1

CHAPTER 1

INTRODUCTION

1.1 OVERVIEW
Image Processing is a technique for various applications to enhance raw
images received from cameras/sensors placed on satellites, space probes, and
aircraft or pictures taken in healthy day-to-day life. The computational analysis of
objects in images is a very challenging issue as it usually involves automatic tasks
for segmentation. This refers and includes the detection of the objects represented,
extraction of representative features from the objects, matching between images,
rigid and non-rigid alignment of images, temporal tracking and motion analysis of
features in image sequences. Hence,image processing methodology has been
largely preferred. It is a form of signal processing for which the input is an image,
and the output of image processing may either be an image or a set of
characteristics or parameters related to the picture. Most image-processing
techniques treat the image as a two-dimensional signal.The quality of input images
plays a crucial role in the success of any computational image analysis task. This
is because the higher their quality is, the easier and simpler the task can be. Thus,
to improve the original quality of the input images, suitable methods of
computational image processing, such as noise removal, geometric correction,
edges and contrast enhancement and illumination correction or homogenization,
are required.

Despite inherent difficulties, computational methods of image processing

and analysis provide a wide range of important applications for the society. The
 2

essential topics within the field of image processing include Image restoration,
Image enhancement, Image compression, etc.
Image processing deals with manipulation and analysis of images,by using
a computer algorithm, and improving the pictorial information for
clearunderstanding. This area is characterized by the need for an extensive
experimental work to establish the viability of the proposed solutions to a given
problem. Thus, Image processing manipulatesimages to extract information to
emphasize or de-emphasize certain aspects of the information contained in the
image or perform image analysis to obtain hidden information.

Another aspect of image processing involves compression and coding of

visual information. With the growing demand for various imaging applications,
storage requirements of digital imagery are growing explosively. Compact
representation of image data, and their storage and transmission through
communication bandwidth are the crucial and active areas of development today.
Interestingly enough, image data generally contain a significant amount of
excessive and redundant information in their canonical representation. Image
compression techniques help to reduce the redundancies in raw image data which
ultimately reduces storage and communication bandwidth.

1.2 HISTORY OF IMAGE PROCESSING

In some sense, "image processing" dates back to the earliest use of graphics
by humans. With the cost of processing being relatively high, it changed in the
1970s when digital image processing proliferated as cheaper computers and
dedicated hardware became available. The fast computers and signal processors
were available in the 2000s making the digital image processing as the most
common form of image processing. Subsequently, it is used because it is not only
the most versatile method but also the cheapest. The conventional methodologies
 3

used median filteringin many research papers since the authors considered it as an
efficient methodology for removing the noise.

In the early days, topics like median filtering existed in many research
papers since the authors considered it as an efficient methodology for removing
the noise.

1.3 METHODS OF IMAGE PROCESSING

There are two methods available in Image Processing, namelyanalog

image processing and digital image processing.

1.Analog Image Processing

Analog image processing is basically a common image processing task

conducted on two-dimensional analog signals by analog means, i.e., the alteration
of the image through electrical means. The most common example is the television
image. Analog or visual techniques of image processing can be used for the hard
copies like printouts and photographs. When creating images using
analogphotography, the image is burnt into a film using a chemical reaction,
triggered by controlled exposure to light. It is processed in a dark room, using
select chemicals to create the actual image. The popularity of this process is
decreasing due to the advent of digital photography, which requires less effort.

2.Digital Image Processing

Digital image processing is the use of computer algorithms to perform

image processing on digital images.Digital image processing has three major
advantages which include reliably good image quality, small processing costs, and
the potential to control any part of the operation. The image is stored as a data file
in digital photography. This file is adapted to construct an actual image using
 4

photographic tools. The shades and variations are all registered when the picture is
taken, and this information is converted into an image by the program. Digital
image processing methods deliver big advantages such as flexibility,
reproductiveness and preservation of original data accuracy.

Various Image Processing techniques are:

A. Image representation
B. Image pre-processing
C. Image enhancement
D. Image restoration
E. Image analysis
F. Image reconstruction
G. Image data compression

(A) Image Representation

An image defined in the "real world" is considered to be a function of two

real variables as in the case of f(x,y) with f as the amplitude (e.g., brightness) of
the image at the actual coordinate position (x,y). The effect of digitization is
shown in Figure 1.1.
 5

Figure 1.1 Effects of Digitization

The 2D continuous image f(x,y) is divided into N rows and M columns.

The intersection of a row and a column is called a pixel. The value assigned to the
integer coordinates [m,n] with {m=0,1, 2,...,M-1} and {n=0,1,2,...,N-1} is f[m,n].
In fact, in most cases f(x,y)--which might be considered as the physical signal that
impinges on the face of a sensor. Typically, image files such as BMP, JPEG,
TIFF, etc., have some header and picture information. A header usually includes
details like format identifier (typically first information), resolution, number of
bits/pixel, compression type, etc.

(B) Image Pre-Processing

It used to remove noise and eliminate irrelevant, visually unnecessary information.

Noise is unsolicitedinformation that can result from the image acquisition
processing.

1. Scaling

Image scaling is the process of resizing an image. Scaling is a non-

trivial process that involves a trade-off between efficiency, smoothness, and
sharpness. With bitmap graphics, as the size of an image is reduced or enlarged,
the pixels that form the image become increasingly visible, making the image
appear "soft" irrespectiveof pixels averaged or jagged if not. With vector graphics,
the trade-off may be in processing power for re-rendering the image, which may
be noticeable as slow re-rendering, with still graphics, or slower frame rate and
frame skipping in computer animation. This technique of resizing is referred in
literaturesof many terminologies, such as image interpolation, image resizing,
image re-sampling, digital zooming, image magnification or enhancement etc. An
image interpolation algorithm is used to convert an image from one resolution
 6

(dimension) to another resolution without losing the visual content in the picture.
Image interpolation algorithms can be grouped in two categories namely non-
adaptive and adaptive. In non-adaptive algorithms, computational logic is fixed
irrespective of the input image features, whereas in adaptive algorithms
computational logic is dependent upon the intrinsic image features and contents of
the input image. When the image is interpolated from a higher resolution to a
lower resolution, it is called as image down-scaling or down-sampling. On the
other hand, when the image is interpolated from a lower resolution to a higher
resolution, it is referred as image up-scaling or up-sampling. Image interpolation
has a variety of applications in the areas of computer graphics, editing, medical
image reconstruction. For instance, scaling up is used to enlarge images for HDTV
or medical image displays while scaling down is applied to shrink images to fit
mini-size LCD panel in portable instruments. It is also a part of many commercial
image processing tools or freeware graphic viewers such as Adobe Photoshop CS2
software, IrfanView, Fast Stone Photo Resizer, Photo PosPro, XnConvert etc.
Numerous digital image scaling techniques have been presented, of which the
most popular methods are pixel replication based nearest neighbor replacement
algorithm, Pixel interpolation based Bi-linear, Filter/Kernel based Cubic, Bicubic,
B-Spline, Box , Triangle, Lanczos.

2. Rotation

Rotation is used in image mosaic, image registration, etc. One of

the techniques of rotation is 3-pass shear rotation, where the rotation matrix can
decompose into three separable matrices. In 3-pass shear rotation, there is no
scaling, i.e., no associated resampling degradations. Shear can be implemented
very efficiently.
 7

3. Mosaic

Mosaicing is one of the techniques of image processing which is

useful for tiling digital images. Mosaicing is blending together of several
arbitrarily shaped images to form one large radiometrically balanced image, so
that the boundaries between the original images are not visible. Any number of
geocoded images can be blended together along userspecified cut lines (polygons).
Mosaicing is a special case of geometric correction where registration takes place
in the existing image. If ground control points (GCP) are collected, the input
image is transformed according to the derived polynomial into the output image. If
no GCPs are provided, but both images already have compatible georeferencing,
then an appropriate translation and scaling will be applied instead of polynomial
transformation. This technique is generally used on several images like remote
sensed images, bio-medical images or other digital images.

(C )Image Enhancement

Image enhancement techniques can be divided into two broad categories:

1. Spatial domain methods, which function directly on pixels

2. Frequency domain methods, which function on the Fourier transform of

an image

1.Spatial domain methods

The value of a pixel with coordinates (x,y) in the enhanced image is the
result of performing some operation on the pixels in the neighbourhood of (x,y) in
the input image, F. Neighbourhoods can be ofany shape, but usually, they are
rectangular.

2.Frequency domain methods

 8

Image enhancement in the frequency domain is straightforward. The

Fourier transform of the image to be enhanced is simply computed, the result is
multiplied by a filter, and the inverse transform is taken to produce the enhanced
image. The idea of blurring an image by reducing its high-frequency components
or sharpening an image by increasing the magnitude of its high-frequency
components is intuitively easy to understand. However, computationally, it is
often more efficient to implement these domain operations. Understanding
frequency domain concepts is important and leads to enhancement techniques that
might not have thought of by restricting attention to the spatial area.

(D) Image Restoration

Image restoration refers to the removal or minimization of degradations in

an image. This includes the de-blurring of images degraded by the limitations of a
sensor or its environment, noise filtering, and correction of geometric distortion or
non-linearity due to sensors. It is the process of taking an image with some known
or estimated degradation and restoring it to its original appearance. Image
restoration is often used in the field of photography or publishing, where an image
is somehow degraded but needs to be improved before it can be printed.

(E) Image Analysis

Image analysis methods extract information from an image by using

automatic or semi-automatic techniques termed as scene analysis, image
description, image understanding, pattern recognition, computer/machine vision.
Image analysis differs from other types of image processing methods, such as
enhancement or restoration, in that the final result of image analysis procedures is
a numerical output rather than a picture.

(F) Image Reconstruction

 9

Image reconstruction encompasses the entire image formation process and

provides a foundation for the subsequent steps of image processing. The goal is to
retrieve image information that has been lost in the process of image formation.
Therefore, image reconstruction requires a systems approach that takes into
account the entire process of image formation, including the propagation of light
through inhomogeneous media, the properties of the optical system, and the
characteristics of the detector. In contrast to image enhancement, where the
appearance of an image is improved to suit some subjective criteria, image
reconstruction is an objective approach to recover a degraded image based on
mathematical and statistical models.

(G) Image Compression

The objective of image compression is to reduce the size of digital images

to save storage space and transmission time. Lossless compression is preferred for
artificial images like technical drawings, icons and also preferred for high-value
content, such as medical imagery or image scans made for archival purposes.
Lossy methods are especially suitable for natural images such as photos in
applications where the minor loss of fidelity is acceptable to achieve a substantial
reduction in bit rate. The lossy compression that produces unnoticeable differences
can be called visually lossless. Run-length encoding, Huffman encoding, and
Lempel Ziv encoding are the methods for lossless image compression. Transform
coding such as DCT and Wavelet transform are applied, followed by quantization,
and symbol coding can be cited as a method for lossy image compression. A
general image compression model is shown in Figure 1.2.
 10

Mapper Quantizer Symbol Symbol Inverse

Coder Decoder Mapper
f(x,y) f(x,y)
f(x,y)

Figure 1.2 General Image Compression Model

Compression is obtained by eliminating any one or more of the below three
fundamental data redundancies:
Coding redundancy:This is presented when the less than the best (that is the
smallest length) code words were used.
Inter-pixel redundancy: This results from the correlations between the pixels of
an image.
Psycho-visual redundancy: This is because of the data which is neglected by the
human visual system (that is, visually not required information).
Image/data Compression attains redundancy for more effective codingas shown in
Figure 1.3.

Source Digital Data Apply

Image redundancy Encoding
reduction in
digital image

Reconstructed
Apply Decoding Perform
digital image
Transmission,
storage and
archiving

Figure 1.3 Shows the Image Compression Effective coding.

 11

1.4 TYPES OF IMAGE COMPRESSION

Image compression can be further classified or divided into two separate

types, such as lossy compression and lossless compression. Lossy compression is
a method of data compression in which the size of the file is reduced by removing
some important data along with the unwanted data in file. In doing
so, image quality is sacrificed to decrease the file size.In this technique, the
compressed image is same as actual/original uncompressed image, yet not exact to
the previous one. This is due to possible loss of a piece of information related to
the image within the compression process. So, they are normally applied for
photographs. The very natural example of the lossy compression is a JPEG.
Whereas in Lossless compression, it compresses an image by encoding it's entire
information from the actual file. Hence, in case if the image is get decompressed
again, then it will be exactly the same as the actual image. Examples of the
lossless technique of image compression are PNG and GIF, i.e., GIF format
images are generally 8-bit images. When a particular image com
pression format is used, which is essentially based on what has been
compressed.
1.4.1 LOSSLESS

The technique of Lossless compression with the compressing of data that is,
when get decompressed, will be the replica of actual data. Here,the binary dataget
compressed. Thesedata are required to be reproduced exactly when they get
decompressed again. On the contrary, the images and the music required need not
be generated 'exactly.' A resemblance of the actual image is sufficient for the most
objective, as far as the error or problems between the actual and compressed image
is avoidable or tolerable.
This type of compression is also noiseless as they never add noise to signal
or image. It is also termed as the entropy coding, as it uses the techniques of
 12

decomposition/statistics to remove/reduce the redundancy. It is also used for some

specific applications along with the rigid needs like medical-imaging. Lossless
compression is usually a two-step algorithm. The first step transforms the original
image to some other format, in which the inter-pixel redundancy is reduced. The
second step uses an entropy encoder to remove the coding redundancy. The
lossless decompression is a perfect inverse process of the lossless compressor
which is shown in Figure 1.4.

Original Transform Entropy

Image Encoding

compressor

Channel

Reconstructed Inverse Entropy

Image Transform Decoding

decompressor

compressor

Figure 1.4 Lossless Compression Techniques

Following methods are involved in lossless compression:

1. Run-length encoding
 13

2. Huffman encoding

3. LZW coding

4. Area coding

1.Run-length encoding:

This is a very simple procedure used for sequential data and is very useful
in case of redundant data. This method replaces sequences of identical symbols
(pixels) which are called as runs by shorter symbols. The run-length code for a
grayscale image is represented by a sequence {V, R}, where V is the intensity of
pixel and R is the number of the consecutive pixel with the intensity V An
example is shown in Figure 1.5 below:

1 1 1 1 0 0 2 2 2 2

{1,4} {0,2} {2,4}

Figure 1.5 Run Length Encoding

The steps of the algorithm for RLE are as follows:.

Step 1: Input the string.

Step 2: From the first symbol or character, give a unique value.

Step 3: Read the next character or symbol; if the character is last in the string, then
exit otherwise.
 14

A: If: the next symbol is the same as the previous symbol, then give the same
unique value as previously. B: Else if: the next symbol is not the same as giving its
new value that is not matched from the previous value.

Step 4: Read and count additional symbols.

Step 5: Go to step 3 until a non-matching value to the not same symbol for
previous.

Step 6: Display the result, that is the count of occurrence of a single symbol, with
that particular symbol.

2. Huffman encoding

This is a common method for coding symbols based on their statistical

occurrence of frequencies (probability). The pixels in the image are treated as
symbols. The symbols that occur more frequently are assigned a smaller number
of bits, while the symbols that occur less frequently are assigned a relatively larger
number of bits. Huffman code is a prefix code. This means that the (binary) code
of any symbol is not the prefix of the code of any other symbol. Most image
coding standards use lossy methods in the earlier stages of compression, and use
Huffman coding as the last step.

The example of Huffman coding with the algorithm is as follows.

Step 1: Input the string

8 3 20 15
l k j h

Step 2: Sorting the data by frequencies

 15

8 3 15 20
l k h j

Step 3: Choose the two smallest frequencies to count.

3 8
k l

Step 4: Merge them together with some of them and update the data.

11

3 8
k l

Step 5: Repeat steps 2, 3, 4.

The final Huffman tree is as follows:

11

20 26
j

11
15
h

3.LZW Coding
3 8
k l is a fully dictionary-based coding. It is divided
LZW (Lempel-Ziv-Welch)
into two subcategories namely static and dynamic. In static, the dictionary is fixed
 16

during the encoding and decoding processes. In dynamic, the dictionary is updated
if the change is needed. LZW compression replaces strings of characters with
single codes. It does not perform any analysis of the incoming text. Instead, it just
adds every new string of characters from the table of strings. The code that the
LZW algorithm outputs can be of any arbitrary length, but it must have more bits
in it than a single character. LZW compression works best for files containing lots
of repetitive data. LZW compression maintains a dictionary. In this dictionary all
the stream entry and code are stored as shown in Figure 1.6.

7 8 3 22 5 7 8 3 9 22 5 10

C1 C2 C1 9 Unique codeC2 10
Unique code C2
C1

Figure 1.6 LZW Coding

LZW codes (C1 and C2) can be achieved by starting the encoding process
with empty dictionary. The first phrase to be entered into the dictionary should
have code values 1, which can be represented using only single bit. The 2nd
and3rd phrases to be entered into the dictionary should have code values 2 and 3
respectivelyand can be represented by 2 bits. Similarly, 3 bits are required for 4th
to 7th phrases to be entered into the dictionary and so on. This is in contrast to
LZW coding, where first 256 phrases entered initially into the dictionary are
 17

represented by 8 bits per phrase. Therefore, significant reduction of number of bits

can be done by optimization of LZW.

The steps in the LZW algorithm are as follows:

Step 1: Input the stream.

Step 2: Initialize the dictionary to contain an entry of each character of the stream.

Step 3: Read the stream, if the current byte is the end of the stream, then exit.

Step 4: Otherwise, read the next character and produce anew code. If a group of
characters is frequently occurring, then give them a unique code.

Step 5: Read the next input character of the stream from dictionary; if there is no
such character in a dictionary, then.

A: Add a new string to the dictionary.

B: Write the new code for the new entered string.

C: Go to step 4.

Step 6: Write out code for encoded string and exit.

4. Area coding

Area coding is an enhanced version of run-length coding, reflecting the

two-dimensional character of images. This is important and more advanced over
other lossless methods. For coding an image, it does not make too much sense to
interpret it as a sequential stream, as it is, in fact, an array of sequences, buildingup
a two-dimensional object. The algorithms for area coding try to find rectangular
regions with the same characteristics. These regions are coded in a descriptive
form as an element with two points and a certain structure. This type of coding can
 18

be highly effective, but it bears the problem of a nonlinear method, which cannot
be implemented in hardware. Therefore, the performance in terms of compression
time is not competitive, although the compression ratio is high.

1.4.2 LOSSY

This technique of Lossy compression decreases the bits by recognizing

unnecessary information and by eliminating it. The system of decreasing the size
of the file of data is commonly termed as the data-compression, though its formal
name is the source-coding, that is coding gets done at the source of data before it
gets stored or sent. In these methods, a few loss of information is acceptable.
Dropping non-essential information from the source of data can save the storage
area. The Lossy data-compression methods are aware of the researches on how
people anticipate data in the question. As an example, the human eye is very
sensitive to slight variations in the luminance as compared to that there are so
many variations in the color. The Lossy image compression technique is used in
the digital cameras to raise the storage ability with the minimal decline of the
quality of the picture. Similarly, the DVDs which use the lossyMPEG-2, Video
codec technique is applied for the compression of the video. In the lossy audio
compression, the techniques of psychoacoustics have been used to eliminate the
non-audible or less audible components of the signal.

Lossy compression, as the name implies, leads to the loss of some

information. The compressed image is similar to the original uncompressed image,
but not just like the previous as in the process of compression since some
information concerning the image might have been lost. They are typically suited
to images. The most common example of lossy compression is JPEG. An
algorithm that restores the presentation to be the same as the original image is
 19

known as lossy technique. Reconstruction of the image is an approximation of the

original image and therefore there is a need for measuring the quality of the image
for the lossy compression technique. Lossy compression technique provides a
higher compression ratio than lossless compression as shown in Figure 1.7.

Original Transform Quantization Entropy

Image Coding

compressor

channel

Reconstruct Inverse Dequantisation Entropy

ed Image Transform Coding
Decompressor

Figure 1.7 Lossy Image Compressions

Types of lossy compression techniques are given below.

1. Transformation coding.
2. Vector quantization.
3. Fractal coding.
4. Block truncation coding.
5. Subband coding.
6. Chroma subsampling.
 20

1.Transformation Coding

DFT and DCT are certain types of transforms used in changing the pixels of
the original image into frequency domain coefficients. There are several properties
in this type of coefficients. One is the compaction property. This is the basis for
achieving the compression.

2.Vector Quantization

This is the method to develop a dictionary of Fixed-sizes vectors called

code vectors. An image is divided into non-overlapping blocks,followed by
determination of each value dictionary and generation of the index thedictionary,
which is used as the encoding for an original input image.

3.Fractal Coding

Fractal coding introduces the idea of decomposition of an image into

segments by using standard methods of image processing like color separation,
edge detection, and texture analysis. Each segment is stored in a library of fractals.
Although the fractal coding method has the advantage of potential high
compression ratio, resolution independence, and fast decoding, its main drawback
is its time-consumption in the encoding process. In order to reduce the
computational complexity of fractal encoding, converting the global search to
local search is an effective way to solve this problem. It mainly consists of
classification techniques and feature vector techniques. For the former ones, the
range blocks and domain blocks are first divided into different categories, and the
block matching is only carried out within the same category.

4.Block Truncation Coding

 21

In this method, firstly the image is divided and then a block of pixels is
arranged. Then, a threshold and reconstruction values are found out for each
block; then, a bitmap of the block is derived, and all those pixels are replaced,
which have the value greater than or equal to the threshold value by 1or 0.

5.Subband Coding

The image got analyzed to produce the components which contain

frequencies of well-defined blocks and sub-bands. Quantization and coding are
applied to each of the bands, and then each sub-band will be designed separately.

5.Chroma subsampling

This method contains the advantage of the human visual system's lower
acuity for color differences. This technique basically used in video encoding, for
example, JPEG encoding and etc. Chromo Sub-sampling is a method that holds
color information of lower resolution and intensity information. Further, the
overwhelming majority of graphics programs perform 2×2 chroma sub-sampling,
which breaks the image into 2×2 pixel blocks and only stores the average color
information for each 2×2 pixel group. The entire procedure is exhibited in Fig 1.8.

2:2
 22

Cb Cr

Figure 1.8 Block diagram of 2× 2-pixel Chroma Sub sampling technique

1.4.3 IMAGE COMPRESSION FILE FORMATS

There are hundreds of image file types of which PNG, JPEG, and GIF
formats are the most often used to display images on the internet.
1.JPEG/JFIF Format

JPEG(Joint photographic experts group) is a compression method. JPEG-

Compressed images are typically stored in the JFIF(JPEG File Interchange
Format) file format.JPEG Compression is of alossytype compression. The
JPEG/JFIF filename is an extension of JPG or JPEG. Every digital camera can
save images in the JPEG/JFIF format. The amount of compression can be
 23

specified, and this amount of compression affects the visual quality of the result as
well.

2.JPEG 2000 Format

JPEG(Joint Photographic Expert Group) is a compression standard enabling

both lossless and lossy storage. The compression methods used are different from
the ones in standard JFIF/JPEG; they improve quality and compression ratios, but
also require more computational power to process.

3.Exif Format

The Exif(Exchangeable image file format)format is a file standard similar

to the JFIF format with TIFF extensions; it is incorporated in the JPEG-writing
software used in most cameras. Its purpose is to record and standardize the
exchange of images with image metadata between digital cameras and editing and
viewing software. The metadata is recorded for individual images and includes
such things as camera settings, time and date, shutter speed, exposure.

4.TIFF Format

The TIFF(Tagged Image File Format)format is a flexible format that

normally saves 8 bits or 16 bits per color for 24-bit and 48-bit totals,
respectively.TIFFs can be lossy and lossless; some offer relatively good lossless
compression for bi-level(black and white) images.TIFF image format is not widely
supported by web browsers.
 24

5.RAW Format

RAW refers to (Raw Image Formats)that are available on some digital

cameras.These formats usually use lossless compression and produce file sizes
which are smaller than the TIFF formats. Most camera manufacturers have their
own software for decoding their raw file format.

6.GIF Format

(Graphics Interchange Format) is limited to an 8-bit palette, or 256 colors.

This makes the GIF format suitable for storing graphics with relatively few colors,
such as simple diagrams, shapes, logos, and cartoon style images. The GIF format
supports animation and is still widely used to provide image animation effects. It
also uses lossless compression.

7.BMP Format

The BMP file format handles graphics files within the Microsoft window
OS. Typically, BMP files are uncompressed. Hence, they are large; the advantages
are their simplicity and wide acceptance in window programs.

8.PNG Format

The PNG(Portable Network Graphics)file format was created as the free,

open-source successor to GIF. The PNG file format supports 8-bit palette images
with optional transparency for all palette colors and 24-bit true-color or 48 bit true
 25

color with and without alpha channel while GIF supports only 256 colors and
single transparent color.

Format Features Disadvantages

TIFF(Tagged Image File The flexible formatsaves 8 Not used in web pages
Format)(lossy and lossless) or 16 per color (RGB) because TIFF files require
totally 24 or 48 bits the large size
GIF(Graphics Interchange Grayscale and black-white It does not work with color
Format) image; it works with 8 bits
per pixel or less, which
indicates 256 or fewer
colors. It states simple
graphics, logos, and
cartoon-style images.
PNG(Portable Network Same 8 bits, 24 bits, and 48
Graphics)(Lossless) bits color image. Provides
motion video compression
that compresses the images,
documents, and video stills.

BMP(Bitmap)(don’t Graphics file related to Large in size, it does not

compress) Microsoft window operating support true colors.
systems, simplicity.BMP
images is binary files.
JPEG(Joint photographic It supports 8 bits grayscale Black and white documents,
expert group)(Lossy) and 24 bits of color images. line art animations
Provides motion video
compression will compress
the images, documents, and
video stills.
 26

RAW(lossless/lossy) File size is smaller than These are not standardized

TIFF Format.Available on images, and it will differ
digital cameras. with manufacturers. So
these images require
manufacture software to
view the images.
Table1.1 Features and its disadvantage

1.4.4 VARIOUS COMPRESSION ALGORITHMS

Data compression a method that takes an input data D and generates the
data C (D) with the lower number of bits as compared to input data. The reverse
process is called decompression, which takes the compressed data C (D) and
reconstructs the data D', as shown in Figure 1.9.

Compression System Decompression

System

D
C(D) D
Figure 1.9 Compression Algorithms

1.JPEG: DCT-Based Image Coding Standard

JPEG enables a compression technique which is able to do compressing
continuous-tone image or data along with a pixel having a depth of the 6 to 24bits
with enough efficiency and speed.

A Discrete Cosine Transform (DCT) describes a fixed series of the data

points in terms of some of the cosine functions fluctuating at various frequencies.
DCTs are vital for various implementations in science and engineering areas,
 27

from the lossy compression of audio. For example, MP3 and the images, and
JPEG, in which the small high-frequency elements maybe get discarded to the
spectral approached for their numerical solution of the partial differential
equations. By using the cosine instead of the sine functions, it is complicated in
these implementations: for the compression,which it returns, finds that the cosine
function is so much effective as mentioned here. Some functions are required to
the exact a typical signal, while for the differential equations, the cosines function
explains a specific selection of the boundary conditions.

JPEG image compression performs in part through rounding off the non-
essential bits of the information. Here is an associated trade-off between the
information loss and the reduction of size. Various number of famous compression
techniques have achieved these intuitive differences, which consists of those that
are used in the music files, video, and images. Hence, the technique of
JPEG'slossy encoding forces to be very prudent with a gray-scale portion of the
image and be very trivial with color.

DCT fragmented the images into parts of separate frequencies in which less
significant frequencies are cancelled through the quantization process. More
significant frequencies are used for retrieving the image during the process of
decompression.

JPEG Process Steps for color images

This section presents jpeg compression steps

 An RGB to YCbCr color space conversion ( color specification )
 The original image is divided into blocks of 8 x 8.
 28

 The pixel values within each block range from[-128 to 127], but pixel
values of a black and white image range from [0-255] so, each block is
shifted from[0-255] to [-128 to 127].
 The DCT works from left to right, top to bottom; thereby, it is applied to
each block.
 Each block is compressed through quantization.
 The quantized matrix is entropy encoded.
 The compressed image is reconstructed through the reverse process. This
process uses the inverse Discrete Cosine Transform (IDCT).

2.Image Compression by Wavelet Transform

For various natural signals, the technique of wavelet transform is a very

efficient tool when compared to the Fourier transform technique. The wavelet
transform technique enables the multi-resolution representation by the use of the
set of analyzing functions, which are translations and dilations of some specific
functions or wavelets. The technique of wavelet transforms is found in various
forms. The complicated sampled form of wavelet transform enables very
sophisticated representation, althoughit has various limitations also.
3.Huffman Algorithm

The general idea in the Huffman encoding algorithm is to allocate very short code-
words to those blocks of input along with the high possibilities. Similarly, the
long code-words are allocated to those who are having the low probabilities.

The Huffman coding process was based on the two observations mentioned below:
 Very frequently found symbols will have the shorter code- words as
compared to the symbol which found less frequently.
 29

 Two symbols which are very less in number may have equal length.

The Huffman code is prepared by combining together the two least possible
characters, and that is repeated in this process as far as there is only one character
is remaining. A code-tree is hence prepared, and then a Huffman code is
generated from the labeling of code tree. It is the best prefix code that is generated
from the set of the probabilities, and which has been used in the different
applications of the compression.

These generated codes are of different lengths of code, which are using an
integral number of the bits. This concept results in a decrease in average length
of the code, and hence the whole size of the compressed data becomes smaller
as compared to the solution to the issue of constructing the codes with less
redundancy.

theoriginal one. The Huffman’s algorithm is the first that

1.5 RESEARCH OBJECTIVES

The present research has been carried out in two stages. The main
objectives of stage 1,2 and 3 are stated below:

 To implement a cost effective method to seamlessly analyze the

image compression techniques.
 To develop the concept of lossless image compression with less
computational complexity using tetragonal and orthogonal
transformation.
The main intention was the compression of image without losing the
details of the image by adopting effective image processing approaches.
 30

1.6 RESEARCH PROBLEMS

A compression problem involves finding an efficient algorithm to remove

various redundancies from certain type of data. The general question to ask here
would be, for example, given a string s, what would be the alternative sequence of
symbols which takes less storage space? The solutions to the compression
problems would then be the compression algorithms that will derive an alternative
sequence of symbols which contains fewer number of bits in total, plus
the decompression algorithms to recover the original string.
How many fewer bits? It would depend on the algorithms, but it would also
depend on how much the redundancy can be extracted from the original data.
Different data may require different techniques to identify the and to remove the
redundancy in the data. Obviously, this makes the compression problems ‘hard’ to
solve because the general question is difficult to answer easily when it contains too
many instances. Fortunately, certain constraints and heuristics can be considered
while designing algorithms.There is no ‘one size fits all’ solution for data
compression problems. In data compression studies, there is a need to analyze the
characteristics of the data to be compressed and hope to deduce some patterns in
order to achieve a compact representation. This gives rise to a variety of data
modeling and representation techniques, which are at the heart of compression
techniques.

1.7 THESIS ORGANIZATIONS

Chapter 1: Introduction: In this chapter, some overviews including

introduction to image compression,motivation, objectives and methodology are
presented.
 31

Chapter 2: Literature Survey: Some literature reviews, reviews of

contribution of other researchers and related works regarding this work are
depicted.

Chapter 3: Research Methodology: In this chapter, a novel unified

technique is proposed for the process of the image compression.

Chapter 4: In this chapter tetrolet compression technique was

implemented for compressing the image.

Chapter 5: In this chapter orthogonal compression technique was

implemented for compressing the image.

Chapter 6: Performance analysis: This chapter provides the detailed

analysis of performance estimation of the proposed techniques.

Chapter 7: Conclusion and future work: This chapter concludes with

important results that have been obtained from the proposed work. This chapter
also suggests further improvements to increase the accuracy

1.8 SUMMARY
Image processing is the study of representation and manipulation of
pictorial information. Digital image processing is performed on digital computers
that manipulate images as arrays or matrices of numbers. High computational
speed, high video resolution, more efficient computer language to process the data,
and more efficient and reliable computer vision algorithms are some of the factors
that let fields such as medical diagnosis, industrial quality control, robotic vision,
astronomy, and intelligent vehicle / highway system to be included as a part of the
large list of applications that use computer vision analysis to achieve their goals.
 32

More and more complex techniques have been developed to achieve new goals
unthinkable in the past.

Machine Vision researchers started using more efficient and faster

mathematical approaches to solve more complicated problems. Convolution
methods, which are widely used in computer vision, can be sped up by using
Fourier Transforms and Fast Fourier Transforms. The idea that an image can be
decomposed into a sum of weighted sine and cosine components is very attractive.
The functions that produce such decompositions are called basis functions.
Examples of basis functions are Fourier and Wavelet Transforms. A wavelet, like
Fourier Transform, has a frequency associated with it, but in addition a scale factor
has to be considered. Only some basis functions produce a decomposition that has
a real importance in image processing. However, the present information
constriction strategies may be far from definitive points of confinement.
Fascinating issues like getting exact models of pictures, ideal portrayals of
such models, and quickly figuring such ideal portrayals are the fantastic
difficulties confronting the information constriction group. The goal of both
lossless and lossy compression techniques is to reduce the size of the compressed
image, to reduce storage requirements, and to increase image transmission speed.
The size of the compressed image is influenced by the compression ratio, with
lossless compression methods yielding ratios of 2:1 to 3:1, and lossy or
irreversible compression having ratios ranging from 10:1 to 50:1 or more. It is
well known that as the compression ratio increases, less storage space is required
and faster transmission speeds are possible, but at the expense of image quality
degradation.
 33

CHAPTER 2

LITERATURE SURVEY

2.1 INTRODUCTION

Image processing is a method to perform certain operations on an image, in

order to get an enhanced image or to extract some useful information from it. It is
a type of signal processing in which input is an image and output may be
an image or characteristics/features associated with that image. In lossy image
compression, the high compression ratio is achieved at the cost of some error in
the reconstructed image. It is always not necessary to obtain the exact image after
reconstruction. In transform coding, initially, DCT was a popular image
compression technique. DCT shows simplicity and satisfactory performance in
compression.

2.2 OVERVIEW OF IMAGE COMPRESSION TETROLET AND

ORTHOGONAL TRANSFORMATION

Dinesh V. Rojatkar et al. (2015) proposed the Digital image compression

technology which emerged from the need for fast communication and live
processing of digital image information over the internet. Time has passed, and a
lot of techniques now are available that reduce the compression ratio and increase
the usability of fast computation. Yet, the approaches are bound to certain limits
which ledto lots of innovative ways that are yet to overcome these limits. Today's
world is very dependent on digital media storage; that's why it requires more
efficient algorithms for image or data compression. Due to reserved bandwidth
and capacity, images need to be compressed and soft-encoded before further using
 34

it in the transmission process. This paper points out Lossy as well as Lossless
compression techniques as used in the fields of image processing.

Ravi et al. (2017) proposed the Image Processing technique to improve

raw images received from cameras placed on satellites, space probes, and aircraft
or pictures taken in normal life for various applications. Basically, all image
processing operations can be grouped into some techniques. These techniques are
Image representation, Image preprocessing, Image enhancement, Image
restoration, Image analysis, Image reconstruction, and Image data compression.
The past studies dealt with the initial processing of raw image data to correct for
geometric distortion, to standardize the data radiometrically, and eliminate noise
present in the data. The aim of the information extraction operations is to replace
visual analysis of the image data with quantitative techniques for automating the
identification of tone in a scene. This involves the analysis of multispectral image
data and the application of statistically-based decision rules for determining the
earth cover via identifying each pixel in an image. The objective of the
categorization process is to classify all pixels in a digital image into several earth
themes. This paper entails the analysis of various image processing techniques and
algorithms. In this paper, different types of the existing method of image
processing techniques are analyzed.

Jyoti A. Patil et al. (2018)made an analysis in which the transmission of a

large amount of image in less time, less bandwidth with better quality without any
information loss are made. For many years, people debated the problem in storage
and transfer. Many people think that lossy image compression is the only solution
to overcome the problem. On the other hand, people have failed to notice that
lossy compression degrades the quality of the image is compressed. Rethinking
approaches have emerged to lossless image compression using Huffman Coding
 35

algorithm and fix quality degradation problem. Replacing the lossy compression
technique gives a better compression ratio without degradation of quality. For
solving the problem, Huffman coding algorithm approach uses lossless image
compression. Using a lossless Huffman coding compression algorithm, the image
is compressed up to 40.08%. Image compression plays a vital role in saving
memory and time, while the transmission Huffman algorithm is comparatively
better in the overall compression technique.

Shaikh et al. (2015)illustrated that the Image compression is one of the

most important steps in image transmission and storage. "A picture is worth more
than a thousand words "is a common saying. Images play an indispensable role in
representing vital in the formation and need to be saved for further use or for
transmission over a medium. In order to have an efficient utilization of disk space
and transmission rate, images need to be compressed. Image compression is the
technique of reducing the file size of an image without compromising the image
quality at an acceptable level. Image compression has been used for a long time,
and many algorithms have been devised. In this paper, the authors have converted
an image into an array using the Delphi image control tool. An algorithm is
created in Delphi to implement the Huffman coding method that removes
redundant codes from the image and compresses a BMP image file (especially
grayscale image), and it is successfully reconstructed. This reconstructed image is
an exact representation of the original,due to the application of lossless
compression technique.

Rachit Patel et al. (2016) proposed the Image Compression using the
Huffman coding technique as a simpler and the easiest compression technique.
Compression of the image is an important task as its implementation is easy and
 36

obtains less memory. The objective of this paper is to analyze Huffman coding
technique which is basically used to remove the redundant bits in data. This is
achieved by analyzing different characteristics or specification like Peak Signal to
Noise Ratio (PSNR), Mean Square Error (MSE) Bits Per Pixel (BPP) and
Compression Ratio (CR) for various input images of different sizes. Further, the
paper shows the devising of a new method of splitting an input image into equal
rows and columns. The final stage sumsup all individual compressed images,
which not only provide better result but also secures the information content. An
image compression technique has various advantages in the field of image analysis
and also for the security purpose for the image.

NehalMarkandeya et al. (2017) Image can be represented with the

minimum number of bits by using image compression. When images are
transferred over the network, it requires space for storage and time to transmit the
image. This research analysis investigates image compression using block
truncation coding. Three algorithms were selected namely, the original Block
Truncation Coding (BTC), Absolute Moment Block Truncation Coding
(AMBTC), and Huffman coding. To justify the objective, a comparison is
performed among these algorithms. Block truncation coding (BTC) and Absolute
Moment block truncation coding (AMBTC) techniques rely on application of
divided image into non-overlapping blocks. They differ in selecting the
quantization level in order to remove redundancy.The input image is divided in the
same rows and columns in Huffman coding, and at the end of each point, all of the
compressed photos are kept healthy, which produce better result outcomes.Here,
the image compression using Huffman coding provides better image quality than
image compression using BTC and AMBTC. Moreover, the Huffman coding is
quite faster compared to BTC.
 37

Anitha et al. (2015) proposed a novel image compression based on the

Huffman encoding and decoding technique. Image files contain redundant and
inappropriate information. Image compression addresses the problem of reducing
the amount of data required to represent an image. Huffman encoding and
decoding is very easy to implement, and it reduces the complexity of memory. A
major goal of this paper is to provide ways of exploring the Huffman coding
technique using MATLAB.

Vikash Kumar et al. (2017)Images include information about the human

body, which is used for different purpose such as medical examination security.
Image compression is used in some applications such as profiling information and
transmission systems. With regard to the importance of images information,
lossless or loss compression is preferred. Lossless compressions are JPEG, JPEG-
LS, and JPEG2000 are few well-known methods for lossless compression.
Differential pulse code modulation can be used for image compression with
Huffman encoder, which is one of the latest techniques which provide good
compression ratio, peak signal to noise ratio, and minimum mean square error.
This paper tries to answer the following question. Which entropy coding,
Huffman, is more suitable compared to others from the compression ratio,
performance, and implementation points of view? The authors implemented and
tested Huffman algorithmsand compared them with other existing methods with
respect to parameter compressionratio, peak signal-noise ratio.

Donapati S et al. (2013) proposed the Image compression algorithms

involving three basic processes viz. Color Space Conversion, Frequency
 38

transformation, and Entropy encoding. The Color Space Conversion (CSC) is one
of the most important processes and can influence the compression and quality of
the image. JPEG XR uses a CSC, which is different from the traditional image
compression algorithms. JPEG XR also supports many image input formats like
RGB, YUV, CMYK, Monochrome, and arbitrary N-channel color formats. JPEG
XR uses the same Frequency Transformation and entropy encoding for all the
input image formats. In the case of JPEG XR, there is a flexibility of using
different CSC for different input image formats like RGB and CMYK. The JPEG
XR bypasses the CSC for images having uncorrelated data like YUV. There are
lot of existing works done on comparison of JPEG XR with other image
compression algorithms like JPEG, but there is very little research or literature on
comparing the image compression ratios and quality of images in different formats
using JPEG XR compression algorithm. In this study, an analysis and comparison
of the compression ratios of the images of different input formats in particular to
RGB input format and YUV 444 format has been carried out to explore the effects
of CSC on the image compression while using the JPEG XR. An analysis of
effective compression (better compression ratio) is also carried out on various
images of unique visual characteristics in different input formats when processed
using JPEG XR.

J. H. Pujar et al. (2010)proposes a novel algorithm for medical image

segmentation based on vigorous smoothening by identifying the type of noise and
edge diction ideology which seems to be a boom in medical image diagnosis. The
main objective of this algorithm is to consider a particular medical image as input
and make the preprocessing to remove the noise content by employing suitable
filter after identifying the type of noise and finally carrying out edge detection for
image segmentation.
 39

Bose, Chaudhuri, and Hocquenghem (BCH) proposed an error correction

and detected algorithm to improve the compression ratio. The proposed approach
applies the BCH algorithms repeatedly until “inflation” is detected. The
experimental results show that the proposed algorithm could achieve an excellent
compression ratio without losing data when compared to the standard compression
algorithms.

Janaki. R et al. (2011)deals with the concept that Image compression is an

important task in the medical imaging system and consists of mechanisms that
reduce storage requirements while maintaining image quality. Medical image
compression can be performed in a lossless or lossy fashion, and is a method that
combines both in the Region of Interest (ROI) techniques. ROI techniques
separate relevant and irrelevant details of an image and apply lossy technique to
irrelevant parts and lossless technique to the relevant part. This paper proposes an
enhanced, active contour-based ROI algorithm to separate the image as
background and foreground. The foreground is compressed using a lossless JPEG
algorithm, and the background is compressed using an enhanced lossy JPEG
algorithm that uses a wavelet neural network followed by a post-processing
algorithm. Experimental results in terms of Compression Rate, Peak Signal to
Noise Ratio, and speed of compression show that the proposed scheme is an
improved version when compared with the traditional algorithm.

UmharaRasool Khan et al. (2017)With the advent of digital information

age, there has been anever-increasing demand of intensive multimedia based web
applications and efficient techniques for the encoding of images and signals. There
are stronger efforts to develop better compression techniques with the sole
objective of producing high compression performance while minimizing the
 40

amount of image data. Image compression is a technique to reduce redundancy

and irrelevancy of image data to enable its efficient transmission and storage. With
the help of image compression techniques, more images can be stored in a given
amount of disk or memory spaceandlossy or lossless image compression can be
preferred, depending upon the application. However, there is degradation in the
quality of a compressed image in comparison to the original image in lossy
compression techniques. Lossless compression techniques do not deliver high
enough compression ratios but exhibit a better quality of reconstructed images in
comparison to lossy techniques. Wavelet algorithm is one of the indispensable
techniques to achieve compression while fulfilling approximately all the
challenges. It enables a multi-resolution analysis of data. In this paper, wavelet-
based compression techniques are studied in detail, and a comparison of
performance is made in terms of image quality metrics viz. PSNR, MSE, BPP, and
Compression ratio. A discrete wavelet transform is easier to implement in contrast
to the continuous wavelet transform. In addition to the above, it also provides
sufficient information for the analysis and synthesis of images with a significant
reduction in the computation time. In this paper, Haar, Daubechies, Biorthogonal,
Coiflets, and Symlet wavelets have been employed to an image, and thereof results
obtained are analyzed in terms of image quality metrics. Further, a comparison of
the image quality has been performed by using four wavelet-based encoding
algorithms, namely Set Partitioning in Hierarchical Trees (SPIHT), Embedded
Zero Tree Wavelet (EZW), Wavelet Difference Reduction (WDR) and Adaptively
Scanned Wavelet Difference Reduction (ASWDR). This paper guides the
developers to choose a better wavelet compression system for their particular
applications.
 41

Sonja Grgic et al. (2001)examines a set of wavelet functions (wavelets) for

implementation in a still image compression system and to highlight the benefit of
this transform relating to today's methods. The paper discusses important features
of wavelet transform in compression of still images, including the extent to which
the quality of the image is degraded by the process of wavelet compression and
decompression. Image quality is measured objectively, using peak signal-to-noise
ratio or picture quality scale, and subjectively, using perceived image quality. The
effects of different wavelet functions, image contents, and compression ratios are
assessed. A comparison with a discrete-cosine-transform-based compression
system is given. The results provide a good reference for application developers to
choose a good wavelet compression system for their application.

Vijaya Kumar et al. (2017) deals with the main objective of source coding
to represent the symbols or messages generated from an information source in a
suitable form so that the size of the data is reduced. In image compression, JPEG
is normally used, where huge number of zeros are generated in medium and high
frequency region of the transformed image using the combination of DCT
(Discrete Cosine Transform) and quantization. This is required to reduce the run-
length coding of an image. The process is lossy compression but provides good
illusion at a glance. In this paper, the Haar wavelet for image compression is
adopted. Haar Transformation is employed with an idea to minimize the
computational requirements by applying different compression thresholds for the
wavelet coefficients, and these results are obtained in a fraction of seconds and
improves the quality of the reconstructed image. The results show that the
reconstructed images have high compression rates and better image quality.
 42

S. Sridhar et al. (2014) proposed a solution for the vital problem in

evaluating the picture quality of an image compression system in describing the
amount of degradation in the reconstructed image. The study elaborates that
Wavelet transforms are a set of mathematical functions that have established their
viability in image compression applications owing to the computational simplicity
that comes in the form of filter bank implementation. The choice of wavelet family
depends on the application and the content of the image. The proposed work is
carried out by the application of different hand-designed wavelet families like
Haar, Daubechies, Biorthogonal, Coiflets, and Symlets, etc. on a variety of
benchmark images. Selected benchmark images of choice are decomposed twice
using an appropriate family of wavelets to produce the approximation and detail
coefficients. The highly accurate approximation coefficients so produced are
further quantized, and later Huffman is encoded to eliminate the psychovisual and
coding redundancies. However, less accurate detailed coefficients are neglected. In
this paper, the relative merits of different Wavelet transform techniques are
evaluated using objective fidelity measures such as PSNR and MSE. Results
obtained provide a basis for application developers to choose the right family of
wavelet for image compression matching their application.

Ahmed A. Nashat et al. (2016)proposed the Discrete Wavelet Transform

(DWT), which is known to be one of the best compression techniques. It provides
a mathematical way of encoding information in such a way that it is layered
according to the level of detail. In this paper, the authors used Haar wavelets as the
basis of transformation functions. Haar wavelet transformation is composed of a
sequence of low pass and high pass filters, known as a filter bank. The redundancy
of the DWT detail coefficients is reduced through thresholding and further through
Huffman encoding. The proposed threshold algorithm is based upon the statistics
 43

of the DWT coefficients. The quality of the compressed images has been evaluated
using some factors like the Compression Ratio (CR) and Peak Signal to Noise
Ratio (PSNR). Experimental results demonstrate that the proposed technique
provides a sufficiently higher compression ratio compared to other compression
thresholding techniques.

Jens Krommweh (2010)discussed getting an efficient image

representation by introducing a new adaptive Haar wavelet transform, called
Tetrolet Transform. The implemented filter bank algorithm is simple but
enormously effective. In every level of the filter bank algorithm, the low-pass
image is divided into 4×4 blocks. Then, a local tetrolet basis is determined in each
block, which is adaptable to the image geometry in this block. An analysis of the
adaptivity costs leads to modified versions of this method. Numerical results show
the strong efficiency of the tetrolet transform for image compression.

B. Niteesh et al. (2016)discuss the Tetrolet transform which is a wavelet-

based efficient and effective transform, and which utilizes supports formed by
connecting four identical squares known as tetrominoes such that each is
connected to at least one other square along its boundary. Haar wavelets are
defined on these tetrominoes so as to form a localized orthonormal basis. The
procedure is applied on small 4×4 partitions of the low pass image and the sparsest
covering from each partition is stored the non-redundancy in the wavelet basis
result in sparse image representation. Consequently, a high compression ratio is
achieved after the application of a wavelet shrinkage procedure on the Tetrolet
coefficients. In order to reduce the processing cost, certain modifications in the
original Tetrolet transform are also discussed. Tetrolet transform is an effective
 44

technique that uses tetrominoes for image compression with less edge blurring.
The reduction in file size allows more images to be stored in a given amount of
disk or memory space.

AvinashGhorpade et al. (2014)illustrates that the Wavelet Transform has proved

to be a very useful tool for image processing in recent years. Digital images
require large amounts of memory to store and when retrieved from the internet,
can take a considerable amount of time to download. Compression makes it
possible for creating file sizes of manageable, storable, and transmittable
dimensions. The Haar wavelet transform provides a method of compressing image
data so that it takes up less memory. The most distinctive feature of Haar
Transform lies in the fact that it lends itself easily to simple manual calculations.
Modified Fast Haar Wavelet Transform is one of the algorithms which can reduce
the calculation work in Haar Transform. This paper attempts to describe the
algorithm for image compression using the Modified Fast Haar Wavelet
Transform (MFHWT).

HidayahRahmalan et al. (2010) came out with an analysis about the Orthogonal
moment, which is known as better moment functions compared to the non-
orthogonal moment. Among all the orthogonal moments, Tchebichef Moment
appears to be the most recent moment function, that still attracts interest among
the computer vision researchers. The author proposes a novel approach based on
discrete orthogonal Tchebichef Moment for efficient image compression. The
image compression is useful in many applications, especially related to images
that are needed to be seen in small devices such as in mobile phones.Meanwhile,
 45

the method incorporates simplified mathematical framework techniques using

matrices, as well as a block-wise reconstruction technique to eliminate possible
occurrences of numerical instabilities at higher moment orders. In addition, a
comparison between Tchebichef Moment compression and JPEG compression is
performed and the result shows significant advantages of Tchebichef Moment in
terms of its image quality and compression rate. Tchebichef Moment provides
more compact support to the image via sub-block reconstruction for compression.
Tchebichef Moment compression is able to perform potentially better for a broader
domain on real digital images and graphically generated images.

NurAzman Abu et al. (2010)introduces the embedding watermark scheme along

the edge based on the concept of psychovisual threshold. This paper will
investigate the sensitivity of minor changes in DCT coefficients against JPEG
quantization tables. Based on the concept of psychovisual threshold, there are still
deep holes in JPEG quantization values to embed a watermark. This paper locates
and utilizes them to embed a watermark. The proposed scheme has been tested
against various non-malicious attacks. The experiment results show the watermark
is robust against JPEG image compression, noise attacks and low pass filtering.

J. S. Rivera-López et al. (2017)mention that the Image reconstruction can help to

determine how well an image may be characterized by a small finite set of its
moments. Also, the number of descriptors needed to describe an image can be
identified. In this work, a comparative analysis using a different set of discrete
orthogonal moments in terms of normalized image reconstruction error (NIRE) is
presented. Color image reconstruction is performed with different colorchannels
 46

and various orders of different discrete orthogonal moments. Finally, the results
obtained by the reconstruction of three color images with different families of
orthogonal moments and error analysis to compare their capacity of descry

MacarenaBoi et al. (2010),described a method for image compression . In the

Wavelet Transform technique, the coefficients below a certain threshold are
removed. A new type of global threshold to improve the wavelet compression
technique is proposed. The aim is to maintain the retained energy and to increase
the compression ratio with respect to other global thresholds commonly used.

AbasiJulius(2016) Medical image compression is very useful and has a great

significance nowadays for data storage and easy transfer of data or images
between two points. In order to transfer images, first it was compressed at the
transmitting point, and upon having reached the destination, the decompression
process is done to reconstruct the original image or an approximate of it. In this
paper, the compression of medical images by using Discrete Haar Wavelet
Transform (DWT) is proposed as the best way. Much research has been done in
image compression techniques of which Discrete Cosine Transform, Gaussian
Pyramids, Block Truncation Coding, and Singular Value Decomposition are
notable.

Paulo A. M. Oliveira et al. (2015)introduceda low-complexity approximation for

the discrete Tchebichef transform (DTT). The proposed forward and inverse
transforms are multiplication-free and require a reduced number of additions and
bit-shifting operations. Numerical compression simulations demonstrate the
 47

efficiency of the proposed transform for image and video coding. Furthermore,
Xilinx Virtex-6 FPGA based hardware realization shows a 44.9% reduction in
dynamic power consumption and 64.7% lower area when compared to the
literature.

Anju B (2012)demonstrates that the image compression plays an important role in

the compression of medical images. Medical images are mainly used for the
diagnosis of diseases and surgical planning. Medical images are usually stored
digitally. Medical Image compression plays an important role in telematics,
especially in telemedicine. It is necessary that medical images need to be
compressed for reliable transmission. In the medical image compression, diagnosis
is effective only when the compressed image preserves all the information of the
original image. This results in a lossless compression technique. While lossy
compression techniques are more efficient in terms of storage and transmission
needs, there is no warranty that they can preserve the characteristics needed in
medical image processing and diagnosis. Compression plays an important role as
medical imaging moves to filmless imaging.CT or MRI medical imaging is used
nowadays, which can produce pictures of the human body in digital form. The
lifting scheme is used for the design of both orthogonal and bi-orthogonal filters. It
is implemented for the orthogonal filters using two lifting steps. The performance
of the proposed filters is then compared with the conventional filters in terms of
compression ratio, PSNR, etc. The bi-orthogonal filters are implemented using the
SPIHT algorithm replacing the EZW coding for image compression. SPIHT
algorithm requires fewer bits to capture the same amount of information when
compared with the proposed EZW method.
 48

Hameed et al. (2016) deals with the idea of the evolutionary optimized
coefficients of discrete orthogonal Tchebichef moment transform (TMT). The
objective of using this transform in this study is to ameliorate the quality of the
traditional moment-based image compression methods. Most of the existing
methods compute moment-transform coefficients for the input image and then
select the coefficients sequentially downward to a certain order, based on the
desired compression ratio. However, the proposed method divides the input image
into nonoverlapping square blocks of a specific size in order to circumvent the
problem of numerical instability and then computes the TMT coefficients for each
block. In this work, a real-coded genetic algorithm is employed to optimize the
TMT coefficients of each block, which produces reconstructed images of better
quality for the desired compression ratio. Here, the optimization is carried out by
minimizing the mean square error function. Standard test images of two different
sizes (128×128 and 256 ×256) have been subjected to the proposed compression
method for the block sizes (4×4 and 8×8) in order to assess its performance. The
results reveal that the proposed real-coded genetic algorithm-based method
outperforms others, namely the conventional sequential selection method and
simple random optimization method, for the chosen input images in terms of the
task of compression.

Ranjan K. Senapati (2010) illustrate that the Discrete Tchebichef Transform

(DTT) based on orthogonal Tchebichef polynomials can be an alternative to
Discrete Cosine Transform (DCT) for JPEG image compression standard. The
properties of DTT are not only very similar to DCT, but also have higher energy
compactness and lowers computational advantage using a set of the recurrence
relation. Through extensive simulation, image reconstruction accuracy (Peak
 49

Signal to Noise Ratio), and the number of bits required to encode the coefficients
for both DCT and DTT are verified. It has been demonstrated that DTT requires a
lesser number of bits to encode the coefficients than DCT for a given compression
ratio.

RanjanSenapati et al. (2013)presents an efficient orthogonal sparse 8×8

transform matrix for color image compression, particularly at lower bit rate
applications. The transform matrix is made sufficiently sparse by appropriately
inserting additional zeros into the matrix proposed by Bouguezel. The algorithm
for fast computation is also developed. It is shown that the proposed transform
matrix provides a 7% reduction in computation over the matrix by Bouguezel,, and
45% over signed discrete cosine transform (SDCT). By using various natural test
images, it is shown that the rate-distortion performance is comparable with the
matrix proposed by Bouguezel and approximated DCT at low bit rates. Further, it
outperforms SDCT by a large margin almost at all bit rates. 

G. A. Papakostas et al. (2005)A new method for extracting feature sets with
improved reconstruction and classification performance in computer vision
applications is presented in this paper. The main idea is to propose a procedure for
obtaining surrogates of the compressed versions of very reliable feature sets
without affecting their reconstruction and recognition properties significantly. The
surrogate feature vector is of lower dimensionality and thus there are more
appropriate for pattern recognition tasks. The proposed Feature Extraction Method
(FEM) combines the advantages of the multiresolution analysis, which is based on
the wavelet theory, with the high discriminative nature of Zernike moment sets.
 50

The resulted feature vector is used as a classification feature in order to achieve

high recognition rates in a typical pattern recognition system. The results of the
experimental study support the validity and strength of the proposed method.

AshlinDeepa et al. (2014) The Character recognition is one of the most important
areas in the field of pattern recognition. Recently, Indian Handwritten character
recognition is getting much more attention, and researchers are contributing a lot
in this field. But Malayalam, a South Indian language, has very fewer works in
this area and needs further attention. Malayalam OCR is a complex task owing to
the various character scripts available and, more importantly, the difference in
ways in which the characters are written. The dimensions are never the same and
maybe never mapped on to a square grid, unlike English characters. The selection
of a feature extraction method is the most important factor in achieving high
recognition performance in character recognition systems. Different feature
extraction methods are designed for different representations of characters. As an
important component of pattern recognition, Feature Extraction has been paid
close attention by many scholars, and currently it has become one of the research
hot spots in the field of pattern recognition.

Robby Candra et al. (2017)mention that the continuous growth of

multimedia data capacity, and the large capacity of communication lines had
always become the obstacle in the communication of multimedia data. To
overcome this, the data compression process in a shorter time is required. One part
of the JPEG algorithm, that is able to determine compression ratio and
compression process speed, is the Zigzag Coding or Zigzag Scan algorithm. This
 51

research is focused on developing the zigzag scan algorithm with the mapping
method. Zigzag Scan with mapping method is a sorting process of DCT-quantized
data result according to the position sequence determined in a zigzag. The study
exhibits the implementation of the Zigzag Scan mapping method into FPGA using
ROM that serves as zigzag address generator, and RAM that serves read and write
data according to a Zig-zag address generated by ROM. The efficiency of the Zig-
zag Scan with the mapping method has been successfully developed. It is able to
accelerate the sorting process of the DCT-quantized coefficients period because
input data can be immediately located in sequence position, which has been
determined without any value comparison and repetition process. Zig-zag Scan
with mapping method process time is 250 MHz or approximately 4ns per byte data
(12 ns per pixel) with the delay time (latency) of 64 clocks. It means that the
generated IC Zig-zag Scan prototype can be operated in realtime JPEG/MPEG
compression with a maximum of 3 megapixels per frame for video with 25fps
(force per second) speed. The generated IC Zig-zag Scan component (IP Core)
needs ten slices of Flip-flop (30 times less than Arafa method and 2 times less than
Ketul method) and LUT (look up table) as many as 39 slices (20 times less than
Arafa method and 2 times less than Ketul method).

Walaa M. and Abd-Elhafiez et al. (2012)presents the performance of different

block-based discrete cosine transform (DCT) algorithms for compressing the color
image. In this RGB component of a color, images are converted to YCbCr before
DCT transform is applied. Y is the luminance component while Cb and Cr are
chrominance components of the image. The modification of the image data is done
based on the classification of image blocks to edge blocks and non-edge blocks.
Then, the edge block of the image is compressed with low compression, and the
 52

nonedge blocks are compressed with high compression. The analysis results
indicate that the performance of the suggested method is much better, where the
constructed images are less distorted and compressed with a higher factor.

Gunapriya et al. (2014) present the compression of color medical images with
different color spaces. Even though multimedia data storage and communication
technologies have attained rapid growth, compression of color medical images
remains a challenging task. In the proposed method, color medical images are
converted to different color spaces such as YCbCr, NTSC and HSV. Then,
decomposition of different color space images is done using the curvelet
transform. The decomposed images are then compressed using the Huffman
coding. The results obtained for different color spaces are compared in terms of
compression ratio and bits per pixel.

Amit Kumar Mandal et al. (2013)In this experimental study, a system has been
developed that uses threshold values to segment an image. An image is treated as a
one-dimensional array of pixel values. In this work, the segmentation is performed
in the YCbCrcolor space. In this experiment, the segmented image will have 2
different colors, which are black and white, and for this reason, the segmentation
is done using local thresholding value for the Cb component of YCbCr. A mask is
used to determine the neighbors of each pixel in the image. The mask also
determines an operation to be applied to the neighborhood of every pixel in the
image. Now the mask and the operations are used to determine the local threshold
for every pixel in the image. For each pixel location, the threshold will be
different. This value is compared with the color value of the pixel. If the value of
 53

the pixel in this location is greater than or equal to the specified threshold for the
pixel, then it is labeled as 1. Otherwise, if it is smaller, then it is labeled as 0. In
this way, an image with two color values is achieved.

Abdul-Wahab Sami Ibrahim and Hind JumaaSartep

(2017)proposed a colouring algorithm for adding colour to a grayscale image and
video that produces a vibrant and realistic colour image. In general, colorizing
gray image involves grayscale image, colour space and source colour image. The
problem of colouring grayscale image is clearly under-constrained, as previous
approaches have either relied on significant user interaction in their proposed
colouring algorithm. They tried to reduce the human labour needed in manual
colouring of a grayscale image, as human interaction is needed only to find a
source colour image. The proposed algorithm then performs the job of transfer
colour traits from source colour image to grayscale image. The proposed algorithm
of colour traits transfer to the grayscale image has been performed by using two
different colour spaces together in the colouring process. These colour spaces are
YCbCr and HSV. With the generation of colour palette for the colour source
image and grayscale image, colour transfer depends on the comparison between
the colour palettes of the colour source image and target grayscale image along
with different pixel window size start from (2 x 2), (3 x 3) up to (10 x 10).

DeepikaChoudhary et al. (2013)introduce the Image Processing as

a technique to enhance raw images received from cameras/sensors placed on
satellites, space probes, and aircraft or pictures taken in normal day-to-day life for
various applications. Image processing has a number of applications such as
 54

Remote Sensing, Medical Imaging, Non-destructive Evaluation, Forensic Studies,

Textiles, Material Science, Military, Film industry, Document processing, Graphic
Arts, Printing Industry, etc.

H.B. Kekre et al. (2013)discussed the image compression using a hybrid wavelet
transform. A hybrid wavelet transform matrix is generated using two-component
transform matrices. One of the component transform matrices contributes to global
properties, whereas the second one contributes to local properties of an image.
Different sizes of component transform matrix can be used to generate hybrid
transform matrix so that its size is same as size of the image.Different colour
images of size 256×256 are used for experimentation. The proposed hybrid
wavelet transform is applied on red, green and blue planes of image separately.
Then, in each plane transformed coefficients are sorted in descending order of
their energy and the lowest energy coefficients are eliminated. Root mean square
error between original image and reconstructed image is calculated to check the
performance at different compression ratios. By varying the size of pair of
component transform matrices, hybrid transform matrix is constructed and results
are observed.Also, by changing the component matrix which contributes to local
properties of image and with size variation, results are observed and compared. It
has been observed that if more focus is given on local features, then better results
can be produced due to the implementation of the Hybrid Wavelet Transform.
Focusing on local features can be done by selecting larger size of orthogonal
component transform that contributes to local properties.
 55

Manjari Singh et al. (2016)illustrate thatthe Image compression is an

implementation of the data compression which encodes actual image with some
bits. Thepurpose of the image compression is to decrease the redundancy and
irrelevance of image data to be capable to record or send data in an effective form.
Hence, the image compression decreases the time of transmit in the network and
raises the transmission speed. In Lossless technique of image compression, no data
get lost while doing the compression. To solve these types of issues, various
techniques are used for the image compression. For compressing the image
commonly, two types’ of approaches are explained, called as lossless and the lossy
image compression approaches. These techniques are easy in their applications
and consume very little memory. An algorithm has also been introduced and
applied to compress images and to decompress them back, by using the Huffman
encoding techniques.

Athira B. Kaimal et al. (2013)deals witha fact that the increasing attractiveness
and trust on digital photography have given rise to their use for visual
communication. They require storage of large quantities of data. Due to limited
bandwidth and storage capacity, images must be compressed before storing and
transmitting. Many techniques are available for compressing the images, yet, in
some cases these techniques will reduce the quality and originality of image. As a
result of compression, some traces like blocking artefacts, and transformation
fingerprints are also introduced into the reconstructed image. It mainly affects the
medical imaging by reducing the fidelity and thereby introducing diagnostic
errors. Many hybrid techniques are also developed to overcome these problems.
This paper addresses various compression techniques as they are applicable to
various fields of image processing.
 56

Mehala et al. (2013) proposed the New Image Compression Algorithm Using
Haar Wavelet Transformation technique. The proposed 8×8 transform matrix can
be obtained by appropriate inserting of some 0’s and 1/2’s into the Haar Wavelet.
The proposed Haar Wavelet algorithm is based on integers, and made sufficiently
sparse orthogonal transform matrix. A Haar Wavelet algorithm for Fast
computation is developed, besides calculating various measures like Compression
Ratio, PSNR, Threshold Value and Reconstructed Normalization. The proposed
algorithm has been implemented in Mat Lab.

KamrulHasanTalukde et al. (2007) trace the increasing growth of technology

and the entrance into the digital age. The study analyses the concern in handling of
a vast amount of information every time which often presents difficulties. So, the
digital information must be stored and retrieved in an efficient and effective
manner, in order to be put to practical use. Wavelets provide a mathematical way
of encoding information in such a way that it is layered according to level of
detail. This layering facilitates approximations at various intermediate stages.
These approximations can require more space to store. Here, a low complex 2D
image compression method using wavelets as the basis functions and the approach
to measure the quality of the compressed image are presented. The simplest
wavelet form namely the Haar Wavelet is the particular waveletchosen and used
here. The 2D Discrete Wavelet Transform (DWT) has been applied and the detail
matrices from the information matrix of the image have been estimated. The
reconstructed image is synthesized using the estimated detail matrixproduced by
the Wavelet transform. The quality of the compressed images has been evaluated
 57

using some factors like Compression Ratio (CR), Peak Signal to Noise Ratio
(PSNR), Mean Opinion Score (MOS), Picture Quality Scale (PQS) etc.

B. Penna et al. (2007)deal with the point that the Transform-based lossy
compression has a huge potential for hyperspectral data reduction. Hyperspectral
data are 3-D, and the nature of their correlation is different in each dimension.
This variation calls for a careful design of the 3-D transform to be used for
compression. In this paper, the transform design and rate allocation stage for lossy
compression of hyperspectral data are examined. Firstly, a set of 3-D transforms
is selected, which is obtained by combining various ways wavelets, wavelet
packets, the discrete cosine transform, and the Karhunen-Loegraveve transform
(KLT). Then, the coding efficiency of these combinations is evaluated. Secondly,
a low-complexity version of the KLT is proposed, in which complexity and
performance can be balanced in a scalable way, allowing one to design the
transform that better matches a specific application. Thirdly, both these
algorithms, as well as other existing transforms are integrated in the framework of
Part 2 of the Joint Photographic Experts Group (JPEG) 2000 standard, by taking
advantage of the high coding efficiency of JPEG 2000, and exploiting the
interoperability of an international standard. An evaluation framework based on
both reconstruction fidelity and impact on image exploitation is introduced, to
evaluate the proposed algorithm by applying this framework to AVIRIS scenes. It
is shown that the scheme based on the proposed low-complexity KLT significantly
outperforms previous schemes as to rate-distortion performance. As for impact on
exploitation, multiclass hard classification, spectral unmixing, binary
classification, and anomaly detection are considered as benchmark applications.
 58

Ghadah Al-Khafaji et al. (2017)analyzed that the lossy based image compression
of has become an increasingly intensive research area. This is due to importance
given to daily visual media applications, including TV, video film, the internet etc.
These applications are basically based on losing some unrecognized or unwanted
information, and managing non-noticeable distortion quality changes is traded off
against high compression ratios. A lossy image compression is introduced, based
on utilizing three techniques of wavelet, polynomial prediction and block
truncation coding, in which each technique exploited according to redundancy
presence. The test results showed promising performance achievement in terms of
higher compression with lower noticeable error or degradation.

Mahinderpal Singh et al. (2014)illustrate that the image Compression is a

method, which reduces the size of the data or the amount of space required to store
the data. Digital images, in their raw form, require an enormous amount of storage
capacity. There are various transformation techniques used for data compression.
Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) are the
most commonly used transformations.The Discrete Cosine Transform (DCT) is a
method for transforming a signal or image from spatial domain to frequency
component. DCT has high energy compaction property and requires less
computational resources. On the other hand, DWT is a multi-resolution
transformation. In this paper, a mixed DWT-DCT algorithm for image
compression and reconstruction is proposed taking benefit from the advantages of
both algorithms.
 59

Sreenivasulureddy et al. (2007describes the algorithms for Image

Compression using Transform Coding Methods viz. Modified Hermite Transform
(MHT), Discrete Cosine Transform (DCT) and Wavelet Transform (WT).In MHT
and DCT, the given image is divided into NxN sub-blocks and transformation is
applied to each block.Then, a threshold value is selected such that to minimize the
mean square value between the Original image and the Reconstructed image.
Wavelet Transform decomposes signal into a set of basis functions. These basis
functions, also called as wavelets, are obtained from a single prototype wavelet by
dilations and contractions as well as shifts. By using Discrete Time Wavelet
Transform, any given image can be decomposed into ‘n’ levels. The lower levels
of decomposition correspond to higher frequency sub bands. In particular, level 1
corresponds to highest frequency sub-band and it is the finest level of resolution.
Correspondingly, nth level corresponds to the lowest frequency sub-bands and has
the coarsest resolution.In this work, different wavelet Transforms like Symlet,
Haar, Daubechies and Coifletare used to compress the 1-D,2-D and color images
and the results are compared with DCT and MHT results.

NidhiSethi et al. (2011) discuss that thecompressing an image is significantly

different than compressing raw binary data. General purpose compression
programs can be used to compress images, but the result is less than optimal. This
is because images have certain statistical properties which can be exploited by
encoders specifically designed for them. Also, some of the details of the image can
be sacrificed for the sake of saving a little more bandwidth or storage space. This
also means that lossy compression techniques can be used in such areas. The
discrete wavelet is essentially sub band–coding system and sub-band coders have
been quite successful in speech and image compression. In this paper HAAR
 60

Wavelet Transform is implemented. The results in terms of PSNR(Peak Signal

Noise Ratio) and MSE (Mean Square Error)show that the Haar transformation can
be used for image compression. The quantization is done by dividing the image
matrix into blocks and taking the mean of the pixel in the given block. It is clear
that DWT has potential application inthe compression problem and the use of Haar
transform is ideally suited.

ShabanamTamboli et al. (2013) proposes a simple but efficient calculation

scheme for 2D-Haar wavelet transformation in image compression. The proposed
work is aimed at developing computationally efficient and effective algorithms for
lossy image compression using wavelet techniques. The work is particularly
targeted towards wavelet image compression using Haar Transformation with an
idea to minimize the computational requirements by applying different
compression thresholds for the wavelet coefficients. Results are obtained in
fraction of seconds enabling to improve the quality of the reconstructed image.
The promising results obtained concerning reconstructed images quality as well as
preservation of significant image details achieved high compression rates and
better image quality.
 61

CHAPTER 3

A SIMPLE UNIFIED TECHNIQUE FOR COMPRESSION OF

IMAGES

3.1 OBJECTIVES

A fundamental goal of lossy image compression is to reduce the bit rate,

save time in transmission, and save storage of memory of digital images while
maintaining an acceptable fidelity or image quality.Spatial encoding is probably
the most well-known and the most intuitive coding method.Transform coding is
used to convert spatial image pixel values into transform coefficient values. Since
this is a linear process and no information is lost, the number of coefficients
produced is equal to the number of pixels transformed. A mixed approach is
discussed in this chapter that combines the features of the Transform and the
Spatial coding technique. It is a simple yet efficient algorithm. The suitability of
the algorithm is tested for both the textured and the un-textured images. The
performance measures such as the Compression Ratio (CR), Maximum
Compression Achieved (MCA), and Signal to Noise Ratio (SNR) are evaluated.

3.2 INTRODUCTION

Image compression techniques fall into two broad categories. They are
information preserving (lossless compression), and lossy compression. A simple
purpose of loss image compression is to minimize the bit rate (i.e. the amount of
bits required to process each pixel). This would conserve time and capacity in
digital image processing while preserving an reasonable image quality or quality.
 62

The system of decreasing the size of the file in data is commonly termed as the
data-compression, though its formal name is the source-coding, that is coding
done at the source of data before it gets stored or sent. In these methods, a few loss
of information is acceptable. Dropping non-essential information from the source
of data can save the storage area. The Lossy data-compression methods are aware
of the researches on how the people anticipate data is the question. As an example,
the human eye is very sensitive to slight variations in the luminance as compared
to so many variations in the color. The Lossy image compression technique is used
in the digital cameras to raise the storage ability with the minimal decline of the
quality of the picture. Similar is the case of DVDs, which use the lossy MPEG-2
Video codec technique for the compression of the video. In the lossy audio
compression, the techniques of psychoacoustics have been used to eliminate the
non-audible or less audible components of the signal. There are many algorithms
developed for image data compression. Each Algorithm is based on a basic
principle of redundancy reduction of the given image. Based on this principle, the
algorithm can be divided into two basic techniques, namely, Spatial Coding
Technique and Transform Coding technique. Spatial Coding Technique operates
directly on the pixels of an image.

Transform Coding is a lossy coding scheme based on the utilization of

correlation between inter-pixels. In this Transformation coding technique, various
transforms are used to modify the value of image pixel into frequency domain
coefficients. These transformed coefficients have various desirable properties.
One of the significant properties of DWT is the energy compaction that most of
the energy of the natural image is represented by the transform coefficients. For
achieving the higher image compression, only few transform coefficients are
occupied, and the exclusive coefficients are discarded. These engaged coefficients
 63

are proposed for further quantization practice and entropy encoding. The decoder
reconstructs the original image by applying the inverse transform.Transform
Coding is used to map the image into a set of transform coefficients, which are
then quantized and coded.

As shown in Figure3.1, this prediction – transformation – decomposition process

is completely reversible. There is a loss of information due to the process of
quantization. The entropy coding after the quantizing is lossless. When the
decoder has input data, entropy decoding is applied to compressed signal values to
get the quantized signal values. Then, de-quantization is used on it, and the image
is recovered, which resembles the original.

Original Data Prediction/ Quantization

Transformation/
Decomposition

Compressed Data
Lossless Coding

Figure 3.1: Lossy Image Compression

Lossy compression methods include some basic consideration (performance-wise):

 64

1. Speed of encoding and decoding

2. Compression ratio
3. SNR ratio

Lossy compression includes the following methods:

1. Block truncation coding

2. Code Vector quantization
3. Fractal Coding
4. Transform Coding
5. Sub-band coding

Commonly used frequency-domain transforms include the Discrete Wavelet

Transform (DWT) and the Discrete Cosine Transform (DCT). Discrete Sine
Transform is used to alter the pixel specifications from the spatial domain into the
frequency domain. One is the energy compaction property; some few coefficients
only have the energy of the original image signal that can be used to reproduce
itself. Only those few significant coefficients are considered, and the remaining is
discarded. These coefficients are given for quantization and encoding. DCT
coding has been the most commonly used in the transformation of image
data.DWT yields excellent spatial localization and multi-resolution characteristics.
DCT offers good compression whereas DWT offers good scalability. The idea of
combining two transforms is based on the fact that combined transforms could
compensate the drawbacks of each and yield a good Compression Ratio and high
image quality. DCT shows simplicity and satisfactory performance in
compression. It introduces blocking artefacts specifically at the low bit rate. This
drawback was overcome by wavelet transform. Since the last two to three decades,
 65

wavelets have come into picture and have become attractive techniques for image
compression, as they give time and frequency analysis of data. Wavelet transform
can be directly applied to the whole image without blocking it. Wavelet-based
Coding is more robust under transmission and decoding errors. The Multi-
resolution property of Wavelet transforms helps to view the image at different
scales. A recent trend is to use a hybrid technique for image compression in which
one transform is combined with another transform to incorporate the advantages of
both transforms.

In recent research, better quality of compression is observed with the use of

Wavelet transforms and hybrid Wavelet transforms. Hybrid based image
compression is proven to be better with a few hybrid wavelets transforms. The
Hybrid Wavelet Transform (HAB) will be generated from a combination of two
orthogonal transforms. Comparison of compression using different Transform
techniques like DCT and Kekre, Kekre and Haar, Kekre and DCT, Kekre and
Walsh, DCT, and Walsh, DCT, and HAAR reveals that the Mean Square Error
evaluated for all combinations yield the proof that DCT and Kekre is the best
combination. The Hybrid wavelet transform is orthogonal, noninvolutional, and it
gains positive traits from both the orthogonal transforms. In addition, its extended
performance comparison of HWT for image compression is tested with varying
proportions of constituent transforms.

In this work, the Transform technique is implemented using HaarWavelet,

and the spatial technique is implemented using the Average Window Technique.
There are two methods to design simple and efficient coding algorithms. One is to
modify a simple coding algorithm by keeping its simplicity while improving its
efficiency; the other is to modify an efficient coding algorithm by keeping its
 66

efficiency while reducing its complexity. The Mixed Approach discussed in this
chapter is a combination of the Transform and Spatial coding technique, which is a
simple yet efficient algorithm.

The suitability of the Algorithm is tested for different types of images, such
as textured and untextured images. The performance measures such as
Compression Ratio (CR), Maximum Compression Achieved (MCA), Signal to
Noise Ratio (SNR), Error Rate (ER) are evaluated. Compression Ratio is defined
by a number of bits required to represent the compressed image to a number of bits
required to represent the original image. The proposed coding scheme employs
wavelet decomposition, and then it converts the quantized wavelet coefficients
into a plane consisting of 1’s and 0’s using Average Window Coding.

Reasons for preferring Wavelet

Wavelets are preferred in image compression because they have lot of advantages
like:
 Image compression using DWT gives a multi-resolution analysis of an
image. Therefore, an image can be compressed at different levels of resolution and
can be consecutively processed from low resolution to high resolution.
 It is robust to common image processing.
 Real-time images are space and band limited. Wavelets are localized in
time and frequency domains so as to capture local features in an image.
 The main advantage of a wavelet basis is the support for a multi-resolution
analysis of the image. Compared to Fourier Transform, where the problem of
windowing is minimized through Wavelet-based decomposition by differing
window size buffer, wavelets allow analyzing the image at different resolutions.
 67

3.3 PROBLEM DEFINITION

The image compression technology is used for various multimedia and

communication applications. The compressions of individual images have been
used extensively, and the fewer on the problem of compressing image sets. This
chapter presents a unified technique for compression of images. Image
compression processes the removal of redundant information from the image so
that only essential information can be stored to reduce the size, transmission
bandwidth, and time. There are two methods to design simple and efficient coding
algorithms. First, a basic coding algorithm must be modified to make it easier
because it increases its performance; therefore an effective coding algorithm must
be altered by maintaining its output while raising its sophistication. Transform
coding (TC) is an efficient coding scheme based on the utilization of inter-pixel
correlation. In transform coding, an image is transformed from one domain
(usually spatial or temporal) to a different type of representation, using some well
–known transform. Then, the transformed values are coded to provide greater data
compression.This chapter uses a mixed approach which combines the functions of
the technique of transformation and spatial codingThe mixed approach is an
efficient approach to compress the images.

3.4 PROPOSED METHODOLOGY

A mixed approach is fundamentally carried out in two steps: a mapping (or
transformation) of the original image into a sequence of descriptors being another
representation of the image, followed by spatial Coding of the descriptors. The
transformation aims to eliminate redundant information in the image. Transform
coefficients are the descriptors on which spatial Coding is essential for
 68

compression. The proposed coding scheme employs wavelet decomposition, and

then it converts the quantized wavelet coefficients into a bit plane consisting of 1's,
and 0's using a spatial coding technique, namely Window Average Coding. In
order to evaluate the performance of the proposed Algorithm and to compare them
with known existing Algorithms, experiments were performed on standard test
images. Standard images obtained from Brodatz albums are taken as test images
for the work.

3.4.1 Transform technique

Image transform is usually referred to as the class of unitary matrices used
for representing the images. In lossy image compression, the transform co-
efficients can be quantized according to their statistical properties producing a
much-compressed representation of the original image data. The purpose of
transformation is to convert the data into a form where compression is easier. The
net effect is to reduce the redundancy of representation. In this work, the Haar
Wavelet Transform is used to compress the image. When Haar Wavelet is applied
to the image, significant coefficients are obtained. These significant coefficients
alone are used for the reconstruction process in the image.

3.4.2 Transform based image compression

Transform refers to changing the coordinate basis of the original signal,

such that a new signal has the whole information in few transformed coefficients.
The processing of the signals in the transform domain is more efficient as the
transformed coefficients are not correlated. Figure 3.2 shows the transform-based
image coder.

Original image Linear transform Quantization

Compressed image Entropy coding

 69

(a) Encoder

Reconstructed image Inverse transform Inverse Quantization

Compressed image Entropy decoding

(b) Decoder
Figure 3.2 Block Diagram of transform-based image coder (a) Encoder (b)
Decoder

3.4.3 Wavelet transforms

A wavelet is an effectively limited duration waveform that has zero average
value.
 70

Sine Wave Wavelet(db10)

Figure 3.3: Wavelet Representation

Figure 3.3 shows the basic concept of Wavelet. Wavelet is considered as a

wave that oscillates with varying amplitude starting with zero. It can be visualized
as a "brief oscillation," that can be seen by a seismograph or heart monitor.
Wavelets are defined by mathematical functions that cut up data into different
frequency components and then study each component separately with a resolution
matched to its scale. There is an advantage of Wavelet over traditional Fourier
methods in analyzing physical situations where the signal contains discontinuities
and sharp peaks. Generally, wavelets are intentionally crafted to have specific
properties that make them useful for signal processing. Wavelets can be combined,
using a "shift, multiply and sum" technique referred to as convolution, with
portions of an unknown signal to extract information. Wavelets can be used to
extract information from various kinds of image. Sets of wavelets are needed to
analyze information of image. A set of complementary wavelets is used to
reconstruct the image without error so that the reconstruction process may be
reversible. Thus, sets of complementary wavelets are useful for Wavelet-based
image compression/decompression methods, where it is desirable to recover the
original image with minimal allowable loss.
 71

The basic purpose of DWT is to transform the image into slowly varying
and rapidly varying contents of frequency. The separation process is performed in
two orthogonal directions separately. Average pixel information represents a
smooth component that is derived from a low pass filter, which,in turn, represents
average information. Differentials of adjacent pixels by the detail component are
derived from a high pass filter. Images are mathematically transformed to extract
details from them. These transforms give details that are readily available in it.
Images captured mostly are time-domain images in nature, yet this is not the best
way to represent it for most image processing applications.

In most cases, the main distinguished information is not visible in the signal
represented in the frequency domain. It is known intuitively, that the frequency is
associated with variation in the rate of information. Rapidly changing information
is considered as high frequency, whereas if it does not change rapidly, i.e., it
changes smoothly, this is considered as low-frequency information. If there is no
change in information, i.e., it does not change at all, then it can be referred as
having zero frequency, or no frequency. Fourier Transform gives the detail of the
frequency contents that exist in the image that is taken for analysis. In some
applications, where the time localization of the spectral contents is needed, a
transform giving both the time and frequency values of the image is required. The
transform that gives such detail is the Wavelet transform. It represents the
information in both time and frequency. Normally, the Wavelet Transform is used
to analyze images that are inactive in nature.

The following are the Special Features of the Wavelet Transform:

 Very good image estimation with just a small number of coefficients
 72

 It can be used to take out and encode edge information, which provides
significant visual cues in separate images.
 The coefficients of wavelet decomposition give information that is
sovereign of the original image resolution.

Thus, the image compression method based on DWT enables analysis of the
image with different resolutions. Finally, wavelet coefficients are used to calculate
the required linear time in the image size. They cannot minimize the size of data
required to represent the image. Algorithms based on both vector and scalar
quantization’sare falling in the category of lossy compression methods requiring a
minimum size of data to represent the image.

3.4.4 HAAR wavelet

The Haar system is an orthonormal system on the interval. The Haar

wavelet is a first known wavelet that has an orthonormal basis. It is a sequence of
functions supported on the small subintervals of length. The Haar basis functions
are step functions with jump discontinuities. The Haar wavelet transform is used
to compress one and two-dimensional signals. Haar wavelet basis can be used to
represent this image by computing a wavelet transform. To do this, first, the pixels
can be summed up together, to get the new lower resolution image with pixel
values. Clearly, some information is lost in this averaging process. For this
purpose, some detail coefficients need to be stored for recoveringthe original four-
pixel values from the two averaged values. In the given example, 1 is chosen for
the first detail coefficient, since the average computed is 1 less than 9 and 1 more
than 7. This single number is used to recover the first two pixels of the original
four-pixel image. Similarly, the second detail coefficient is -1, since 4 + (-1) = 3
 73

and 4 - (-1) = 5. Thus, the original image is decomposed into a lower resolution
(two-pixel) version and a pair of detail coefficients. Table 3.1 displays the full
decomposition achieved when this process is repeated recursively on the averages.
Table 3.1 Full Decomposition
Resolution Averages Detail Coefficients
4 [9 7 3 5] [2,-2,3,-3]
2 [8 4] [1 -1]
1 [6] [2]

Thus, for the one-dimensional Haar basis, the wavelet transform of the original
four-pixel image is given by [6 2 1 - 1]. The method used to compute the wavelet
transform by recursively averaging and differencing coefficients is called as filter
bank. The image can be reconstructed to any resolution by recursively adding and
subtracting the detail coefficients from the lower resolution versions.

The family of N Haar function h k ( t) is defined on the interval 0 ≤ t ≤ 1. The shape of

the Haar function of an index k is determined by two parameters:p and q,where

k =2 p +q−1(3.1)
and k is in a range of k =0,1,2 , … ., N −1

when k=0, the Haar function id defined as a constant h 0 ( t )=1/ √ N when K>0, the
Haar function is defined as

{ }
P /2
2 ( q−1 ) 2P ≤ t<(q−0.5)/2 p
1
h k ( t )= −2
P /2
( q−0.5 ) 2P ≤t <q /2 p (3.2)
√N
0 otherwise
 74

From the above equation, it can be seen that p determines the amplitude and width
of the non-zero part of the function, while q determines the position of the non-
zero part of the Haar function. The discrete Haar functions formed the basis of the
Haar matrix H

H 2N =
[ H N ⨂ [ 1,1 ]
I N ⨂ [ 1 ,−1 ]](3.3)

H ( 0 )=1
where

[ ]
1 0 ⋯ 0 0
0 1 ⋯ 0 0
I N = ⋮⋮ ⋱ ⋮⋮ (3.4)
0 0 1 0
⋯
0 0 0 1

where N=2k What is HN

and⨂ is the Kronecker product

The Kronecker product of A ⨂ B, where A is m∗n matrix and B is a p∗q matrix,

is expressed as

[ ]
a 11 B ⋯ a1 n B
A ⨂ B= ⋮ ⋱ ⋮ (3.5)
am 1 B ⋯ amn B
 75

[]
ϕ
h0,0
h1,0
h1,1
H N= ⋮ (3.6)
h k−1,0
h k−1,1
⋮
hk−1 ,2 k−1
−1

Where ϕ =[ 1 1 1 … 1 ] is a 1∗N matrix and is a H N Haar function.

The Haar matrix is real and orthogonal, i.e.,
¿
H =H
H −1=H T i.e., H T H=I
A un-normalized 8-point Haar matrix H 8 is shown below

[ ]
1 1 1 1 1 1 1 1
1 1 1 1 −1 −1−1 −1
1 1 −1 −1 0 0 0 0
H [ m, n ] = 0 0 0 0 −1 1 −1 −1 (3.7)
1 −1 0 0 0 0 0 0
0 0 1 −1 0 0 0 0
0 0 0 0 1 −1 0 0
0 0 0 0 0 0 1 −1

From the definition of the Haar matrix H, it can be observed that unlike the
Fourier transform, H-matrix has an real element only and is non-symmetric.

The first row of the H matrix measures the average value, and the second row H
matrix measures a low-frequency component of the input vector. The next two
rows are sensitive to the first and the second half of the input vector respectively,
 76

which corresponds to moderate frequency components. The remaining four rows

are sensitive to the four sections of the input vector, which corresponds to high-
frequency components. The Haar function with narrower width is responsible for
analyzing the higher frequency content of the input signal. The inverse 2k point
Haar matrix is described as H −1=H T D
D [ m, n ] =0ifm ≠ n (3,8)
D [ 0,0 ] =2 (3.9)
−k

D [ 1,1 ] =2 (3.10)
−k

D [ n , n ] =2
−k + p p p+1
if 2 < n<2 (3.11)

For k=3,un-normlised inverse 8-points Haar transform.

[ ]
1/8 0 0 0 0 0 0 0
0 1 /8 0 0 0 0 0 0
0 0 1/ 4 0 0 0 0 0
D= 0 0 0 1/4 0 0 0 0 (3.12)
0 0 0 0 1/2 0 0 0
0 0 0 0 0 1/2 0 0
0 0 0 0 0 0 1/2 0
0 0 0 0 0 0 0 1/2

Definition 3.1 Haar Wavelet

The Haar wavelet is constructed from the Haar Wavelet Transformation which is
generated by the scaling function φ=x (0,1 ) ( x ) forj , k ∈ Z

φ= {01otherwise
0≤ x≤1
}(3.13)
 77

A family of shifted and translated scaling function can also be defined by

j
{ φ j , k ( x) } j ,k ∈ Z by φ j , k ( x )=2 2 φ ( 2 j x−k ) (3.14) as shown as Figure 3.

22

1.5

0.5

-0.5

-1 -0.5 0 0.5 1 1.5 22

Figure 3.4: Scaling Function or Father Wavelet

it is clear that

{ }
−j −j
φ ( 2 j x−k )= 1 k 2 ≤ x<(k + 1)2 (3.15)
0 otherwise
This collection can be introduced as the system of Haar scaling functions.
Definition 3.2 Haar Function
ψ ( x ) =X (0,1 / 2) ( x )− X ( x)
Let ( 2,11 ) be the Haar function

{
1 0≤ x ≤ 1/ 2
ψ ( x ) = −1 1 ≤ x ≤ 1 ( 3.16)
2 2
0 otherwise

forj,k j , k ∈ Z by translation and dilation we can define

ψ j , k ( x )=2 ψ ( 2 x−k ) (3.17)
j /2 j
 78

The collection {ψ j ,k ( x ) } j ,k ∈ Z is introduced as the Haar system on R. The Haar

scaling function is shown in Figure 3.5

-2.5

22

1.5

0.5

-0.5

-1

-1.5

22

-2.5

-1 -0.5 0 0.5 1 1.5 22

Figure 3.5: Haar Scaling Function

3.4.5 HAAR wavelet algorithm

Input = Image of size N×N

Output = Image of size N×N
 79

Step1: Read a greyscale image.

Step 2: Apply Wavelet transform for the image.
Step 3: Obtain significant co-efficient.
Step 4: Expand the Wavelet co-efficient with zero padding
Step 5: Colum and row re-decomposition
Step 6: Reconstructed Image obtained

3.4.6 Properties of HAAR Wavelet

Functions φ j , k ( x) and ψ j , k ( x) are supported on the dyadic interval

I j , k [ K 2−J , ( K + 1 ) 2− j ) since for each j , kϵ Ζ

j
2
φ j , k ( x )=2 xI j ,k ( x )(3.18)
and

ψ j , k (t )=2
j /2
( xI j+1,2 k ( x )−xI j +1,2k+1 ( x ) ) (3.19)
It means that they are not vanish on I j , k

The system { ψ k , j∨ j ,k ∈ Ζ } is complete in L2 ( R)

Proof: To show the completeness of this system, it is to be shown that span

{ ψk , j: j , kϵ Ζ }is dense in L2 ( R). Let us consider the vector space V j i.e

V j=span { xI j ,k :kϵZ } (3.20)

It is obvious that simple functions are dense in L2 ( R). Hence, the set of all finite
linear combinations of characteristic functions of intervals is also dense in L2 ( R ) .
 80

And also, it is known that for every interval I, the function I can be approximated
by functions in U jϵz V j . So

2
span {❑❑ V : jϵz }=L ( R ) (3.21)

As mentioned earlier,the sequence {Vj } j ∈ Z is increasing.i.e for every

j ∈ Z ,Vj ⊆ Vj +1.This property holds for the Haar scaling function. Moreover,
∩ j ∈Z Vj=0 . To show this, let a function f ∈ ∩ j ∈Z V is considered. It is clear that f is a

constant on arbitrarily intervals, and f ∈ L2( R). Therefore f =0.In the next step of
the proof, the vector space W j is defined by

Wj=span { ψ j ,k :k ∈ Z } (3.22)

Thus, it is shown above that the system {ψ j ,k :k ∈ Z } is an orthonormal system in

L2 ( R). So Wj ⊥W ' jforj ≠ j−1Then it is enough to show that.

2
W j=L ( R ) (3.23)
By definition,

φ 0,0 ( x )= X [ 0,1 ) ( x ) (3.24)

1 1
¿ X [ 0,2) ( x ) + ( X [ 0,1 ) ( x )−X [ 1,2) ( x ) ) (3.25)
2 2

1 1
¿ φ−1,0 ( x ) + Ψ −1,0 ( x ) ∈V −1+ W −1
√2 √2

By translation, it can be seen

 81

1 1
ϕ 0 , k ( x) ϕ−1, k ( x)+ ψ (x )∈V −1+ W −1 .(3.26)
√2 √ 2 −1 ,k
So V 0 ⊆ V −1 +W −1. And also by the above relation V −1 ⊆ V 0, and W −1 ⊆V 0, which
means V −1 +W −1 ⊆ V 0 .
Hence

V 0=V −1 +W −1 (3.27)
As it is known thatV −1 ⊥W −1 , the above sum is a direct sum.
V 0=V −1 ⊕ W −1. (3.28)

So, by repeating these arguments at the level N, the orthogonal decomposition.can

be obtained by

V N =V N −1 ⊕W N−1, (3.29)

which is the very important property of the Haar basis. By induction

V N =V N −1 ⊕W N−1=( V N −2 ⊕ W N−2 )⊕W N−1=V M ⊕ ( ⊕ N−1 Wj ),(3.30)

j=M

For any M ≤ N −1. So by the above relation, if M →−∞ , then V ❑ . M .→ 0 . Hence

j=−∞
V N =⊕ N −1 Wj . (3.31)

Let N →+ ∞ , we have

2
L ( R)=⊕ j ∈Z Wj . (3.32)
 82

Hence, the property is done.

3.4.7 SPATIAL TECHNIQUE

Spatial transform modifies the spatial relationship between pixels in an

image. Spatial transform is also referred to as mapping pixels location in an input
image to a new location in an output image. In this work, Average Window
Coding is used to implement the Spatial Technique. The image to be compressed
is divided into small non-overlapping blocks. A two-level (one bit) quantizer is
used. The blocks are coded individually, each into a two-level signal. Meanvalues
for each individual blocks are calculated and the mean values are updated to obtain
the compressed image. Mean is obtained by using the equation

N −1
1
X=
N
∑ Xi (3.33)
i=0

The reconstructed image is obtained by applying mean values for all non-
overlapping windows.

3.4.7.1 WINDOW AVERAGE CODING ALGORITHM

Input = Image of size N×N

Output =Image of size N×N

Step1: Read a grayscale image.

Step2: Divide the original image into non-overlapping windows of size 2×2
 83

Step3: Obtain the mean of the Window by using formula

Step4: Replace each mean value in the spatial domain representing the entire
Window. The compressed image is obtained.
Step5: Reconstruction takes place by applying mean values for all non-
overlapping windows
Step6: Reconstructed Image obtained

3.4.8 MIXED APPROACH

In a mixed approach technique, a combination of the Transform coding

technique and Spatial coding technique is used. The image is first passed through
the Haar wavelet; the obtained compressed image is then passed through the
Average Window technique where the image is compressed much more. In the
reconstruction process, the compressed image is first passed through the inverse
window average technique. The reconstructed image is then again passed through
the inverse Haar transform technique, where the final reconstructed image is
obtained. Mixed Approach image compression and Reconstruction blocks are
shown in Figures 3.6 and 3.7, respectively.

Wavelet
Input Transform Transforme
Image Coding d Image

Window
Storage Average
Coding
 84

Figure 3.6 Block Diagram of Mixed Approach to Image

Compression Scheme

Inverse Window
Storage Average Coding Wavelet Code
Expansion

Inverse
Reconstructed
Wavelet
Image
Transform

Figure 3.7 - Block Diagram of Mixed approach Image Reconstruction Scheme

3.4.8.1 MIXED APPROACH ALGORITHM

Input = Image of size N×N

Output = Image of size N×N

Step 1: Read a grayscale image.

Step 2: Apply the Wavelet transform for the original image of size N×N.
Step 3: Perform average window coding in the wavelet coefficient by dividing
them into non-overlapping windows of size 2×2
Step 4: Obtain the mean of the Window by using formula (3.1)
Step 5: Replace each mean value in the spatial domain representing the entire
Window
Step 6: Obtained the compressed image
 85

Step 7: Reconstruction takes place by applying mean values for all non-
overlapping windows
Step 8: Wavelet co-efficient are applied to inverse window average coding
Step 9: Reconstructed image is obtained of size N×N

3.5 EXPERIMENTS AND RESULTS

Experiments were performed on different types of images—first, the
images of sizes N × N were subjected to wavelet coding. In the resulting
coefficient array matrix, N/2 × N/2 elements in the top left corner alone were
chosen for subjecting it to Window Average coding. The above procedure was
repeated by changing block sizes in the wavelet coefficient array matrix. The
suitability of the Algorithm was tested for different types of images such as
textured (Calfskin) and un-textured images (Lena), and the Performance Measures
were evaluated. For the un-textured image "Lena" 8 bits per pixel image was given
as the input, and the reconstructed image obtained by using different Algorithms
are shown in Figure 3.8 and the Performance Measures evaluated are listed in
Table 3.2.

Input Image Wavelet Transform

 86

Window Average Mixed Approach

Figure 3.8 - Original (Input) and reconstructed images obtained using various
Image Compression algorithms on “Lena” image
Table 3.2 – Performance Measures obtained for “Lena” image using different
algorithms

Performance Measure

Algorithm
MCA(morphologic
al component E.R. SNR C.R.
analysis)
Transform
Coding using 75 0.0195 71.453 4
Wavelet
Spatial Coding using
75 0.0195 71.453 4
Window Average
Mixed Approach
93.75 0.0338 66.775 16
[Transform + Spatial]

By definition, Compression Ratio is the ratio of uncompressed data size

(Suncomp) to the compressed data size (Scomp) , thus:
 87

CR =(Suncomp / Scomp) (3.34)

CR is a relative measure (and dimensionless) and it is many times represented as a

normalized ratio. 
The signal is equal to the mean of the pixel values and the noise is the standard
deviation or error value of the pixel values. Either the ratio or the
SNR=10×log10(signal/noise) can be used to express the result in decibel.
(3.35)
The Error Rate  (ER) is the number of incorrectly or image compressed with
information loss. The expected value of the ER is a denoted as p, probability
packet error that can be expressed as N,
(3.36
Graphical representation of Compression Ratio (C.R.), Signal to Noise Ratio
(SNR) and Error Rate (E.R.) for Transform, Spatial and Mixed Approach are
shown in Figure 3.9

Technique vs Performance Measures

80

70 71.45 71.45
66.77
60

50 C.R
SNR
40 E.R

30

20 16

10 4 4
0 0.01 0.01 0.03
Transform Spatial Mixed

Figure 3.9 Graphical Representation of Performance Measures for un-

textured image
 88

For textured image "Calfskin", 8 bits per pixel image is given as the input,
and the reconstructed image obtained using different Algorithms are shown in
Figure 3.10 and the Performance Measures evaluated are listed in Table 6.2.
Graphical representation of Compression Ratio (CR), Signal to Noise Ratio
(SNR), and Error Rate (ER) for Transform, Spatial, and Mixed Approaches are
shown in Figure 3.11.

Input Image Wavelet Transform

Window Average Mixed Approach

Figure
3.10 - Original (Input) and reconstructed images obtained using various
Image Compression algorithms on “Calfskin” image
 89

Table 3.3 – Performance Measures obtained for “Calfskin” image using

different algorithms

Performance Measure

Algorithm

MCA E.R. SNR C.R.

Transform
75 0.0474 67.028 4
Coding using
Wavelet

75 0.0474 67.028 4
Spatial Coding using
Window Average

93.75 0.0821 62.187 16

Mixed Approach
[Transform + Spatial]
 90

Technique vs Performance Measures

80

70 67.02 67.02
62.18
60

50 C.R
SNR
40 E.R

30

20 16

10 4 4
0 0.04 0.04 0.08000000000000
01
Transform Spatial Mixed

Figure 3.11 Graphical Representation of C.R. and SNR for textured image

3.6. SUMMARY

Experiments have been performed with different types of images, such as

textured and untextured images for verifying the efficiency of the Algorithm.
While using the transform coding and spatial coding Algorithm for image
compression application individually, the value of MCA estimated was 75%, and
ant Compression Ratio was 4:1. But when the techniques were combined in the
mixed approach of the image compression algorithm, it was found that the MCA
increased to 93.75% with Compression Ratio of 1:16. Even though there is a small
variation in the ER and SNR values, the increased value of MCA, compared to the
fall in the quality of image reflected by ER. and SNR is appreciable and
compromising. The Mixed Algorithm offers a 'trade off' between the quality of the
 91

image and also offers a high value of MCA. When tested for different sizes of
images, this Algorithm, gives the same type of result showing its robustness.
 92

CHAPTER 4

LOSSLESS MEDICAL IMAGE COMPRESSION ALGORITHM

USING TETROLET TRANSFORMATION

4.1 OBJECTIVES

In many analytical applications, the reuse of medical images plays an

important role. To store the images, image compression is the technique used to
save storage space and storage time. The image compression is used in image
transmission for fast transmission. Also, in the health care sector, the compressed
image helps to make it completely computerized. The noise interference is the
major problem in image compression. Thus, some compression techniques are
used to remove the noise got added in medical images. This will improve the
quality of the compressed image. Also, in image compression, the medical image
can be processed to a deep degree by de-noising, edge preservation, and high
compression rate. This chapter aims to develop an efficient computational Tetrolet
transform, which can be used in lossless compression. For analysis purposes,
various wavelet techniques are used, and the result is compared with these
techniques. From the analysis, it is inferred that the proposed technique can
achieve high-quality images with low noise.

4.2 INTRODUCTION
As the medical field is advancing, today medical images are stored in some
areas, and are used for future diagnosis of patients. Thus, the large volume of
images is generated and reused today.As huge production of medical images has
become essential, it is also essential to undertake the process of compression
before storing or transmitting medical images through the internet. By
 93

compression, the transmission time will be reduced. Thus, the compression has an
important role for efficient storage and transmission. Among different
compression methods practiced now, the wavelet compression technique is more
used in modern medical image compression. The wavelet technique is becoming
more popular because of exceptional image quality at a high compression rate.
But, since the 3D images are also introduced in the medical images instead of
wavelet, wavelet encoders are used for the compression.

An image compression system mainly consists of two distinct structural

blocks, which are encoder and decoder. Usually, an image f(x, y) is fed into the
encoder, and the encoder creates the set of symbols from input data and uses them
to represent the image. Figure 4.1 shows the block diagram of the compressing
image system.The image compression techniques or algorithms that are used, are
based on the JPEG and JPEG2000 standards.Nowadays, JPEG is widely used in
many application areas. JPEG-LS is called lossless compression also, exactly
what is seen is what is compressed in that case, and there is no irrelevant
information added.In data compression, less data is used to represent the original
data, taking account of the information contained on it. The JPEG file format was
named for the committee that wrote the standard. It was originally designed to deal
with color images, as it could store 24 bits per pixel of color data (instead of 8 bits
per pixel) and restore millions of colors. Compression level is usually expressed in
ratios (2:1, 10:1, 20:1, and so on)

Transformation: The discrete cosine transform cuts the image into blocks
of 64 pixels (8×8) and processes each block independently, shifting and
simplifying the colours so that there is less information to encode.
 94

Quantization: The values in each block are then divided by a quantization

coefficient. This is the compression step where information loss occurs. Pixels are
changed only in relation to the other pixels within their block.

Encoding: The reduced coefficients are then encoded, usually with

Huffman coding (entropy encoding that finds the optimal system of encoding
based on the relative frequency of each character).With high ratio compression,
the block boundaries become obvious, causing the “blockiness” or blocking
artifactwhich is often observed in JPEG images.

f(x,y)

Mapper Quantizer Symbol

Coder
Compressed
image

Encoder
Decoder

f(x,y)
Symbol Inverse
decoder Mapper

Figure 4.1 General Image Compressing System Block Diagram

JPEG 2000 is the new standard for wavelet compression issued by the
JPEG Committee. It arose out of the need to harmonize the wavelet compression
algorithm. JPEG 2000 uses a multilevel DWT with octave-scaled decompositions.
JPEG 2000 is the new ISO standard for image compression commonly used to
 95

compress medical images. Version 1 of the standard provides the core coding
system, specifying both lossy and lossless compression. Version 2 provides
extensions to the standard use for a variety of applications. For 3-dimensional data
sets, there are Version 2 extensions that allow the use of several types of
decorrelating transformations in the third dimension. Specifically, wavelet
transforms, linear transforms, and dependency transforms are all classified under
Part 2. These multi-component transformations in Version 2 of JPEG 2000 can be
effective in compressing volumetric datasets because the correlation between
adjacent images can be exploited to achieve better compression than other existing
methodology.

In this type, the codes are transferred based on the cosine bases and wavelet
bases. But in fractal coding, the codes are transferred using affine transformations.
Also, the redundancy (unwanted information) is removed by using a compression
technique. Redundancy indicates duplication and irrelevancy. In redundancy, the
part of the image information is not noticed by the Human Visual System (HVS).
In the medical image compression, lossless compression is preferred, whereas, for
multimedia applications, lossy compression can also be used. The self-similarity
of the images is used in the fractal image compression (FIC) technique.

Fractal image compression is a promising technique for compact storing of

images. It is based on fractals and uses self-similarity that can be found in any
image. Fractal image compression is a lossy method, which is not a problem for
applications like moving images and also for presenting images on the internet.
Standard image compression methods can be evaluated using their compression
ratio. It is the ratio of the memory required to store an image as a collection of
pixels and the memory required to a representation of the image in compressed
form. The compression ratio for the fractal scheme is easy since the image can be
 96

decoded at any scale. Monochrome images can be encoded because color images
are typically extensions of the grayscale representations of an image.

FIC is considered as a block-based method in which the images are divided

into range and domain blocks. Using the matching domain block, the
transformation is carried out. The equivalent fractal codes are transmitted for
range block, which is non-overlapping over the whole image. Thus, the medical
images can be compressed using a fractal approach. The wavelets are commonly
used in image processing applications as compression. The major advantage of
wavelet-based image compression is that it can process multi-scale analysis due to
the presence of frequency spectra. To get details of the images, the multi-
resolution property of the wavelet is used. The blocking artifacts can be reduced
by using DWT.

Lossless compression is used when data of pre-compression and post

compression are accurate. This technique is used when binary data such as
executables, documents, etc. are compressed. On the other hand, images, music,
video, and so on cannot be reproduced exactly. In the case of image, an
approximation of the original image is enough. In this case, lossy compression
techniques must be used for the sake of using minimum storage space. They
consist of removing the redundant information in adjacent pixels to minimize the
number of bits.

Wavelet techniques have been proven to be indispensable for image

processing. Based on this, the Wavelet Transform is used in this work as the
compression technique. Cite, for example, in 1992, and the FBI used wavelet
compression to store and to retrieve 250 million fingerprint files requiring 250
 97

terabytes of space. Using wavelets, the FBI obtained a compression ratio of about
20:1. It can be noted that the storage of a fingerprint with a grayscale of 8 tones
and with a resolution of 500 dots per inch occupies 10 megabytes.

The Discrete Wavelet Transform (DWT) uses Mallat’s multi-resolution

analysis to represent a signal in terms of its frequency components. DWT projects
a signal on different approximation spaces called Vj with different levels of
resolution. Each of these includes another approach at half resolution. The
difference between the two is captured by the spaces Wj, called the subspaces of
detail. If the signal behaves smoothly, the difference between two successive
projections of the signal is very small, and therefore the detail coefficients will be
small or zero.

The main benefit of image compression is reduction in the data storage

space. This will reduce communication costs in transmitting high volumes of data
over long-haul links via higher efficient use of the accessible bandwidth in data
links. It will also reduce the communication cost due to the data rate reduction.
Thus, the quality of multimedia arrangement through short-bandwidth
communication channels will also increase. Due to the advanced compression
techniques, high-performance compression has made new opportunities for
creative applications such as video teleconferencing, digital library, telemedicine
digital entertainment, and digital archiving. The image compression has great
implications in database access. The performance of the database can be enhanced
by using compression. This is because more compressed records can be packed in
a given buffer area in traditional computer execution. This will increase security in
compressed image transmission.
 98

In this chapter, the novel image compression method is proposed. In this

method, the feature is stored instead of pixels to find the relationship in images.
Commonly it is used in content-based feature matching. Another compression
method is proposed in which compression is not carried out based on images pixel
by pixel; instead, it is described. Finally, it is retrieved from the base station based
on the descriptors in a hierarchical scheduling optimization scheme for
compression of images present. The wavelets and the bandelets are described.
Here, the performance is compared. The comparative study of the performance of
contourlet transform and wavelet transform is given. The performance of the
contourlet transform is better than the wavelet transform. The combination of
wavelet and curvelet transform is introduced in the analysis. Further,Tetrolet
transformation based medical image compression is proposedin this chapter to
improve the quality of the image compression, to reduce the storage space, and for
better denoising. The performance is analyzed with the parameters like mean
square error, peak signal to noise ratio, feature similarity index, and mean
structural similarity index.

4.3 PROBLEM DEFINITION

The quality of medical images gets reduced due to interferences in the

noise. If the quality of the images is lost, then it will be difficult to diagnose the
images. The major challenge in the digital image transmission is the transmission
of large size data sets. For example, size of MRI and CT images are 5 to
12Mbytes, and a single X-ray size is considered as 24Mbytes. The compression is
used to send the digital images effectively, and also to reduce the storage space.
This will reduce network traffic as well as improve efficiency. For removing
thenoise, the medical image wavelet transform is used. Different multi-resolution
analyses are also there to remove the noise or denoising process.
 99

The compression ratio of the lossy compression is 50:1 or more. But the
original images cannot be completely recovered. Hence, by using the lossless
compression, original images can be completely recovered. The compression ratio
of lossless compression is around 2:1. In the medical industry, compression
without loss is normal. This is because the retrieval of the whole original signal is
essential for the diagnosis process.

The major contributions are:

 Comparative study of image denoising abilities of wavelet transforms.

 A three-stage methodology implements image denoising.
 The multi-resolution analysis is carried out in the standard images like
magnetic resonance, mammography, and fundus images.

4.4 PROPOSED METHODOLOGY

4.4.1 System Model

Figure 4.2 shows the block diagram of the proposed system. The main part
of the block diagram is the input image, compression module, compressed image,
inverse transformation, and reconstructed image. The tile-matching and
rearrangement will turn Haar model into Tetrolet transformation. The input image
is divided into 4×4 blocks. For each block, a tetromino partition is assigned, which
is updated to the image geometry in this block.

Compression
Pre-processing
Identify matching
filter
Compressed Image
Tetrolet Transformation
 Encoding

100

Reconstructed image Inverse Transformation

Figure 4.2 Block Diagram of Proposed System

4.4.2 Identification of matching tile

Thetetrolet decomposition algorithm is the basis of the Tetrolet transform.

Customary Haar transformation is performed in the 4×4 input image matrix. Then,
the absolute sum of the twelve detail coefficients is calculated, and is given by:

CD ∑ ¿=|s ( vertical )|+|s (horizontal )|+|s (diagonal )|(4.1)¿

Where CDsumis the current detail. The new tiles are selected using a succeeding
tile. By using the Haar transform, new details are obtained, and are stored in
NDsum. After evaluating all the tiles, the process is terminated. After that, matching
tiles are selected.

4.4.3 Tetrolet transform algorithm

 101

Tetrolet transform is a basic four lattice board proposed by Jens Krommweh Harr.
Wavelet transform concept, according to the local geometric features of images,
adaptively selects the corresponding four lattice plate of the square area of sparse
representation. Compared with the traditional multi-scale transform wavelet, the
curvelet and contourlet using the same number of transform coefficients
reconstruction can get better image quality.

The tetrolet transform is designed with a simple structure of a Haar filter. In

the two-dimensional classical Haar case, the low-pass filter and the high-pass
filters are just given by the averaging sum and the averaging differences of each
four-pixel values, which are arranged in a 2×2 square. It can make the function
system adaptable to the local structure instead of selecting the prior basis or frame.
However, it can obtain the anisotropic decomposition with more directional
selectivity. Tetrolet transforms divide the image into 4×4 blocks, then
decomposition is done with more directional selectivity. Tetrolet transforms divide
the image into 4×4 blocks, then a tetromino partition is determined in each block,
which is adapted to the image geometry in this block.

If the image size cannot be divided by four, zero-padding should be

applied. There are five shapes called tetrominoes, each of which contains four
pixels, is used for TT (Figure 4.3. (a)). Four of them are chosen, and the selected
tetrominoes should be located into a 4×4 sized pixel blocks. One hundred
seventeen different combinations are possible, and one of them is given in
Figure4.3.(b). After locating the tetrominoes, pixels of the 4×4 sized blocks are
rearranged according to TT rules with the selected combination of tetrominoes,
and Haar WT is applied on the reformed part of the image locally. In-order to
calculate Tetrolet Coefficients, the input image is divided into 4×4 pixels. It is
 102

considered that the total block size is N = 4, and matching tetrominoes tile of sub-
images are stored in the level-1 Haar approximation. For all the image
transformation groups, approximate, vertical, horizontal, and diagonal coefficients,
the approximation matrix is to be set as the base image and repeated.

(a)

(b)

Figure 4.3 Tetrominoes (a) Shapes of the tetrominoes,(b)a combination of

tetrominoes for a 4×4 pixel block.

4.4.4 Inverse transformation

It is used to retrieve the original image from the compressed image.

Normally, reconstruction is used to make 2-D and 3-D images. The initial step is
selection of approximation and detail coefficients of the sub-image. The inverse
level-1 Haar transformation is performed. Based on the matched tile, the
computation of the elements is rearranged. After that, all sub-blocks are combined
to form a single image.
 103

Based on Haar wavelet-based transform, the Tetrolet transform is made.

ThisTetrolet transform is formed by connecting four identical square tiles. To form
localized orthonormal tetrominoes, matching and rearranging process are
subsequently used in Haar transformation. For all 4×4 sub-blocks of low pass
image, the procedure is performed. After that, sparsest covering from each portion
is stored as a non-redundancy element in the wavelet basis functions. High
compression performance is achieved by applying the tetrolet transform. The
Tetrolet functions are normally used in image watermarking, texture segmentation,
pattern recognition, image projection, and image reconstruction.

Wavelet transformation

Wavelet is considered as a waveform which is said to have zero average

value. The common feature of the wavelet transform is good image estimation
with a small number of the coefficient. It is used for obtaining and encoding
theimage edge data so that the images can be easily differentiated .There are two
types of wavelet transform viz. continuous wavelet transformations and discrete
wavelet transformation.

Wavelet families

The degree of smoothness, energy compaction, regularity, orthogonality,

and symmetry are the common property of wavelets. The important types of
wavelets are listed below:

HaarWavelet

In mathematics, the Haar wavelet is a sequence of rescaled "square-shaped"

functions, which together form a wavelet family or basis. Wavelet analysis is
 104

similar to Fourier analysis in that it allows a target function over an interval to be

represented in terms of an orthonormal basis. The Haar transform can be used
for image compression. The basic idea is to transfer the image into a matrix in
which each element of the matrix represents a pixel in the image.

The expression is given by:

1 1
Ψ ( t )=1 when0< t< ;−1 when < t<1=0 otherwise( 4.2)
2 2

It is found that this transformation is a real function, and it is anti-symmetric with t

½. It is discontinuous in the time domain and low localization in the frequency
domain.

Daubechies wavelet

It is based on orthonormal scaling and wavelet functions families. These

families have maximum regularity to give the length of support of quadrature
mirror filters.

Morlet wavelet

It is formed by multiplying Fourier basis with a Gaussian window, and the

expression is given by:

Ψ ( t )=exp ( j w 0 t ) exp t [ ]
−t 2
2
( 4.3)

√( )( ( ))
2
Π −( w +w 0 )
exp (−( w−w0 ) )+ exp
2
Ψ ( w )= ( 4.4)
2 2

The expression of Mexican hat wavelet is given by:

 105

2
exp (−t )
Ψ ( t )=(1−t¿¿ 2) (4.5)¿
2

Fourier transform of the (4.5) equation can be in the form of

Ψ ( t )=−w exp
2
( )
−w2
2
( 4.6)

Shannon wavelet

The expression is given by

Ψ ( t )=sin ( 2 Π )−sin ( 2 Πt ) /¿ Πt ( 4.7)¿

This has poor time resolution, whereas the frequency localization is excellent.
Thus, a new technique based on the Haar wavelet transform called Tetrolet
transform is formed in this chapter. This technique is used for providing efficient
image representation.

4.4.5 TheTetrolet Transform

The basic notations and concepts of the Tetrolet transform are described here.
Consider two-dimensional square data sets. The index set of a digital image

I =⊏ Z 2( 4.8)

a=( a [ i , ] ) ( Ji, j ) ∈ I (4.9)

Where N = 2J,J∈ N. The associate degree index of confirming a 4-

neighborhood is given ( i , j ) ∈ I by:

N 4 ( i , j ) ≔ ( 1 , j ) , (i+1. j ) , ( i, j−1 ) , ( i , j+1 )(4.10)

An index that lies at the boundary has 3 neighbors; the associate degree of an
index at a vertex of the image has 2 neighbors.
 106

J : I → with N ( ( i , j ) ) ≔ jN +i . A set E=r ∈ N (4.11)

The subsets I v ⊂ I could be a disjoint partition of I if I v ∩ I μ =∅

In this chapter, it tends to contemplate disjoint partitions E of the index set I that
satisfy 2 conditions:

Every set I v contains four indices, i.e., I v =4

Each index of I v contains a neighbor in I v , and it is given by:

∀ ( i , j ) ∈ I ,∃ ( i' , j ' ) ∈ I v : ( i' , j ' ) ∈ N 4 ( i , j ) (4.12)

These subsets Iν are discussed here because the tiling problem of the square values
[0,N)2are determined which is known as tetrominoes could be a well-known
drawback being closely associated with partitions of the index set I = 2. For a
straightforward one-dimensional classification of the four components in one
tetrominoes setIν, tends to apply the bijective mapping J as follows. For I ν= letL:
Iν→ is ruled that to order the values J ( i 1 , j 1 ) , … … , J ( i 4 , j 4 ) is tended by size and
maps them to specified the littlest index identified with zero.

4.4.5.1 Tilings by tetrominoes

The tetrominoes are shaped by using the 4 unit squares. This is connected
by edges, not just at their corners. Irrespective of rotations and reflections, there
are five different shapes, thus referred to as free tetrominoes as shown in Figure
4.4. By considering the isometrics, it's clear that each square (0, N) 2 will be lined
by tetrominoes if and as long as N is even. Larsson showed that there are unit 117
solutions for a disjoint covering of a four × 4 board with four tetrominoes. For an
8 × 8 board, there is a tendency to figure 4.4 1174 > 108 as a rough edge of
potential tilings. Thus, for the sake of handling the number of solutions, it'll be
cheap to limit to an image partition into 4 × 4 squares. As pictured in Figure 4.5,
 107

there are 22 basic solutions within the 4×4 board (disregarding rotations and
reflections).

Figure 4.4. Fundamental forms Tiling of 4×4 board

 108

Figure 4.5 The 22 fundamental tilings filling a 4 × 4 board

The two-dimensional classical Haar wavelet decomposition results in a

special tetromino partition. For introducing the distinct tetrolet transformation, it is
essential to have a tendency to recall the traditional Haar case that is consistent
with the subsequent tetrolet plan.

The Haar Wavelet model considers variations of samples or variations of

native sample averages of the associate input image matrix. Thus,Haar remodels
coefficients, aside from k=l=0, square measure variations on rows and columns of
native averages of pixels in a picture. This square measure is manifested as many
 109

"edge extractions" of the original image. Secondly,for pictures hair transformation

is used as a constituent in rows and columns will move up the process. The
transformation may be an easy constituent add and distinction average for level
one. But for levels which are larger than one, the transformation matrix contains
root components. These components lead to decimal values (pixels values square
measure integers) with even minor decimal worth leading to important value
changes on higher levels. To exhaust this limitation, the repetitious technique
represented in the Approximation matrix for higher levels is used. In general,
Haar may be a family of N functions, ℎ(𝑧),( 𝑘 = 0,....,𝑁 − 1);

k =2 A + H−D( 4.13)

Based on the above formula, p and q are unambiguously outlined for every k with
the condition that 2𝑝 is the largest power of two contained in 𝑘 and 𝑘>0.

A V
[
In general, Haar Transform = H D ( 4.14) ]
=
[ A

H D ]
[ HA VD ] (4.15)
For Level 3 (N=8), the approximation matrix contains 15 coefficients in addition
to the approximation pass:

[
HaarTransform= [ H D ] [ H D ]
A V A V
]
[ HA VD ] [ HA VD ] (4.16)
 110

4.4.5.2 The parameters for simulation

Noise Addition and Thresholding Methods: normally, in medical and test images,
random, Gaussian, and rician noise are applied. In order to get an efficient image
representation, a new adaptive Haar wavelet transform, called Tetrolet Transform
is implemented. Tetrolets are Haar-type wavelets whose supports are tetrominoes
which are shapes made by connecting four equal-sized squares.
Numerical results show the strong efficiency of the tetrolet transform for image
approximation.

X n=X + signa∗N T (4.17)

Where sigma indicates the noise ratio, the factor used to indicate the amount
of noise. The random noise (Nr) is applied to the Gaussian equation :

2
−(x−μ)
1 2σ
2

P ( x) = e ( 4.18)
σ √2 π

Where μ is the mean, and σ is the variance.

Peak-signal-to-noise-ratio (PSNR)

It is used to measure the quality of reconstruction images. It is also used to

measure the loss images quality measurement because of its simplicity and
mathematical convenience. Commonly, it depends on the Mean Square Error
 111

(MSE) of the reconstructed image. MSE is defined as the difference between the
reconstructed image and original image, whose expressions are given by:

M N
1
∑ ∑ [ y ( i , j )−x ( i , j ) ] (4.19)
2
MSE=
MN i=1 j=1

L2
PSNR=10 log ( 4.20)
MSE

Where L is the length of the pixel, M,N is the pixel mean variable

Compression Ratio (CR)

It is used to analyze the performance of the compressed image. The

performance is calculated using parameters like compression efficiency, and it is
measured in terms of compression ratio or by bit rate. It is defined as the ratio
between size to the original image and the size of the compressed image and bit
rate.

¿ the originalimage
CR= (4.21)
¿ compressed image

As described in equation (16), it is defined as the ratio between sizes of the

original image divided by the size of the compressed image. This ratio shows how
much compression is achieved. This indicates the picture quality. Commonly it is
defined that if the compression ratio is high, then the quality of the resulting image
will below. During the compression process, the trade-off between compression
ratio and picture quality is an important factor. Some compression schemes
produce compression ratios that are highly dependent on the image content. This
aspect of compression is called data dependency.
 112

Entropy

This is used to check the similarity of digital image and the original image. The
entropy can be calculated by using MATLAB. The statistical measure of
randomness is termed as entropy. The expression is given by:

E=−∑ ( p∗log 2 ( p ) ) (4.22)

Where p- represents the probability rate.

Start

Take the Input – divide it into image 4×4 blocks

Find sparsest tetrolet representation for the image

Change the low-pass and high-pass coefficient of each block into a 2×2 block.

Store the tetrolet coefficient

Stop
 113

Figure 4.6 Flow Chart of Proposed System

Figure 4.6 shows the flow chart of the tetrolet transform-based image
compression. The basic operation of the proposed method is the conversion of
input into the sparsest tetrolet.After that, the low pass and the high pass
coefficients get changed. Finally, the coefficients are stored and aligned.

4.5 RESULT AND DISCUSSION

The result of image compression is described in this section. The

simulations were performed using MATLAB simulation environmentand were
carried out with metrics for both the proposed protocol and existing. From the
comparison, it is clear that the existing protocol has low performance as compared
to the proposed tetrolet transform. For the simulation, different parameters like
PSNR, MSE, entropy, and the CR are considered. Three different images were
compressed using the proposed tetrolet transform. The images selected are MRI,
CT, and normal image (cameraman). Table 4.1 gives the image quality metrics.
The specification of the input image is given by256×256. With a set of
Approximate coefficients after transformation, the reconstructed image achieves a
remarkable PSNR of -19.6598 because the oriented edges are well adapted.

Table 4.1 Image Quality Metrics

Type of PSNR MSE ENTROPY CR

Images
MRI -19.6598 284.4 3.35 2.94
9
CT -15.8503 226.1 4.22 3.50
6
 114

Normal -12.1530 176.2 6.80 4.42

Image 8

Figure 4.7 shows the input images and the compressed image for three images.
Optimum decomposition level is evaluated using thetetrolet transform. It is found
that the optimal scale number is taken as four for MRI, CT, and normal images.
Table 4.2 shows the simulation result of the compression ratio. Table 4.3 shows
the simulation result of encoding time.

Input Image Compressed Image

(a) Cameraman Image (b) Compressed Cameraman

Image
 115

(c) MRI Image (d) Compressed MRI Image

(e) CT Image (f) Compressed CT Image

Figure 4.7 Input and Compressed Image

 116

Table 4.2 Analysis of Compression Ratio

Parameters Compression Ratio

Tetrolet Haar(Kadam FQT (wavelet-

and rathod based fractal
2019) quad tree
combined with
Huffman coding)
(Kadam and
rathod 2019)

MRI 4.12 7.55 4.79

CT 4.24 7.54 4.69

Normal Image 3.12 7.30 3.25

Table 4.2 shows the simulation result of the compression ratio. The analysis
is performed in three different images. From the table, it is clear that for all the
three images MRI, CT, and the normal image, the proposed tetrolet transform has
a low compression ratio. If the compression ratio is high, the quality of images
will be low. Similarly, the quality of the image will be high if the compression
ratio is low. Thus, the proposed tetrolet transform is capable of generating the
highest quality compressed image as compared to the existing FQT. Figure 4.8 and
Table 4.3show the statistical analyses of the compression ratio. From the graph, it
can be found that the proposed tetrolet transform has a low compression ratio.
From the analysis of the compression ratio for MRI, CT, and normal image, it is
 117

clear that the proposed tetrolet transform has different values. By using tetrolet
transform, the normal image can be compressed in high quality.

Compression Number of File size (% of File size(KB) Compression

level images original) ratio

original 100 100 288 1:1

J90 100 13.1 40 1:7

J70 100 6.4 18 1:16

J50 100 4.4 12 1:20

J30 100 3.3 9 1:30

Table 4.3 Average compression ratio

 118

Figure 4.8 Comparison Graph of Compression Ratio

Table 4.3 gives an analysis result of the encoding time parameter. The encoding
time is the time taken for the compression of the image. Table 4.4 shows three
different techniques used for compression. From the table, it is clear that the
proposed tetrolet transform is capable of yielding higher encoding time. This
shows that images can be compressed faster using the tetrolet transform. Thus, the
proposed tetrolet transform has a faster response compared to the existing
technique FQT.
 119

Table 4.4 Analysis of Encoding Time

Parameters Encoding Time

Tetrolet Haar (Katham FQT (wavelet-based

and rathod 2019) fractal quad tree
combined with
Huffman coding)
(Katham and rathod
2019)

MRI 6.89 6.5 2.25

CT 7.02 6.13 4.10

Normal Image 6.51 5.76 2.7

 120

Figure 4.9 Comparison Graph of Encoding Time

Figure 4.9 shows the statistical analysis of encoding time. The graph also shows
that by using the tetrolet transform CT image can be encoded faster as compared
to other image sets

4.6 SUMMARY

This chapter illustrated an adaptive image compression technique that has

non-redundant tetromino basis functions and a fast filter bank technique. The aim
of this work is to compress the medical image with high quality or low noise. The
group of tetrolet coefficients has the ability to work superior for both real data
arrays and high-end graphics. Thus,the results also show that by selecting a proper
relaxing parameter, the low cost of adaptivity using minimum Tetroletcoefficients
 121

can be achieved. The simulation was carried out in MATLAB. From the analysis,
it is clear that the proposed tetrolet transform has high performances compared to
other existing techniques. This compression technique can be used for image
denoising as well.
 122

CHAPTER 5

LOSSLESS MEDICAL IMAGE COMPRESSION ALGORITHM

USING ORTHOGONAL MOMENT TRANSFORMS

5.1 OBJECTIVES

With the development of CT, MRI, EBCT, SMRI, etc., the scanning rate
and distinguishing rate of imaging equipment has significantly enhanced. Using
Compression techniques, medical images can be processed to a profound degree
by de-noising, enhancement, edge extraction, etc., to make good use of the image
information and improved diagnosis. Since medical images are in digital format,
more time-efficient and cost-effective image compression technologies are to be
developed to reduce the mass volume of image data. This chapter proposes the use
of orthogonal moment transform for fast and higher compression rates. This
method incorporates a simplified mathematical approach using a sub-block
reconstruction scheme that eliminates numerical instabilities at higher moment
orders. Hence, Orthogonal Moment performs better for both real digital images
and graphically generated images.

5.2INTRODUCTION
A modern Medical imaging tool has a significant impact on the diagnosis of
diseases and preparation for surgery. As medical images have transformed into
digital formats such as DICOM, optimal settings for image compression are
needed to facilitate long- term mass storage requirements. Also, with the increased
use of medical imaging in clinical practice, and the growing dimensions of data
 123

set, the management of digital medical image data sets requires high compression
rates. The pharmaceutical enterprise depends on a system that makes diagnostic
images available for radiologic interpretation, that transmits images to physicians
throughout the system, and efficiently stores images pending retrieval for future
medical or legal purposes. Computerized medical imaging generates large, data-
rich electronic files. To speed up electronic transfer and minimize computer
storage space, medical images often undergo compression into smaller digital
files. The level of diagnostic detail needed for clinical interpretation of medical
images varies according to modality. In general, nuclear medicine scans require
less detail than Computed Tomography (CT) or Magnetic Resonance (MR). As
interpretation of mammography and radiography depends on high spatial
resolution, these images demand more detail than CT or MR, which,in turn, needs
high contrast resolution for diagnostic analysis. The need for compressing medical
images are

 The growing need for storage

 Efficient data transmission
 Telemedicine
 Tele-radiology applications
 Real-time Tele-consultation
 PACS (Picture Archiving and Communication Systems)

Optimal medical image compression is defined as a degree of compression

that decreases the file size substantially by maintaining a degree of image
distortion that is not clinically significant. In contrast to JPEG medical image
compression model, moment functions are used for detailed image analysis as they
provide better image representation and higher image quality. Image moments
 124

constitute an essential feature extraction method (FEM), which helps to generate

upper discriminative features. One part of JPEG can determine the compression
ratio, and compression process speed is Zig-zag Coding or Zig-zag Scan
algorithm. For video compression process, speed in frame per second (fps) is
crucial. PAL(Phases Alternating Line) uses 30 fps, while NTSC (National
Television System Committee) uses 25 fps. Zig-zag Scan holds a vital role in
sequencing the coefficient result of Discrete Cosine Transform (DCT)-
quantization starting from the value related to low-frequency information to value
compared to high-frequency details. Matrix 8×8 of DCT-quantized result will be
sequenced in 1× 82 size vector using zigzag reading.

A Feature Extraction Method (FEM) can be termed successful if the

resulted features describe the processed object uniquely in a scene. FEM is more
efficient in classification process. Though many FEMs that provide several
discriminative feature sets have been presented, some are based on the
computation of object moments. Moments have been used successfully in many
classification applications, and their ability to describe an object entirely makes
them a powerful tool in image processing tasks especially in computer vision
applications, like object recognition in robotic applications and object
characterization in visual inspection based quality control systems.Their ability to
capture particular characteristics of the described pattern makes them suitable for
image analysis, image watermarking, Pattern Recognition System (PRS), texture
segmentation, monitoring crowd and image projection.

A new class of moment transform called Discrete Tchebichef Transform

(DTT), derived from adiscrete class of Tchebichef polynomials, is an orthonormal
version of orthogonal moments. Owing to its low computational complexity,
 125

Tchebichef Moment Compression is scalable and portable to smaller computing

devices like Personal Digital Assistance (PDA) and mobile phones. The Discrete
Tchebichef Transform (DTT) is a novel polynomial-based orthogonal transform. It
exhibits interesting properties, such as high energy compaction, optimal
decorrelation, and direct orthogonality, and hence is expected to produce good
transform coding results. Advances in the areas of image and video coding have
generated a growing interest in discrete transforms. The demand for high quality
with limited use of computational resources and improved cost benefits has led to
experimentation with novel transform coding methods.The discrete Tchebichef
transform (DTT) is a useful tool for signal coding and data decorrelation. In recent
years, signal processing literature has employed the DTT in several image
processing problems, such as artifact measurement, blind integrity verification,
and image compression. In particular, the 8-point DTT has been considered in
blind forensics for integrity checks of medical images. For image compression, the
8-point DTT is also capable of outperforming the 8-point discrete cosine transform
(DCT) in terms of average bit-length in bitstream codification. Moreover, an 8-
point DTT-based encoder, capable of improved image quality and reduced
encoding/decoding time was proposed, as a competitor to state-of-the-art DCT-
based methods.

The significant benefits of moment based compression are

1. Reduces the amplitude of image features and helps in precise restoration.

2. Being entirely robust to anti-noise, anti-lossy, and anti-mirror operations.

Some essential characteristics of DTT are summarized as follows:

 A discrete domain of definition which matches precisely with image

 126

coordinates space.
 The absence of numerical approximation terms allows a more accurate
representation of image features than others, which is not possible using
conventional transform.

The orthogonal moment functions were employed for various applications

of image processing, which include image segmentation, computer vision, image
analysis, and image compression.Moments are derived from polynomial functions
and computed for a digital image. They are statistical values that represent both
low-frequency information and high-frequency details of an image. Unlike
geometric moments, orthogonal moments represent independent features of the
model and thus have near-zero or very minimum information redundancy in a set.
The invariant property of orthogonal moments concerning image translation,
rotation, and sizing invites scientists/researchers to make use of them for various
image processing applications. Since continuous moments suffer from
discretization error severely, when they are numerically implemented in the
domain of the discretized image coordinate space, few orthogonal
polynomials/moments are directly defined in the discrete area. It has been well
proved that the distinct orthogonal moments are very useful as pattern features in
the analysis of two-dimensional images. The task of compression, the Tchebichef,
outperformed others by reaching the highest value of the Peak-Signal-to-Noise
Ratio (PSNR). This work aimed to build a simple compression method with better
performance. The discrete orthogonal Tchebichef moment is considered in this
study as its compression efficiency.
 127

5.3PROBLEM DEFINITION

Image compression is the science of efficient coding of image data with a

target to reduce the number of bits required in performing an image. The
advantages of image compression are saving time for image transformation and
memory for image storage. The Tchebichef moment transform has good energy
compacts and works better for specific images. This chapter proposes an
orthogonal moment for fast and efficient image compression. This chapter further
analyzes the reconstruction aspects of real and graphical images for compression.
The results show the advantage of orthogonal transform evaluation, and proves
that it provides compact support to sub-block reconstruction of image
compression.

5.4 PROPOSED METHODOLOGY

5.4.1 Orthogonalmoments

In image processing applications, an image moment is defined as a

particular weighted approximation of the image pixel intensities. Orthogonal
moment functions are used in feature representation techniques for image
reconstruction and object identification. Orthogonal moments are often preferred
due to their ability to represent images with minimum redundancy.

Definition

Consider a General moment M pq(f ) of an image f ( x , y ) where p,q are non-

negative, and r=pqis called the order of the moment is defined as

M pq =∬ P pq ( x , y ) f ( x , y ) dxdy (5.1)
(f )

wherep00(x, y), p10(x, y), . . . ,pkj(x, y), . . .( small letter p) are polynomial
basis functions defined on D. These set of Cartesian moments following the
 128

orthogonality condition are called as orthogonal moments expressed

mathematically - two functions ym and yn are orthogonal over an interval a≤x≤b
if :

∫ y m ( x ) y n ( x ) dx=0 m≠ n(5.2)

The orthogonality condition simplifies the reconstruction of the original

image from the generated moment functions. The set of discrete orthogonal
moment functions based on discrete Tchebichefpolynomials has been
successfully introduced as alternatives to continuous orthogonalmoments. The
computational aspects of Tchebichef moments are discussed below:

5.4.1.1 TCHEBICHEFmoments

Let Tmn be defined as Tchebichef moments based on a discrete orthogonal

polynomial set {tn(x)} specified directly on the image space [0, s–1].

s−1 s−1
1
T mn= ∑ ∑ t m (i)t n ( j)f ( i, j ) (5.3)
p ( m , s ) p ( n , s ) i=0 j=0

form,n=0,1,2…….,s-1; where s represents the initialorthogonal polynomial

The Tchebichef orthogonal polynomials set { tn( x) } are generated

recursively with the following set of initial conditions as:

t 0 ( x )=1

2 x +1−s
t 1 ( x )=
s

The general Tchebichef orthogonal polynomials equations are:

 129

( )
2
(n−1)
( 2 n−1 ) t1 ( x ) t n−1 ( x )−(n−1)t n−2 ( x ) 1 2
s
t n ( x )= (5.4 )
S

for n=2,3,……, s-1

The degree of the polynomial is defined by:

β ( n , S ) =S n (5.5)

The squared-norm equation of tn(x) is given by:

s−1
ρ ( n , S )=∑ { t i (x) } (5.6)
2

i=0
 130

( )( )( ) ( )
2 2 2 2
1 2 3 n
S . 1− . 1− 2 . 1− 2 … … 1− 2
S
2
S S S (5.7)
¿
2 n+1

For a description of the properties of Tchebichef polynomials and the definitions

of related terms such as the squared-norm ρ( ), the Tchebichef orthogonal
polynomials set can be generated iteratively according to as follows.

2 x +1−S
t 0 ( x )=1 , t 1 ( x )= (5.8)
S

t n ( x )=
(
( 2 n−1 ) . t 1 ( x ) . tn −1 ( x )−( n−1 ) 1−
( n−1 )2
S2 )
. t n −2 ( x )
(5.9)
S

Discrete orthogonal Tchebichef moment has its advantage in image processing,

which has not been fully explored. Since computer image data operates on
integers, discrete orthogonal Tchebichef moment is suitable for computer image
processing.

5.4.2 TMT compressionalgorithm

In the proposed compression algorithm, an RGB image is divided into

equal-sized blocks of image data, and discrete Tchebichef moments are
performed independently on each block. The quantization tables are proposed to
reduce high frequencies. Next, the Huffman code is used to calculate the average
bit length of TMT coefficient. TMT is implemented to achieve excellent
compression performance. The visual TMT image compression is depicted in
Figure 5.1.
 131

Input RGB YCbCr space TMT Moment

image conversion co-
efficients

Compressed Huffman ZigZag scan Quantisatio

image coding n

Figure 5.1 Block diagram of proposed algorithm

5.4.3 Color spaceconversion

To achieve a good compression ratio, the correlation between the color
components is lowered by converting the RGB image to YCbCr. The RGB image
should be separated into a luminance (Y) and two chrominance (Cb and Cr).
YCbCr can be computed directly from an 8 bit RGB as follows:

[ ][ ][ ] [ ]
Y 0.299 0.587 0.114 R 0
Cb = 0.1687 −0.3313 0.5 G + 128 (5.10)
Cr 0.5 −0.4187 −0.081 B 128

[][ ][ ] [ ]
R 1 0 1.4021 Y 0
G = 1 −0.34414 −0.71414 Cb − 128 (5.11)
B 1 1.7718 0 Cr 128
 132

The JPEG committee was considered for use as a standard color space. The
YCbCr color space was used as the default color space carried out for the JPEG
committee. It has a limited amount on the use of other color spaces that affects the
compressibility of color images. The XYZ and YCbCr color spaces are better than
the RGB color space for image compression.

YCbCr is a color space used as part of the color image pipeline in video and
digital photography systems. YCbCr is not an absolute color space. It is a way of
encoding RGB information. The actual color displayed depends on the actual RGB
primaries used to display the signal. It represents color in terms of one luminance
component (Y) and two chrominance components (Cb and Cr), where Cb is the
chrominance-blue component, and Cr is the chrominance –red component.

The YCbCr image can be converted to/from the RGB image. To convert
from RGB to YCbCr, one variant of this color space (according to ITU-R BT.709)
is considered as following:

Y = (0.2126 ×red)+(0.7152×green)+(0.0722×blue)
Cb = 0.5389×(blue-Y)
Cr = 0.6350×(red-Y)

Reasons for using YCbCr

1. The human eye has a different sensitivity to color and brightness.
Hence,there is a need to transform RGB to YCbCr.
2. In contrast to RGB, the YCbCr color space is lemma-independent, resulting
 133

in better performance.
3. YCbCr is broadly utilized in video compression standards such as MPEG
and JPEG.
4. Y can be stored with high resolution or transmitted at high bandwidth, and
two chrominance components (Cband Cr) that can be bandwidth-reduced,
sub-sampled, compressed or otherwise treated separately for improved
system efficiency.

According to the Human Visual System (HVS), human eyes are more
sensitive to the luminance than the chrominance. Therefore, the accuracy of the
chrominance is reduced to achieve data compaction through sub-sampling, and the
human eyes do not easily perceive the difference. There are four frequently used
YCbCr sub-sampling formats, and all of them are shown in Figure 5.2. It is worth
noting that the name of the format is not always related to the sub-sampling ratio.

Figure 5.2 YCbCr Formats

 134

5.3.4 TMT(Tchebichief Moment Transform)

The image matrixes are partitioned into 2×2 pixels, where the orthogonal
Tchebichef moments are calculated independently. The block size N is
considered as 2.Based on orthogonal moments, a kernel matrix (K 2×2) is given as
follows:

K=
[ t( 0) t (1)
t (2) t (3) ] (5.12)

The image block matrix (F 2×2) with [F(x, y)] denotes the intensity value of
the pixel as:

F=
[ f (0,0) f (0,1)
f (1,0) f (1,1) ] (5.13)
 135

Figure 5.3 8×8 TMT basis function

There are 64 two dimensional TMT basis functions that are generated by
multiplying a horizontal against a vertical set of one-dimensional 8-point TMT
basis functions. In Figure 5.3, neutral gray represents zero, white represents
positive amplitudes, and black represents negative amplitudes. An image contains
low medium and high-frequency components. The low-frequency signal
corresponds to slowly varying color, whereas the high frequencies represent the
detail within the image information. Intuitively, low frequencies are more
important to create a good representation of an image, and the higher frequencies
can largely be ignored to a certain degree. Human eye is highly sensitive to low
frequency distortions rather than too high frequencies. The process of image
reconstruction from its moments is provided as follows:
 136

M −1 N−1
f́ ( x , y )= ∑ ∑ k m ( x ) T mn k n ( y) (5.14)
m=0 n=0

for m=0,1,2,….M- 1 and n=0,1,2,….N-1

Where f́ ( x , y ) denotes the reconstructed intensity distribution.Km represents

the kernel variable of m pixel size, K n represents the kernel variable of n pixel
size. T mn represents the kernel variable of Tchebichef moment transform . The
transforming moment of Tchebichef has its own benefit in the not completely
explored image processing.The discrete orthogonal Tchebichef polynomial
domain consists of real rational numbers, unlike the continuous orthogonal
transform. Discrete orthogonal Tchebichef moment is capable of performing
image reconstruction exactly without any numerical errors. Image reconstruction
provides a measure of the feature representation capability of the moment
functions.

5.3.5 QUANTIZATION

The idea of quantization is to remove the high frequencies of discards

information, which is not visually significant. This is achieved by dividing the
coefficient transform T (u , v ) in each block using quantization matrix Q ( u , v ) . This
process removes the high frequencies present in the original image and comprises
a large number of zeroes as a result of the filtering of high-frequency noise.

T q (u , v )=Round ¿

The quantizer can be classified into a uniform quantizer and non-uniform

quantizer. A uniform quantizer divides the input domain into several equally
spaced intervals. The reconstruction value is taken from the midpoint of the
 137

corresponding interval. One simple uniform quantizer is as follows:

Q ( x) =
[ x
∆ ]
+0.5 , where ∆ is the quantization step¿ ¿

If the quantization step size is 1, then the quantization output function is, as shown
in Figure 5.4, the number of elements in the output domain is finite. On the other
hand, it is easy to understand that the larger the quantization step size, the larger
the loss of the precision.

Q(x)

4.0

3.0

2.0

1.0
0.5 1.5 2.5 3.5

-1.0
-3.5 -2.5 -1.5 -0.5 )
-2.0

-3.0
-4.0

Figure 5.4 The uniform quantity.

In non-uniform quantization, the quantization step is determined according

to the probability of the elements in the input domain. For example, if the
probability of the small value is high, then a small quantization step is assigned to
it. If the probability of the large value is low, then a large quantization step is
assigned to it. Figure5.5 shows one example of the non-uniform quantizer with the
 138

characteristics described above. As can be seen from Figure5.5, the quantization of

small value is smaller, but its probability is high, so that the quantization error can
be reduced efficiently.

Q(x)

4.0

3.0

2.0

1.0
0.5 1.5 2.5 3.5

¿)
-3.5 -2.5 -1.5 -0.5 -1.0

-2.0

-3.0
-4.0

Figure 5.5 Non-uniform quantization

For each layer, the moment coefficients shall be quantized separately. The
quantization process has a key role in JPEG compression, which removes the high
frequencies present in the original image. This is done due to the fact that the eye
is much more sensitive to lower spatial frequencies than to higher frequencies.
 139

This is done by dividing values at high indexes in the vector (the amplitudes of
higher frequencies) with larger values than the values by which the amplitudes of
lower frequencies are divided. The Standard JPEG luminance and chrominance
quantization tables QL and QRrespectively are given below,.

[ ][ ]
16 11 10 16 17 18 24 16
Q L= 12 12 14 19 Q =
R
18 21 26 19
(5.16)
14 13 16 24 24 26 56 24
14 17 22 29 47 66 99 99

The 2-dimensional DCT performed on the 4×4 sub-blocks of the image data
generates 4-bit gains. The quantization table for luminance starts with 24 =16.
However, the Tchebichef moment generates only 2-bit gains. The quantization
tables for Tchebichef moment should start with 22=4 .The proposed luminance and
chrominance quantization tables Q ML andQ MR below are used for Tchebichef
Moment compression. These tables may be generated mathematically in a
friendlier manner.

[ ] [ ]
4 6 8 16 4 8 16 32
Q ML= 6 8 16 32 QMR = 8 16 32 64
(5.17)
8 16 32 64 16 32 64 128
16 32 64 128 32 64 128 256

Eighty traditionally popular images have been selected and the basic
image compression experimental validation has been gone through. These images
 140

are categorized into 40 real images and 40 graphical images. All the images are
raw RGB 3-layer images of size 512×512 pixels.

5.3.6 ZIG-ZAG ORDERING

After the quantization technique, the coefficients are converted into a

stream of binary data, and the DC Coefficients are partitioned with the AC
coefficients. The entire coefficient in each sub-block is arranged in a Zig-Zag
sequence and depicted as a linear array. The main purpose of the Zig-Zag
Scanning is to group the low-frequency coefficients in top of vector, and it is quite
optimal for lossless compression algorithms. The zig-zag output of the quantized
AC coefficient is represented as a series of data set. Next, run-length encoding is
used to minimize the size of a repeating coefficient value in the coefficient set.

Figure 5.6 Zig-Zag Scanning

 141

The zigzag scan is positioned in the transformation process from 8×8

matrix of the quantification process result into 1 × 82 vector, through zigzag scan
reading, as shown in Figure 5.6. The sorting index started from the top left
coefficient and moved in the same direction with an arrow, as observed in Figure
5.6, until it ended at the bottom right DTC matrix coefficient. The efficiency of the
zigzag scan with the mapping method is that it is able to accelerate the time used
for data sorting to group the components from quantified coefficients. An
overview of the mapping process sequence is shown by the scheme in Figure 5.7.
 142

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 62 63

Mapping

0 1 5 6 14 15 27 28 2 4 7 13 16 26 29 42 --- --- 62 63

Input

-26 -3 -6 2 2 -1 0 0 0 -2 -4 1 1 0 0 0 --- --- 0 0

4
0 1 2 3 7 8 9 62 6
5 6 10
11 12 13 14 15

-26 -3 0 -2 -6 2 -4 1 2 -1 --- --- 0 0

Output

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
15 62 63
Figure 5.7 Diagram of the Zig-Zag scan with mapping method
 143

The mapping process is run between the input data sequence and the
zigzag position sequence. The mapping happens according to the same element
position between those two vectors (see mapping arrow in Figure 5.7). Every 64
DCT coefficients from every 8×8-pixel block and the resultant 64 DCT
coefficients must be available before the scanning process. The reading or saving
DCT coefficient input could slow down the scanning process. The architecture of
the zigzag scan is as shown in Figure 5.8.

ZIGZAG
REGISTER 1

MEMORY
2:1 2:1 FOR
MUX MUX SCANNING
ORDER
ZIGZAG_IN ZIGZAG_OUT

ZIGZAG CNT 64 MEMORY

REGISTER 2 READY
SWITCH
MEMORY

COUNTER 64 READY_OUT
CLOCK
RESET

Figure 5.8 Architecture of zigzag scan

 144

5.3.7 HUFFMAN CODING

Huffman coding is a lossless coding technique to generate the shortest

possible code length of the source symbol and the probability of occurrence of
those symbols. Using these probability values, a set of Huffman code for the
source symbols can be generated by Huffman Tree. The Huffman codes are saved
in the Huffman Table. Huffman tables used during the compression process are
stored as header information in the compressed image file in order to decode the
coefficients asymmetrically during the decompression process. For the image with
three components, the encoder can store four sets of Huffman table as shown in
Figure 5.11.
 145

Input Image

Split Equal Rows and Columns

Apply Huffman Coding on Individual

Rows and Columns

Individual Compressed Image

Sum of Compressed Individual Image

Compressed Image

Figure 5.9 Block Diagram of Huffman Coding Compression

The stepwise execution of Huffman coding is shown in Figure 5.9. For

compression of the image, give input as an image, split image into equal rows and
column. Next, apply Huffman coding on individual rows and columns. Individual
 146

rows and columns compress the whole image. The output is produced as a
compressed image.

ALGORITHM

The algorithm which is used for developing to compress an image is in the

following order:

Step 1: Using image box control read image for the process of compression.

Step 2: Call function, which will sort the pixel in Rows and Column. The function
prioritizes the pixel based on the frequency count of each pixel in the image.

Step 3: Call function, which will create an initial heap. Reheap tree according to
the occurrence of each node in the tree, the lower occurrence is attached in a heap.
Create a new node where the left child is lowest in the sorted list, and right is the
second-lowest in the sorted list.

Step 4: Build a Huffman tree based on the prioritized list. Chop-off those two
elements in a sorted list as they are parts of one node and add probabilities. The
result is the probability of the new node.

Step 5: Perform insertion sort on the list with the new node.

Step 6: Repeat STEPS 3, 4, 5 UNTIL only 1 node is left.

Step 7: Perform a traversal of trees to generate a code table. This will determine
the code for each element of tree in the following way.

The code for each symbol obtained by the tracing path of a symbol from the root
of the tree.A1 is assigned to a branch of one direction, and 0 is assigned for a
branch to another direction. For example, a symbol that is reached to branch at
 147

twice; then itis left once represented by pattern '110'. Figure 5.10 depicts codes for
nodes of a sample tree.

(0) (1)

(10) (11)

(110) (111)

Figure 5.10: Huffman Tree

Step 8: First Huffman tree is built. Huffman codes, which require minimum
information to rebuild, may be generated by steps:

Step 1: Length of soft codes resulting from a Huffman tree generated, as

shown in Figure 5.10.
Step 2: Sort the symbols to be encoded by the lengths of their codes (use
symbol value to break ties).
 148

Step 3: Initialize current code as all zeros and assign code values to
symbols from longest to shortest code as follows:

A. If current code length is greater than the length of code for

the current symbol, right shift off extra bits. Assign code to
the current symbol.

B. Increment the code value.

C. Get the symbol with the next longest code.

D. Repeat until all symbols are assigned codes.

Step 9: Encoding pixel of Huffman code has been generated; data is encoded
simply by replacing symbols with its code.

Step 10: Original image is reconstructed i.e., decompression by using Huffman

Decoding.

Step 11: Generate a Huffman tree equivalent to the encoding tree. Huffman code
for some encoded data, decoding can be accomplished by reading encoded data
one bit at a time.

Step 12: Read input character-wise and left to tree until the last element is reached
in the tree.

Step 13: Output the character encodes in the leaf and return to the root, and
continue the step 12 until all the codes of corresponding symbols are known.
 149

Figure 5.11 shows one simple Huffman coding. Firstly, symbols with the
corresponding probability are ordered. Secondly, the symbols a3 and a5are
combined to form a new symbol, and the probability of this symbol is the
combination of its corresponding combined symbols. A similar procedure is
repeated until the final probability becomes 1.0. After that, corresponding
codewords are assigned to each symbol backwardly. Finally, the codeword for
each corresponding symbol is obtained.

Figure 5.11 HuffmanCoding example

5.3.8 IMAGE QUALITY MEASURES

The image reconstruction error can be calculated between the reconstructed image
g(i, j, k) and the original image f(i, j, k) using :

M −1 N −1 2
1
E ( s )=
3 MN
∑ ∑ ∑ ¿∨g (i , j , k )−f ( i, j , k )∨¿(5.18)
i=0 j=0 k=0
 150

Other measurements that represent the reconstruction accuracy are Mean Squared
Error (MSE) - calculate the average of the square of the error.

The MSE is defined as:

M −1 N−1 2
MSE= ∑ ∑ ∑ ¿∨g ( i , j, k )−f ( i , j ,k ) ¿∨¿2 (5.19) ¿
i=0 j =0 k=0

Signal-to-Noise ratio(SNR) is used as a measure of the quality of reconstruction

and compression. The Peak SNR (PSNR) is defined as follows:

( Maxi )
PSNR ( dB )=10 log 10 (5.20)
√ MSE
where Max i is the maximum possible pixel value.

AD is defined as the average difference that occurred between the original image
and reconstructed image, while MD measures the maximum difference that
occurred between the original image and the reconstructed image. The formulae
are defined as :

M −1 N−1 R−1
AD= ∑ ∑ ∑ ¿∨I ( i, j , k ) ∨¿(5.21)
i=0 j =0 k=0

MD=max 0 ≤ i≤ M ∨¿ I ( i, j , k ) (5.22)
 151

5.5 EXPERIMENTAL RESULTS

In this part of the proposed research, an efficient compression technique

based on Discrete Tchebichef Transform (DTT) is proposed and
developed.Default TMT image compression produces clearer and smoothens the
white eye ball surface than JPEG compression. JPEG compression produces ripple
effect on the white eye. The new quantization table on TMT image compression
produces brighter pupil onLena right eye than the previously proposed default
quantizationtable. The TMT image compression, using a new quantization table as
shown in this chapter, serves better visualquality image output than the earlier
TMT image compression. A set of test images is taken to justify the effectiveness
of the algorithm. Figure5.12 (a) shows the original image and Figure 5.12(b)
shows the reconstructed image. Figure5.13 shows various stages of reconstruction
where 5.14(a) and 5.14(b) refer to the input image and initial 16x16 part of the
reconstruction. Fig 5.14(c) and 5.14(d) show the 8x8 reconstruction and
decompressed images. Table 5.1 depicts the quality measurements of various test
images. Figure5.14 shows the comparison of SNR of various input images. From
this graph, it can be analyzed that higher SNR values result in higher compression
performance.

Figure 5.12: Original and Reconstructed Image

 152

Table 5.1. IMAGE QUALITY MEASURES

PARAMETERS Input image 1 Input image Input image

2 3
PSNR 14.1912 13.2848 11.1682
MSE 43.5257 54.1624 70.4904
MD 222.1475 249.0821 249.0485

Figure. 5.13. comparison of SNR (signal noise ratio)

 153

Figure. 5.14 Stages of Reconstruction

5.6 SUMMARY

The function of Tchebichef Moment Transform (TMT) can be used as an

equivalent to DCT for applications in image compression and reconstruction.
The set of Tchebichef Polynomials has the potential to work better for both real-
world imagery and high-end graphics. Two important features of Tchebichef
moments identified are(i) a discrete dimensionality of definition which matches
exactly with the image coordinate, and (ii) absence of numerical approximation
errors for better reconstruction. The experimental results also prove that the
proposed algorithm reduces the time taken to transform images of different sizes
efficiently. Concurrently, it has lower computational complexity since it does not
require any special primitive algorithms as JPEG Compression. This
improvement makes it practical and more elementary implementation for both
software and hardware developers. The future approach of this research is to
implement a similar compression technique using Legendre Moments, which
generate custom quantization tables for low, medium, and high image output
quality levels. The extended model allows a developer to design several
customized quantization tables for a user to choose from according to target
outputspecifications.
 154

CHAPTER 6

PERFORMANCE ANALYSIS

6.1 INTRODUCTION

In this research, the performance evaluation of three stages is described.

The first stage is a mixed approach implementation. The second stage is the
implementation of image compression method. The simulations were performed
using MATLAB simulation environment and were carried out with metrics for
both the proposed protocol and existing. From the comparison, it is clear that the
existing contract has low performance as compared to the proposed tetrolet
 155

transform. In the third stage of this research, an efficient compression technique

based on Discrete Tchebichef Transform (DTT) is proposed and developed.

6.2 PERFORMANCE EVALUATION OF MIXED APPROACH

Experiments were performed on different types of images. First, the images

of size N×N were subjected to wavelet coding. In the resulting coefficient array
matrix, N/2×N/2 elements in the top left corner alone were chosen for submitting
them to Window Average coding. The above procedure was repeated by changing
the block sizes in the wavelet coefficient array matrix. The suitability of the
algorithm was tested for different types of images such as textured (Calfskin) and
un-textured images (Lena), and the Performance Measures were evaluated. For the
un-textured image "Lena", 8 bits per pixel image was given as the input, and the
reconstructed image obtained using different algorithms are shown in Figure. 6.1
and the Performance Measures evaluated are listed in Table 6.2.

Input Image Wavelet Transform

 156

Window Average Mixed Approach

Figure 6.1 - Original (Input) and reconstructed images obtained using various
Image Compression algorithms on “Lena” image

Table 6.1 – Performance Measures obtained for “Lena” image using different
algorithms

Performance Measure
Algorithm

MCA E.R. SNR C.R.

Transform
Coding using 75 0.0195 71.453 4
Wavelet
Spatial Coding using
75 0.0195 71.453 4
Window Average
Mixed Approach 93.75 0.0338 66.775 16
[Transform + Spatial]

Graphical representation of Compression Ratio (C.R.), Signal to Noise Ratio

(SNR) and Error Rate (E.R.) for Transform, Spatial and Mixed Approach are
shown in Figure 6.2.

Technique vs Performance Measures

80
71.45 71.45
70 66.77
60
C.R
50
SNR
40 E.R
30
20 16
10 4 4
0 0.01 0.01 0.03
Transform Spatial Mixed
 157

Figure 6.2 Graphical Representation of Performance Measures for un-

textured image

For textured image" Calfskin", 8 bits per pixel image is given as the input
and the reconstructed image obtained using different algorithms are shown in
Figure 6.3 and the Performance Measures evaluated are listed in Table 6.2.
Graphical representation of the Compression Ratio (CR), Signal to Noise Ratio
(SNR), and Error Rate (ER) for Transform, Spatial, and Mixed Approaches are
shown in Figure 6.4.

Input Image Wavelet Transform

Window Average Mixed Approach

Figure 6.3 - Original (Input) and reconstructed images obtained using various
Image Compression algorithms on “Calfskin” image
 158

Table 6.2 – Performance Measures obtained for “Calfskin” image using

different algorithms

Performance Measure
Algorithm

MCA E.R. SNR C.R.

Transform
Coding using 75 0.0474 67.028 4
Wavelet

Spatial Coding using 75 0.0474 67.028 4

Window Average

Mixed Approach 93.75 0.0821 62.187 16

[Transform + Spatial]

Technique vs Performance Measures

80
67.02 67.02
70 62.18
60 C.R
50 SNR
40 E.R
30
20 16
10 4 4
0.04 0.04 0.0800000000
0 000001
Transform Spatial Mixed

Figure 6.4Graphical Representation of C.R. and SNR for textured image

 159

6.3 PERFORMANCE EVALUATION OF TETROLET TRANSFORM

For determining the performance of the suggested method,PSNR, MSE,

entropy, and the CR parameters are considered. Three separate images, MRI, CT,
and normal image(cameraman), were compressed using the proposed tetrolet
transform. Table 6.3 gives image quality metrics. The specification of the input
image is provided by256×256. With a set of Approximate coefficients after
transformation, the reconstructed image achieves a remarkable PSNR of -19.6598
because the oriented edges are well adapted.

Table 6.3 Image Quality Metrics

Type of Images PSNR MSE ENTROPY CR

MRI -19.6598 284.49 3.35 2.94
CT -15.8503 226.16 4.22 3.50
Normal Image -12.1530 176.28 6.80 4.42

Table 6.3 shows the simulation result of the compression ratio.

Table 6.4 Analysis of Compression Ratio

Parameters Compression Ratio

Tetrolet Haar FQT

MRI 4.12 7.55 4.79

CT 4.24 7.54 4.69

 160

Normal Image 3.12 7.30 3.25

Table 6.4 shows the analysis of the compression ratio. The analysis is
performed in three different images. From the table, it is clear that for all the three
images MRI, CT, and the standard image, the proposed tetrolet transform has a
low compression ratio. When the compression ratio is high, quality of images will
be low. Similarly, the quality of the image will be high if the compression ratio is
low. Thus, the proposed tetrolet transform is capable of generating the highest
quality compressed image as compared to the existing FQT.Figure 6.5 shows the
statistical analysis of the compression ratio. From the graph, it can be found that
the proposed tetrolet transform has a low compression ratio. From the analysis of
the compression ratio for MRI, CT, and normal image, it is clear that the proposed
tetrolet transform has different values. By using tetrolet transform, the normal
image can be compressed in high quality.
 161

Figure 6.5 Comparison Graph of Compression Ratio

Table 6.5 gives the result analysis of the encoding time parameter. The encoding
time is the time taken for compression of the image. Here,three different
techniques are used for compression. Form the table, it is clear that the proposed
tetrolet transform expressesencoding time. This shows that images will be
compressed faster using the tetrolet transform. Thus, the proposed tetrolet
transform shows a faster response when compared to the existing FQT technique.
Figure 6.8 shows the statistical analysis of encoding time. The graph also shows
that by using the tetrolet transform CT image can be encoded faster as compared
to other image sets.
 162

Table 6.5 Analysis of Encoding Time

Parameters Encoding Time

Tetrolet Haar(Kadam FQT (Kadam and

and rathod 2019) rathod 2019)

MRI 6.89 6.5 2.25

CT 7.02 6.13 4.10

Normal Image 6.51 5.76 2.7

Figure 6.6 Comparison Graph of Encoding Time

 163

6.4 PERFORMANCE EVALUATION OF DISCRETE TCHEBICHEF

TRANSFORM (DTT)

A set of test images was taken to justify the effectiveness of the

algorithm. Figure6.7 (a) shows the original image and Figure 6.7(b) shows the
reconstructed image. Here,6.9(a) and 6.9(b) refer the input image and initial 16x16
part of the reconstruction. Fig 6.9(c) and 6.9(d) show the 8×8 reconstruction and
decompressed images. Table 6.8 depicts the quality measurements of various test
images. Figure6.8 shows the comparison of SNR of various input images. From
this graph, it can be analyzed that higher SNR values result in higher compression
performance.

Figure 6.7: Original and Reconstructed Image

Table 6.6 IMAGE QUALITY MEASURES

PARAMETERS IMAGE 1 IMAGE 2 IMAGE 3

PSNR 14.1912 13.2848 11.1682

 164

MSE 43,5257 54.1624 70.4904

MD 222.1475 249.0821 249.0485

Figure 6.8 COMPARISON OF SNR

Figure 6.9 Stages of Reconstruction

 165

CHAPTER 7

CONCLUSION AND FUTURE WORK

Efficient image compression techniques are becoming very vital in areas like
pattern recognition, image processing, system modeling, data mining, etc.
Compression techniques have become the most concentrated area in the field of
computer. Image compression is a technique of efficiently coding digital image to
reduce the number of bits required in representing an image.The image
compression problem arises when there is limited storage and bandwidth for
transmission of images. Image compression remains a challenge for researchers
because it introduces artifacts and causes blurring of the images. This research
discusses the compression technique and transformation. It can be analysed and
observed that the compression techniques are complicated unless identical data
sets and performance measures are used. After analysis of all methods, it is found
that lossless image compression techniques are the most effective among lossy
compression techniques. Further, lossyprovides a higher compression ratio as
compared to lossless. Image compression plays a vital role in saving memory
storage space and saving time while transmission images over the network. Image
compression aims to remove noisy and repeated pixels to compress the image. The
first stage of this research discusses the mixed approach, which is a combination
of spatial and transforms coding techniques. By using this approach tested the
textured and untextured images were tested. Compression ratio(CR), Maximum
compression Achieved(MCA) and signal to noise ratio(SNR) were evaluated.
While using the transform coding and spatial coding Algorithm for image
 166

compression application individually, the value of MCA estimated was 75%, and
ant Compression Ratio was 4:1. But when the techniques were combined in the
mixed approach of the image compression algorithm, it was found that the MCA
increased to 93.75% with Compression Ratio of 1:16. Even though there is a small
variation in the ER and SNR values, the increased value of MCA, compared to the
fall in the quality of image reflected by ER. and SNR is appreciable and
compromising. The Mixed Algorithm offers a 'trade off' between the quality of the
image and also offers a high value of MCA. When tested for different sizes of
images, this Algorithm, gives the same type of result showing its robustness.

The second stage of the proposed work is to develop an efficient computational

tetrolet transform. It is used in lossless compression. The medical image
compression is greatly enhanced all through this research.. The group of tetrolet
coefficients has the ability to work superior for both real data arrays and high-end
graphics. Thus, the results also show that by selecting a proper relaxing parameter,
the low cost of adaptivity using minimum Tetrolet coefficients can be achieved.
The simulation was carried out in MATLAB. From the analysis, it is clear that the
proposed tetrolet transform has high performances compared to other existing
techniques. The results are analysed using wavelet techniques and the noise level
in high-quality image is reduced . This compression technique can be used for
image denoising as well.

The third stage of the proposed work is used to create faster compression
ratio. For that the function of Tchebichef Moment Transform (TMT) can be used
as an equivalent to DCT for applications in image compression and
reconstruction. The set of Tchebichef Polynomials has the potential to work
 167

better for both real-world imagery and high-end graphics. Two important
features of Tchebichef moments identified are (i) a discrete dimensionality of
definition which matches exactly with the image coordinate, and (ii) absence of
numerical approximation errors for better reconstruction. The experimental
results also prove that the proposed algorithm reduces the time taken to
transform images of different sizes efficiently. Concurrently, it has lower
computational complexity since it does not require any special primitive
algorithms as JPEG Compression. This improvement makes it practical and
more elementary implementation for both software and hardware developers.

FUTURE SCOPE

In future works, more focus can be made on the improvement of the

compression ratio using the newer techniques. As the experimental dataset in
this research is somehow limited, application of the developed methods on a
larger dataset could be a subject for future research, which may lead to new
observations and conclusions. Also to implement a similar compression
technique using Legendre Moments, which generate custom quantization tables
for low, medium, and high image output quality levels. The extended model
allows a developer to design several customized quantization tables for a user to
choose from according to target output specifications. The proposed methods
can be further investigated for colour image compression. The implemented
concepts can be applied to develop Region Of Interest (ROI) coding of images.

Further, the work can be extended to video compression. Video data is a

three-dimensional array of color pixels that contain spatial and temporal
redundancy. Similarities can thus be encoded by registering differences within a
 168

frame where the data frame is a set of all pixels that correspond to a single time
moment. With the advancements in compression technology, it is now
straightforward and efficient to compress video files. Thus, the proposed concepts
may be used and future research work can be extended further.
 169

REFERENCES

1. Abdul Hameed, Duraisamy manimegali, Ameerbasha abudhahir,

Subramanian baskar 2016"Evolutionary optimized discrete Tchebichef
moments for image compression applications" Turk J Elec Eng & Comp
Sci Volume 24, pp: 3321 – 3334
2. Abdul-Wahab Sami Ibrahim and Hind Jumaa Sartep2017 “Grayscale
image coloring by using YCbCr and HSV color spaces” International
journal of modern trends in engineering and research ,pp-130-136.
3. A. Abelardo, L. Alparone, and F. Bartolini,1997 "Encoding- interleaved
hierarchical interpolation for lossless image compression," Signal Process,
Vol. 56, Issue. 3, pp. 321–328.
4. Abu Nur Azman, Lang Wong Siaw, Suryana Nanna and Mukundan
Ramakrishnan;2010 "An Efficient Compact Tchebichef moment for
Image Compression," 10th International Conference on Information
Science, Signal Processing and their Applications (ISSPA), pp.448-451
5. Nashat, A. A., & Hassan, N. H. 2016. Image compression based upon
wavelet transform and a statistical threshold. International Conference on
Optoelectronics and Image Processing (ICOIP) (pp. 20-24). IEEE.
6. Ameya K. Naik and Raghunath S. Holambe2013 "Design of Low-
Complexity High-Performance Wavelet Filters for Image" IEEE
Transactions on Image Processing, Vol.22, Issue.5.
7. Amir Averbuch,1996 "Image Compression Using Wavelet Transform and
Multiresolution Decomposition," IEEE Trans. Image Processing; Volume.
5. Issue. 1, ,pp 4-15.
 170

8. Amit Kumar Mandal, Dilip Kumar Baruah2013 "Image Segmentation

Using Local Thresholding And YCbCr Color Space" Int. Journal of
Engineering Research and Applications Volume.3, Issue 6, pp.511-514.

9. A.Ansari and R.S.Anand,2008 "Implementation of Efficient Medical Image

Compression Algorithms with JPEG, Wavelet Transform and SPIHT," Int.
Journal of Computational Intelligence Research and Applications (IJCIRA),
Volume.2, Issue.1, pp. 43-55.
10. M.A. Ansari2008, "DWT Based Context Modeling Of Medical Image
Compression," National Systems Conference, Dec.
11. Anitha. S, Assistant Professor, GAC for women, Tiruchendur, Tamil Nadu,
India, "Lossless Image Compression And Decompression Using Huffman
Coding,".
12. Anju B, Manimurugan S2015 "An Approach to Medical Image
Compression Using Filters Based On Lifting Scheme " IOSR Journal of
VLSI and Signal Processing (IOSR-JVSP) Volume 1, Issue 2,2012, pp: 09-
16
13. Ashlin Deepa R.N, R.Rajeswara Rao2014 “Feature Extraction Techniques
for Recognition of Malayalam Handwritten Characters: Review”
International Journal of Advanced Trends in Computer Science and
Engineering, Volume. 3, Issue.1, pp : 481– 485.
14. Athira B. Kaimal, S. Manimurugan, C.S.C.Devadass2013 "Image
Compression Techniques: A Survey" International Journal of Engineering
Inventions Volume 2, Issue 4, pp: 26-28
15. AvinashGhorpade, PriyankaKatkar,2014 “Image Compression Using Haar
Transform And Modified Fast Haar Wavelet Transform,” IJSTR Volume 3,
Issue 7.
 171

16. Balpreet Kaur, Deepak Aggarwal, Gurpreet Kaur, Amandeep Kaur,”

2011Efficient Image Compression based on Region of Interest” IJCST
Volume. 2, Issue 1, March.
17. Da-Zeng Tian, Ming-Hu Ha2004, "Applications of wavelet transform in
medical image processing," IEEE Proc. of Int. Conf. on Machine Learning
and Cybernetics, Vol.3,pp.1816– 1821.
18. Deepika Choudhary, Anjali Lather, Sandeep Yadav2014 "Overview of
Image Processing" International Journal for Research in Applied Science &
Engineering Technology(IJRASET) Volume 2,:pp:120-125.

19. R.A. DeVore; B. Jawerth; B.J. Lucier1992 “Image compression through

wavelet transforms coding” IEEE Transactions on Information Theory,
Volume: 38, Issue: 2,:pp-719 – 746.
20. Dinesh V. Rojatkar, Nitesh D. Borkar, Buddhabhushan R. Naik, Ravindra
N. Peddiwar2015 “Image Compression Techniques: Lossy and Lossless”
International Journal of Engineering Research and General Science Volume
3, Issue 2, pp-912-917

21. Donati, S. Yagain2013 "A Comparative Study of Effects of CSC on Image

Compression Ratios While Using JPEG-XR,"pp. 158-161.
22. Feig and Heidi Peterson vireshRatnakar1995, "Image compression using
Spatial prediction." IEEE International Conference on Digital Object
Identifier.
23. Garima Chopra and A. K. Pal2011 "An Improved Image Compression
Algorithm Using Binary Space Partition Scheme and Geometric Wavelets"
IEEE TRANSACTIONS on image processing, Vol.20, Issue.1,.

24. dr. Ghadah Al-Khafaji, rafaa Ismael Yahya2017 “Lossy Image

Compression Using Wavelet Transform, Polynomial Prediction And Block
 172

Truncation Coding” IOSR Journal of Computer Engineering (IOSR-JCE)

Volume 19, Issue 4, pp 34-38.
25. M Gunapriya, A Sivanantha Raja2014 "Compression of Color Medical
Images with Different Color Spaces" International Journal of
Computational Intelligence and Informatics, Volume: 3, Issue: 4, pp 233-
238.

26. Gunpreet Kaur, Mandeep Kaur2011"Performance Evaluation of ROI-based

Image Compression Techniques," International Journal of Scientific &
Engineering Research (IJSER), ISSN: 2229-5518, Vol- 2, Issue 10.
27. Gurparkash Singh Kang2010 “Blind digital image watermarking using
adaptive casting energy in different resolutions of wavelet transform” IEEE
International Conference on Digital Image Identifier,pp.210-215.
28. Hidayah Rahmalan, Nur Azman Abu and saw lang wong2010 "using
tchebichef moment for fast and efficient image compression" Pattern
recognition and image analysis, Volume 20, Issue 4,2010, pp 505-512.
29. Honghai Yu and Stefan Winkler2013, "Image Complexity and Spatial
Information," International Workshop on Quality of Multimedia
Experience.
30. Hugo R. Albuquerque, Tsang Ing Ren, and George D. C. Cavalcanti2012
"Image Fusion Combining Frequency Domain Techniques Based on Focus"
IEEE 24th International Conference on Tools with Artificial Intelligence.
31. A.K.Jain2011 “Digital Image Process Fundamentals of Digital Image
Processing,” E.FatemiZadeh, Sharif University of Technology.
32. Janaki. R, Dr.Tamilarasi.2011An "Enhanced Lossy Techniques for
Compressing Background Region of Medical Images Using ROI-Based
Encoding" IJCSET Volume 1, Issue 11, pp- 690-695
 173

33. Jens Krommweh2010 “Tetrolet Transform: A New Adaptive Haar Wavelet

Algorithm for Sparse Image Representation” Volume 21, Issue 4, pp 364-
374
34. Jyoti A. Patil, Dipali M. Lohar, Prachi M. Badgujar, Dipmala V. Patil
2018“Lossless Image Compression Using Huffman Coding Algorithm”
International Journal of Innovative Research in Computer and
Communication Engineering Volume. 6, Issue 4.
35. Kamrul Hasan Talukde, Koichi Harada2010 "Haar Wavelet-Based
Approach for Image Compression and Quality Assessment of Compressed
Image" International Journal of Applied Mathematics, Volume 36, Issue 1.
36. Kaushik Deb, Md. Sajib Al-Seraj, Md. Moshiul Hoque, and Md. Iqbal
HasanSarkar2012 "Combined DWT-DCT Based Digital Image
Watermarking Technique for Copyright Protection," International
Conference on Electrical and Computer Engineering.
37. H.B. Kekre, Tanuja Sarode, Prachi Natu2013 “Image Compression Based
on Hybrid Wavelet Transform Generated using Orthogonal Component
Transforms of Different Sizes” International Journal of Soft Computing and
Engineering (IJSCE) Volume-3, Issue-3,:pp:53-57.
38. Krivoulets, A. 2002. Efficient entropy coding for image
compression. Technical Report ITU-TR, 2002-13.
39. Lang Wong Siaw, Abu Nur Azman, Rahmalan Hidayah2009; "Fast 4x4
Tchebichef Moment Image Compression", International Conference of
Soft Computing Pattern Recognition (SoCPaR), pp. 295-300.
40. Lavanya.A,K.vani,S.Sanjeev2011, "Image fusion of multi-sensor Lunar
image data using Wavelet combined transformation"IEEE International
Conference on recent trend in information technology.
 174

41. MacarenaBoix, BegoñaCantó2010 “Wavelet Transform application to the

compression of images” Mathematical and Computer ModellingVolume
52, Issues 7–8, pp:1265-1270.
42. Mahindra Singh, Meenakshi Garg"2014Mixed DWT-DCT Approached
Based Image Compression Technique" International Journal Of
Engineering And Computer Science Volume 3, Issue 11, pp. 9008-9111
43. Mahipalsinh R. Zala, Shaktisinh S. Parmar,2013 "A New Image
Compression Scheme with Wavelet Transform and SPIHT Algorithm
using PSO," International Journal of Engineering Research and
Applications (IJERA), Vol. 3, Issue 2, March -April,pp.272-275.
44. Manjari Singh, Sushil Kumar, Siddharth Singh, Manish2016 "Various
Image Compression Techniques: Lossy and Lossless" International Journal
of Computer Application Volume 142 – Issue.6, pp:23-26.
45. R.Mehala, K.Kuppusamy2013 “A New Image Compression Algorithm
using Haar Wavelet Transformation” International Journal of Computer
Applications pp:1-4.
46. Mukundan R., Ong S.H., and Lee P.A.2001; "Image Analysis by
Tchebichef Moments," IEEE Transactions on Image Processing, Vol. 10,
No. 9,, pp. 1357–1364.
47. Hunt, O., & Mukundan, R. 2004. A comparison of discrete orthogonal
basis functions for image compression.Proceedings Conference on Image
and Vision Computing, New Zealand, pp. 53-58.
48. Markandeya, N., & Patil, S. 2017. Optimizing Information Rate in
Polynomial Based Image Secret Sharing Scheme Using Huffman
Coding.International journal of engineering and computer science.Vol.6.
 175

49. Sethi, N., Krishna, R., & Arora, R. P. 2011. Image compression using Haar
wavelet transform. Computer Engineering and Intelligent Systems, issue
2(3), PP.1-5.
50. Nikkoo Khalsa, Dr. G. G. Sarate, Prof. D. T. Ingole2010, "Factors
Influencing The Image Compression Of Artificial And Natural Image
Using Wavelet Transform," International Journal of Engineering Science
and Technology (IJEST), ISSN: 6225-6233 Volume. 2 Issue 11.
51. B.Niteesh, V.Ravi Kumar, R, Sai Ram, P.Krishna Sagar, B.Hemanth Nag
2016"Image Compression Using Adaptive Haar Wavelet-Based Tetrolet
Transform" International Journal of Innovative Research in Computer and
Communication Engineering Volume. 4, Issue 4,, pp-6522-6528.
52. Nur Azman Abu; Siaw Lang Wong; Nanna Suryana
Herman; Ramakrishnan Mukundan2010 “An efficient compact Tchebichef
Moment for image compression” International Conference on Information
Science, Signal Processing and their Applications,
53. G. A. Papakostas, D. A., Karras, B. G. Mertzios, and Y. S. Boutalis2005
"An Efficient Feature Extraction Methodology for Computer Vision
Applications using Wavelet Compressed Zernike Moments" icgst, pp 5-15

54. Paulo A. M. Oliveira, Renato J. Cintra, Fábio M. Bayer, Sunera

Kulasekera2015 A “Discrete Tchebichef Transform Approximation for
Image and Video Coding” IEEE Signal Processing Letters Volume: 22,
Issue: 8, pp:1137 – 1141.
55. B. Penna, Tammam Tillo, Enrico Magli, G. Olmo2007 “Transform Coding
Techniques for Lossy Hyperspectral Data Compression” IEEE
Transactions on Geoscience And Remote Sensing, Volume. 45, Issue. 5,:pp
1409-1421
 176

56. J. H. Pujar and L. M. Kadlaskar2010, “A New Lossless Method of Image

Compression and Decompression Using Huffman Coding Technique,”
Journal of Theoretical and Applied Information Technology, Vol. 15, Issue.
1,.
57. Puja D Saraf, Deepti Sisodia, Amit Sinhal and Neetesh Gupta,
2012"Comparisons Of Wavelets Based Image Compression Methods,"
World Journal of Science and Technology, ISSN: 2231 – 2587.
58. Rachit Patel, Virendra Kumar, Vaibhav Tyagi, Vishal Asthana,2016 "A
Fast and Improved Image Compression Technique Using Huffman
Coding," IEEE,, pp: 2283-2286.
59. T. M. P. Rajkumar, Mrityunjaya V Latte,2011 "ROI Based Encoding of
Medical Images: An Effective Scheme Using Lifting Wavelets and SPIHT
for Telemedicine," Computer Theory and Engineering, Vol. 3, No. 3.
60. Ranjan K Senapati, Umesh C Pati, Kamala K Mahapatra,2011 "A Novel
Hybrid HVS Based Embedded Image Coding Algorithm Using DTT and
SPIHT," International Conference on Devices and
Communication(ICDeCom).
61. Ranjan K. Senapati, Umesh C. Pati, Kamala K. Mahapatra2010 "Image
Compression Using Discrete Tchebichef Transform
Algorithm"International Conference on Advances in Communication,
Network, and Computing, pp 104-108

62. Ranjan Senapati, Umesh C. Pati, Kamalakanta Mahapatra2013 “An

Efficient Sparse 8× 8 Orthogonal Transform Matrix for Color Image
Compression” Int.J. Signal and Imaging Systems Engineering,
Vol.6.Issue.1,:pp.16-23.
 177

63. P. Ravi, A. Ashokkumar2017 “Analysis of Various Image Processing

Techniques” International Journal of Advanced Networking & Applications
(IJANA) Volume: 8, Issue: 5, pp: 86-89

64. S. A. Raza Naqvi2013 "Image compression using haar wavelet-based

Chevrolet transform" International Conference on Open Source Systems
and Technologies,.
65. J. S. Rivera-López, C. J. Camacho-Bello2017 "Color Image
Reconstruction by Discrete Orthogonal Moment" Journal of Data Analysis
and Information Processing, Volume 5, , pp:156-166
66. Robby Candra, Saifuddin Madenda, Sunny Arief Sudiro, and Muhammad
Subali2017 “The Implementation of an Efficient Zigzag Scan” Volume. 9
Issue. 2, pp 95-98
67. Ruchika, Mooninder Singh, Anant Raj Singh,2012 "Compression of
Medical Images Using Wavelet Transforms," International Journal of Soft
Computing and Engineering (IJSCE), Volume-2, Issue-2.
68. A. Said and W. A. Pearlman,1996 “A new, fast, and efficient image codec
based on set partitioning in hierarchical trees,” IEEE Trans. Circuits Syst.
Video Technol., Vol. 6, no. 3,:pp. 243–250.
69. Sedig A. et al.,2007 "Semi-automatic region of interest identification
algorithm using wavelets," J. of Optical Engineering, Vol-46, Issue 3,:pp-
035003-6.
70. A. A. Shaikh, P. P. Gadekar,2015 “Huffman Coding Technique for Image
Compression," ISSN,.
71. Shabnam Tamboli, V R Udupi2013“Image Compression Using Haar
Wavelet Transform” International Journal of Advanced Research in
Computer and Communication Engineering Volume. 2, Issue 8,:pp-3166-
3170.
 178

72. I. Singh, P.Agathoklis, and A.Antony1998 "Image compression using a

mixed-transform technique and vector quantization."IEEE International
Conference on Digital Object Identifier,.
73. Sonja Grgic, Mislav Grgic, BrankaZovko-Cihlar2001 “Performance
Analysis of Image Compression Using Wavelets” IEEE Transactions on
IndustrialElectronics, Vol. 48,Issue. 3,:pp-682-695.

74. Mr. T.Sreenivasulu reddy Ms. K. Ramani Dr. S. Varadarajan Dr.

B.C.Jinaga2007 “Image Compression Using Transform Coding Methods”
IJCSNS International Journal of Computer Science and Network Security,
Volume.7 Issue.7.
75. S. Sridhar, P. Rajesh Kumar, and K.V.Ramanaiah,2014 Wavelet Transform
Techniques for Image Compression - An Evaluation, U.Image, Graphics
and Signal Processing Volume 2,:pp- 54-67.

76. Sridhar S, Rajesh Kunar P, Ramanaiah K V2014 "High-Performance DWT

for Hybrid Image Compression System" International Journal of Computer
Science and Network, Volume 3, Issue 6, pp:584-592.

77. Sudeep D. Thepade andShwetaliErandole2013 “Extended performance

Comparison of Tilling based Image Compression using Wavelet Transform
and Hybrid Wavelet Transform."IEEE International Conference on Digital
Image Identifier, pp .1150-1155.
78. R.Sumalatha, M.V.Subramanyam,2011 "Medical Image Compression
Using Multiwavelets For Telemedicine Applications," International Journal
of Scientific & Engineering Research (IJSER), ISSN: 2229-5518, Vol-2,
Issue-9.
 179

79. J. Taquet and C. Labit,2010 "Near-lossless and scalable compression for

medical imaging using a new adaptive hierarchical oriented prediction," in
Proc. 17th IEEE ICIP, pp. 481–484.
80. D.Taubman2000, "High-performance, scalable image compression with
EBCOT," IEEE Trans. Image Process., Vol. 9, no. 7, Jul., pp. 1158–1170.
81. S. Udomhunsakul and K. Hamamoto,2004 "Wavelet filters comparison for
ultrasonic image compression," Conf. IEEE TENCON, Vol. 1,2, pp. 171-
174.
82. Umhara Rasool Khan, Sanna Mairaj, Tamanna Nazeer, Suhaib Ahmed2017
"Wavelet-Based Image Compression Techniques: Comparative Analysis
and Performance Evaluation" International Journal of Emerging
Technologies in Engineering Research (IJETER) Volume 5, Issue 9, , pp-
9-13.
83. M. Unser1995, "Texture classification and segmentation using wavelet
frames," IEEE Trans. Image Processing, Vol- 4, no. 11, pp. 1549-1560.
84. Vijaya Kumar C N, Kumar D Anil Kumar R2017 “Performance Analysis of
Image Compression Using Discrete Wavelet Transform” International
Journal of Advanced Research in Computer Science and Software
Engineering Volume 7, Issue 3, pp:213-218.
85. Vikash Kumar, Sanjay Sharma2017 “Lossless Image Compression through
Huffman Coding Technique and Its Application in Image Processing using
MATLAB” International Journal of Soft Computing and Engineering
(IJSCE) Volume-7 Issue-1, pp:10-13.
86. Walaa M. Abd-Elhafiez, Wajeb Gharibi2012“Color Image Compression
Algorithm Based on the DCT Blocks”:pp 1-5.
 180

87. Weinberger, M.J., Seroussi, G., and Sapiro, G,2000, "The LOCO-I lossless
image compression algorithm: principles and standardization into JPEG-
LS," IEEE Transactions on Image Processing, Vol. 9, Issue 8, pp.1309-
1324.
88. Yen-Yu Chen2007, "Medical image compression using DCT-based
subband decomposition and modified SPIHT data organization,"
International Journal Of Medical Informatics, pp. 717– 725.

LIST OF PUBLICATION
International Journal
1. S.UmaMaheswari and Dr.V.SrinivasaRaghavan, “Lossless MedicalImage
Compression Algorithm Using Tetrolet transformation”, Journal ofAmbient
Intelligence and Humanized Computing. DOI
https://doi.org/10.1007/s12652-020-01792-8(Accepted).