, a medical
ON FRACTALS AND BACK PROPAGATION imaging requires to storing the large quantities of digitized
NEURAL NETWORK clinical data must be compressed before transmission and
storage. But the main disadvantage of the fractal image
A fractal image compression algorithm proposed for MRI
compression is the high encoding time. A fractal image
images, utilizes the self-similarity property of a MRI image
compression for MRI images introduce in this paper has
and it comprises of three phases namely Training, Encoding
reduced the search and encoding time in which the expert
and Decoding .The main aim of this compression technique
system has employed during the encoding process, to
is to reduce the search space and speedup the encoding
increase the speed of encoding process without severely
time. These three phases are detailed in the following sub-
affecting the image quality.
sections.
A. Training Phase REFERENCES
In this phase, similar MRI images are taken as input images [1] Geoffrey Fellows, “WinRAR temporary folder
and each image is partitioned into range blocks and domain artefacts”, Digital Investigation, Volume 7, Issue 1-2,
blocks separately. At the initial stage, the range block of October 2010, Pages 9-13.
one MRI image is compared with the domain blocks of the [2] Wang, X., Yang, M., Cour, T., Yu, K, Han, T.X.
same MRI image to select the best matched domain block “Contextual weighting vocabulary tree based image
for that particular range block in an image and the fractal Retrieval”, Proceedings of the IEEE International
codes are given as output. Then the expert system has Conference on Computer Vision 6126244,pp. 209-216.
trained with the indices of range and its best matched
domain blocks. The time taken in the training process will [3] Wong, Stephen, Zaremba, Loren, Gooden, David,
not be counted with the encoding time. Huang, H.K. “Radiologic Image Compression-a review”,
Proceedings of the IEEE 82(3), pp.194-219.
B. Encoding Phase [4] Ghadah Al-Khafaji “Image Compression based on
During the encoding phase, the expert system takes the Quadtree and Polynomial”, International Journal of
index of the range block of MRI image one by one as its Computer Applications (0975- 8887) Volume 76-No.3.
input and it produces a set of domain blocks for the parallel [5] I. B. Mohammed, “Improved Fractal Image
range block. Then the search will be done only in the Compression Using Range Block Size,” in In 2015 IEEE
resulted set of domain blocks Hence, the search space is International Conference on Computer Graphics, Vision
reduced and speed of the encoding time is increased. and Information Security (CGVIS), pp. 284–289, 2015.
C. Decoding Phase [6] M. Jayamohan, “Domain Classification using B + Trees
It is the reverse process of encoding is done in which the in Fractal Image Compression,” in In 2012 IEEE National
transformation parameters are applied to an initial image Conference on Computing and Communication Systems
with mean value and it will then converge to the fractal (NCCCS), pp.1-6, 2012.

Copyright © 2018. Innovative Research Publications. All Rights Reserve 11