Computer Graphics and Fundamentals of Image Processing(21CS63)

MODULE-4
Overview of Image Processing

Computers are faster and more accurate than human beings in processing numerical data.
However, human beings score over computers in recognition capability. The human brain is so
sophisticated that we recognize objects in a few seconds without much difficulty. Human beings
use all the five sensory organs to gather knowledge about the outside world. Among these
perceptions, visual information plays a major role in understanding the surroundings. Other kinds
of sensory information are obtained from hearing, taste, smell and touch.

With the advent of cheaper digital cameras and computer systems, we are witnessing a powerful
digital revolution, where images are being increasingly used to communicate effectively.

Images are encountered everywhere in our daily lives. We see many visual information sources
such as paintings and photographs in magazines, journals, image galleries, digital libraries,
newspapers, advertisement boards, television, and the Internet. Many of us take digital snaps of
important events in our lives and preserve them as digital albums. Then from the digital album,
we print digital pictures or mail them to our friends to share our feelings of happiness and
sorrow. Images are not used merely for entertainment purposes. Doctors use medical images to
diagnose problems for providing treatment. With modern technology, it is possible to image
virtually all anatomical structures, which is of immense help to doctors in providing better
treatment. Forensic imaging application process fingerprints, faces and irises to identify
criminals. Industrial applications use imaging technology to count and analyse industrial
components. Remote sensing applications use images sent by satellites to locate the minerals
present in the earth.

Images are imitations of real-world objects. Image is a two-dimensional (2D) signal f(x,y), where
the values of the function f(x,y) represent the amplitude or intensity of the image. For processing
using digital computers, this image has to be converted into a discrete form using the process of
sampling and quantization, known collectively as digitization. In image processing, the term
‘image’ is used to denote the image data that is sampled, quantized and readily available in a
form suitable for further processing by digital computers. Image processing is an area that deals
with manipulation of visual information.

Page 1
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Major objectives of image processing is to

 Improve the quality of pictorial information for better human interpretation.

 Facilitate the automatic machine interpretation of images.

Nature of Image Processing

There are three scenarios or ways of acquiring an image

1. Reflective mode Imaging

2. Emissive Type Imaging
3. Transmissive Imaging

The radiation source shown in Figure 4.1 is the light source.

Figure 4.1: Image processing environment

Objects are perceived by the eye because of light. The sun, lamps, and clouds are all examples of
radiation or light sources. The object is the target for which the image needs to be created. The
object can be people, industrial components, or the anatomical structure of a patient. The objects
can be two-dimensional, three-dimensional or multidimensional mathematical functions
involving many variables. For example, a printed document is a 2D object. Most real-world
objects are 3D.

Page 2
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Reflective Mode Imaging

Reflective mode imaging represents the simplest form of imaging and uses a sensor to acquire
the digital image. All video cameras, digital cameras, and scanners use some types of sensors for
capturing the image. Image sensors are important components of imaging systems. They convert
light energy to electric signals.

Emissive Type Imaging

In Emissive type imaging , images are acquired from self-luminous objects without the help of a
radiation source . In emissive type imaging, the objects are self-luminous. The radiation emitted
by the object is directly captured by the sensor to form an image. Thermal imaging is an example
of emissive type imaging. In thermal imaging, a specialized thermal camera is used in low light
situations to produce images of objects based on temperature. Other examples of emissive type
imaging are magnetic resonance imaging (MRI) and positron emissive tomography (PET)

Transmissive Imaging

In Transmissive imaging, the radiation source illuminates the object. The absorption of radiation
by the objects depends upon the nature of the material. Some of the radiation passes through the
objects. The attenuated radiation is sensed into an image. This is called transmissive imaging.
Examples of this kind of imaging are X-ray imaging, microscopic imaging, and ultrasound
imaging.

The first major challenge in image processing is to acquire the image for further processing.
Figure 4.1 shows three types of processing – optical, analog and digital image processing.

Optical Image Processing

Optical image processing is the study of the radiation source, the object, and other optical
processes involved. It refers to the processing of images using lenses and coherent light beams
instead of computers. Human beings can see only the optical image. An optical image is the 2D
projection of a 3D scene. This is a continuous distribution of light in a 2D surface and contains
information about the object that is in focus. This is the kind of information that needs to be
captured for the target image. Optical image processing is an area that deals with the object,

Page 3
 Computer Graphics and Fundamentals of Image Processing(21CS63)

optics, and how processes are applied to an image that is available in the form of reflected or
transmitted light. The optical image is said to be available in optical form till it is converted into
analog form.

Analog Image Processing

An analog or continuous image is a continuous function f(x,y) where x and y are two spatial
coordinates. Analog signals are characterized by continuous signals varying with time. They are
often referred to as pictures. The processes that are applied to the analog signal are called analog
processes. Analog image processing is an area that deals with the processing of analog electrical
signals using analog circuits. The imaging systems that use film for recording images are also
known as analog imaging systems.

Digital Image Processing

The analog signal is often sampled, quantized and converted into digital form using digitizer.
Digitization refers to the process of sampling and quantization. Sampling is the process of
converting a continuous-valued image f(x,y) into a discrete image, as computers cannot handle
continuous data. So the main aim is to create a discretized version of the continuous data.
Sampling is a reversible process, as it is possible to get the original image back. Quantization is
the process of converting the sampled analog value of the function f(x,y) into a discrete-valued
integer. Digital image processing is an area that uses digital circuits, systems and software
algorithms to carry out the image processing operations. The image processing operations may
include quality enhancement of an image, counting of objects, and image analysis.

Digital image processing has become very popular now as digital images have many advantages
over analog images. Some of the advantages are as follows:

1. It is easy to post-process the image. Small corrections can be made in the captured image
using software.
2. It is easy to store the image in the digital memory.
3. It is possible to transmit the image over networks. So sharing an image is quite easy.
4. A digital image does not require any chemical process. So it is very environment
friendly, as harmful film chemicals are not required or used.

Page 4
 Computer Graphics and Fundamentals of Image Processing(21CS63)

5. It is easy to operate a digital camera.

The disadvantages of digital images are very few. Some of the advantages are the initial cost,
problems associated with sensors such as high power consumption and potential equipment
failure, and other security issues associated with the storage and transmission of digital images.

The final form of an image is the display image. The human eye can recognize only the optical
form. So the digital image needs to be converted to optical form through the digital to analog
conversion process.

Image Processing and Related Fields

Image processing is an exciting interdisciplinary field that borrows ideas freely from many
fields. Figure 4.2 illustrates the relationships between image processing and other related fields.

Figure 4.2: Image Processing and other closely related fields

1) Image Processing and Computer Graphics

Computer graphics and image processing are very closely related areas. Image processing deals
with raster data or bitmaps, whereas computer graphics primarily deals with vector data. Raster
data or bitmaps are stored in a 2D matrix form and often used to depict real images. Vector

Page 5
 Computer Graphics and Fundamentals of Image Processing(21CS63)

images are composed of vectors, which represent the mathematical relationships between the
objects. Vectors are lines or primitive curves that are used to describe an image. Vector graphics
are often used to represent abstract, basic line drawings.

The algorithms in computer graphics often take numerical data as input and produce an image as
output. However, in image processing, the input is often an image. The goal of image processing
is to enhance the quality of the image to assist in interpreting it. Hence, the result of image
processing is often an image or the description of an image. Thus, image processing is a logical
extension of computer graphics and serves as a complementary field.

2) Image Processing and Signal Processing

Human beings interact with the environment by means of various signals. In digital signal
processing, one often deals with the processing of a one-dimensional signal. In the domain of
image processing, one deals with visual information that is often in two or more dimensions.
Therefore, image processing is a logical extension of signal processing.

3) Image Processing and Machine Vision

The main goal of machine vision is to interpret the image and to extract its physical, geometric,
or topological properties. Thus, the output of image processing operations can be subjected to
more techniques, to produce additional information for interpretation. Artificial vision is a vast
field, with two main subfields –machine vision and computer vision. The domain of machine
vision includes many aspects such as lighting and camera, as part of the implementation of
industrial projects, since most of the applications associated with machine vision are automated
visual inspection systems. The applications involving machine vision aim to inspect a large
number of products and achieve improved quality controls. Computer vision tries to mimic the
human visual system and is often associated with scene understanding. Most image processing
algorithms produce results that can serve as the first input for machine vision algorithms.

4) Image Processing and Video processing

Image processing is about still images. Analog video cameras can be used to capture still images.
A video can be considered as a collection of images indexed by time. Most image processing
algorithms work with video readily. Thus, video processing is an extension of image processing.

Page 6
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Images are strongly related to multimedia, as the field of multimedia broadly includes the study
of audio, video, images, graphics and animation.

5) Image Processing and Optics

Optical image processing deals with lenses, light, lighting conditions, and associated optical
circuits. The study of lenses and lighting conditions has an important role in study of image
processing.

6) Image Processing and Statistics

Image analysis is an area that concerns the extraction and analysis of object information from the
image. Imaging applications involve both simple statistics such as counting and mensuration and
complex statistics such as advanced statistical inference. So statistics plays an important role in
imaging applications. Image understanding is an area that applies statistical inferencing to extract
more information from the image.

Digital Image Representation

An image can be defined as a 2D signal that varies over the spatial coordinates x and y, and can
be written mathematically as f(x,y). Medical images such as magnetic resonance images and
computerized tomography(CT) images are 3D images that can be represented as f(x,y,z), where
x,y, and z are spatial coordinates. A simple digital image and its matrix equivalent are shown in
Figs 4.3(a) and 4.3(b).

(a) (b)

Figure 4.3: Digital Image Representation (a) Small binary digital image (b) Equivalent
image contents in matrix form

Page 7
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Figure 4.3(a) shows a displayed image. The source of the image is a matrix as shown in Fig.
4.3(b). The image has five rows and five columns. In general, the image can be written as a
mathematical function f(x,y) as follows:

In general, the image f(x,y) is divided into X rows and Y columns. Thus, the coordinate ranges
are {x=0,1,……X-1} and {y=0,1,2,…….Y-1}. At the intersection of rows and columns, pixels
are present. Pixels are the building blocks of digital images. Pixels combine together to give a
digital image. Pixel represents discrete data. A pixel can be considered as a single sensor,
photosite(physical element of the sensor array of a digital camera), element of a matrix, or
display element on a monitor.

The value of the function f(x,y) at every point indexed by row and a column is called grey value
or intensity of the image. The value of the pixel is the intensity value of the image at that point.
The intensity value is the sampled, quantized value of the light that is captured by the sensor at
that point. It is a number and has no units.

The number of rows in a digital image is called vertical resolution. The number of columns is
called horizontal resolution. The number of rows and columns describes the dimensions of the
image. The image size is often expressed in terms of the rectangular pixel dimensions of the
array. Images can be of various sizes. Some examples of image size are 256 X 256, 512 X 512.
For a digital camera, the image size is defined as the number of pixels (specified in megapixels)

Resolution is an important characteristic of an imaging system. It is the ability of the imaging

system to produce the smallest discernable details, that is the smallest sized object clearly and
differentiate it from the neighbouring small objects that are present in the image. Image
resolution depends on two factors- optical resolution of the lens and spatial resolution.

Spatial resolution of the image is very crucial as the digital image must show the object and its
separation from the other spatial objects that are present in the image clearly and precisely.

Page 8
 Computer Graphics and Fundamentals of Image Processing(21CS63)

A useful way to define resolution is the smallest number of line pairs per unit distance. The
resolution can then be quantified as 200 line pairs per mm.

Spatial resolution depends on two parameters – the number of pixels of the image and the
number of bits necessary for adequate intensity resolution, referred to as the bit depth. The
numbers of pixels determine the quality of the digital image. The total number of pixels that are
present in the digital image is the number of rows multiplied by the number of columns.

The choice of bit depth is very crucial and often depends on the precision of the measurement
system. To represent the pixel intensity value, certain bits are required. For example, in binary
images, the possible pixel values are 0 or 1. To represent two values, one bit is sufficient. The
number of bits necessary to encode the pixel value is called bit depth. Bit depth is a power of
two. It can be written as 2 m . In monochrome grey scale images (e.g medical images such as X-
rays and ultrasound images), the pixel values can be between 0 and 255. Hence, eight bits are
used to represent the grey shades between 0 and 255.(as 2 8 =256). So the bit depth of grey scale
images is 8. In colour images, the pixel value is characterized by both colour value and intensity
value. So colour resolution refers to the number of bits used to represent the colour of the pixel.
The set of all colours that can be represented by the bit depth is called gamut or palette.

So, the total number of bits necessary to represent the image is

Number of rows x Number of columns x Bit depth

Spatial resolution depends on the number of pixels present in the image and the bit depth.
Keeping the number of pixels constant but reducing the quantization levels(bit depth) leads to
phenomenon called false contouring. The decrease in number of pixels while retaining the
quantization levels leads to a phenomenon called checkerboard effect (or pixelization error).

A 3D image is a function f(x,y,z) where x,y, and z are spatial coordinates. In 3D images, the
term ‘voxel’ is used for pixel. Voxel is an abbreviation of ‘volume element’.

Types of Images
Images can be classified based on many criteria.

Page 9
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Figure 4.4: Classification of Images

Based on Nature

Images can be broadly classified as natural and synthetic images. Natural images are images
of the natural objects obtained using devices such as cameras or scanners. Synthetic images are
images that are generated using computer programs.

Based on Attributes

Based on attributes, images can be classified as raster images and vector graphics. Vector
graphics use basic geometric attributes such as lines and circles, to describe an image. Hence the
notion of resolution is practically not present in graphics. Raster images are pixel-based. The
quality of the raster images is dependent on the number of pixels. So operations such as
enlarging or blowing-up of a raster image often result in quality reduction.

Based on Colour

Based on colour, images can be classified as grey scale, binary, true colour and pseudocolour
images.

Page 10
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Grayscale and binary images are called monochrome images as there is no colour component in
these images. True colour(or full colour) images represent the full range of available colours. So
the images are almost similar to the actual object and hence called true colour images. In
addition, true colour images do not use any lookup table but store the pixel information with full
precision. Pseudocolour images are false colour images where the colour is added artificially
based on the interpretation of the data.

i) Grey scale Images

Grey scale images are different from binary images as they have many shades of grey between
black and white. These images are also called monochromatic as there is no colour component in
the image, like in binary images. Grey scale is the term that refers to the range of shades between
white and black or vice versa.

Eight bits (28=256) are enough to represent grey scale as the human visual system can
distinguish only 32 different grey levels. The additional bits are necessary to cover noise
margins. Most medical images such as X-rays, CT images, MRIs and ultrasound images are grey
scale images. These images may use more than eight bits. For example, CT images may require a
range of 10-12 bits to accurately represent the image contrast.

(a) (b)

Figure 4.5: Monochrome images (a) Grey scale image (b) Binary image

ii) Binary Images

Page 11
 Computer Graphics and Fundamentals of Image Processing(21CS63)

In binary images, the pixels assume a value of 0 or 1. So one bit is sufficient to represent the
pixel value. Binary images are also called bi-level images. In image processing, binary images
are encountered in many ways.

The binary image is created from a grey scale image using a threshold process. The pixel value
is compared with the threshold value. If the pixel value of the grey scale image is greater than the
threshold value, the pixel value in the binary image is considered as 1.Otherwise, the pixel value
is 0. The binary image is created by applying the threshold process on the grey scale image in
Fig. 4.5(a) is displayed in Fig. 4.5(b) . It can be observed that most of the details are eliminated.
However, binary images are often used in representing basic shapes and line drawings. They are
also used as masks. In addition, image processing operations produce binary images at
intermediate stages.

iii) True Colour Images

In true colour images, the pixel has a colour that is obtained by mixing the primary colours
red,green and blue. Each colour component is represented like a grey scale image using eight
bits. Mostly, true colour images use 24 bits to represent all the colours. Hence true colour images
can be considered as three-band images. The number of colours that is possible is 256 3 (i.e
256x256x256=1,67,77,216 colours)

Figure 4.6 a) shows a colour image and its three primary colour components. Figure 4.6 b
illustrates the general storage structure of the colour image. A display controller then uses a
digital-to-analog converter(DAC) to convert the colour value to the pixel intensity of the
monitor.

Original Image Red Component

Page 12
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Green Component Blue Component

(a)

Figure 4.6: True colour images (a) Original image and its colour components

(b)

(c)

Figure 4.6: (b) Storage structure of colour images (c) Storage structure of an
indexed image

Page 13
 Computer Graphics and Fundamentals of Image Processing(21CS63)

A special category of colour images is the indexed image. In most images, the full range of
colours is not used. So it is better to reduce the number of bits by maintaining a colour map,
gamut, or palette with the image. Figure 4.6(c) illustrates the storage structure of an indexed
image. The pixel value can be considered as a pointer to the index, which contains the address of
the colour map. The colour map has RGB components. Using this indexed approach, the number
of bits required to represent the colours can be drastically reduced. The display controller uses a
DAC to convert the RGB value to the pixel intensity of the monitor.

iv)Pseudocolour Images

Like true colour images, pseudocolour images are also widely used in image processing. True
colour images are called three-band images. However, in remote sensing applications multi-band
images or multi-spectral images are generally used. These images, which are captured by
satellites contains many bands. A typical remote sensing image may have 3-11 bands in an
image. This information is beyond the human perceptual range. Hence it is mostly not visible to
the human observer. So colour is artificially added to these bands, so as to distinguish the bands
and to increase operational convenience. These are called artificial colour or pseudocolour
images. Pseudocolour images are popular in the medical domain also. For example, the Doppler
colour image is a pseudocolour image.

Based on Dimensions

Images can be classified based on dimension also. Normally, digital images are 2D rectangular
array of pixels. If another dimension, of depth or any other characteristics, is considered, it may
be necessary to use a higher-order stack of images. A good example of a 3D image is a volume
image, where pixels are called voxels. By ‘3D image’, it is meant that the dimension of the target
in the imaging system is 3D. The target of the imaging system may be a scene or an object. In
medical imaging, some of the frequently encountered images are CT images, MRIs and
microscopy images. Range images, which are often used in remote sensing applications, are also
3D images.

Based on Data Types

Images may be classified based on their data type. A binary image is a 1-bit image as one bit is
sufficient to represent black and white pixels. Grey scale images are stored as one-byte(8-bit) or

Page 14
 Computer Graphics and Fundamentals of Image Processing(21CS63)

two-byte(16-bit) images, With one byte, it is possible to represent 2 8 , that is 0-255=256 shades
and with 16 bits, it is possible to represent 2 16, that is 65,536 shades. Colour images often use 24
or 32 bits to represent the colour and intensity value.

Sometimes, image processing operations produce images with negative numbers, decimal
fractions, and complex numbers. For example, Fourier transforms produce images involving
complex numbers. To handle, negative numbers, signed and unsigned integer types are used. In
these data types, the first bit is used to encode whether the number is positive or negative.
Floating-point involves storing the data in scientific notation. For example, 1230 can be
represented as 0.1234x104 , where 0.123 is called the significand and the power is called the
exponent. There are many floating-point conventions.

The quality of such data representation is characterized by parameters such as data accuracy
and precision. Data accuracy is the property of how well the pixel values of an image are able
to represent the physical properties of the object that is being imaged. Data accuracy is an
important parameter, as the failure to capture the actual physical properties of the image leads to
the loss of vital information that can affect the quality of the application. While accuracy refers
to the correctness of a measurement, precision refers to the repeatability of the measurement.
Repeated measurements of the physical properties of the object should give the same result.
Most software use the data type ‘double’ to maintain precision as well as accuracy.

Domain Specific Images

Images can be classified based on the domains and applications where such images are
encountered.

Range Images

Range images are often encountered in computer vision. In range images, the pixel values denote
the distance between the object and the camera. These images are also referred to as depth
images. This is in contrast to all other images whose pixel values denote intensity and hence are
often known as intensity images.

Multispectral Images

Page 15
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Multispectral images are encountered mostly in remote sensing applications. These images are
taken at different bands of visible or infrared regions of the electromagnetic wave. Multispectral
images may have many bands that may include infrared and ultraviolet regions of the
electromagnetic spectrum.

Basic Relationship and Distance Matrices

Images can be easily represented as a two-dimensional array of matrix. Pixels can be visualized
logically and physically. Logical pixels specify the points of a continuous 2D function. These are
logical in the sense that they specify a location but occupy no physical area. Normally, this is
represented in the Cartesian first coordinate system. Physical pixels occupy a small amount of
space when displayed on the output device. Digitized images indicating physical pixels are
represented in the Cartesian fourth coordinate system.

For example, an analog image of size 3x3 is represented in the first quadrant of the Cartesian
coordinate system as shown in Fig 4.7.

Figure 4.7: Analog image f(x,y) in the first quadrant of Cartesian coordinate system

Figure 4.7 illustrates an image f(x,y) of dimension 3x3, where f(0,0) is the bottom left corner.
Since it starts from the coordinate position(0,0), it ends with f(2,2), that is x=0,1,2,…..M-1 and
y=0,1,2,…….,N-1. x and y define the dimensions of the image.

In digital image processing, the discrete form of the image is often used. Discrete images are
usually represented in the fourth quadrant of the Cartesian coordinate system. A discrete image
f(x,y) of dimension 3x3 is shown in Fig. 4.8(a)

Page 16
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Many programming environments including MATLAB starts with an index of (1,1). The
equivalent representation of the given matrix is shown in Fig 4.8(b)

Figure 4.8: Discrete image (a) Image in the fourth quadrant of Cartesian coordinate system
(b) Image coordinates as handled by software environments such as MATLAB

The coordinates used for discrete image is, by default, the fourth quadrant of the Cartesian
system.

Image Topology
Image topology is a branch of image processing that deals with the fundamental properties of the
image such as image neighbourhood, paths among pixels, boundary, and connected components.
It characterizes the image with topological properties such as neighbourhood, adjacency and
connectivity. Neighbourhood is fundamental to understanding image topology. Neighbours of a
given reference pixel are those pixels with which the given reference pixel shares its edges and
corners.

In N4(p), the reference pixel p(x,y) at the coordinate position (x,y) has two horizontal and two
vertical pixels as neighbours. This is shown graphically in Fig. 4.9.

0 𝑋 0
[𝑋 𝑝(𝑥, 𝑦)
𝑋]
0 𝑋 0

Figure 4.9: 4-Neighbourhood N4(p)

The set of pixels {(x+1,y),(x-1,y),(x,y+1),(x,y-1)}, called the 4-neighbours of p is denoted as

N4(p) .Thus, 4-neighbourhood includes the four direct neighbours of the pixel p(x,y). The pixel
Page 17
 Computer Graphics and Fundamentals of Image Processing(21CS63)

may have four diagonal neighbours. They are (x-1,y-1), (x+1,y+1),(x-1,y+1) and (x+1,y-1). The
diagonal pixels for the reference pixel p(x,y) are shown graphically in Fig 4.10.

𝑋 0 𝑋
[0 𝑝(𝑥, 𝑦) 0]
𝑋 0 𝑋

Figure 4.10: Diagonal elements ND(p)

The diagonal neighbours of pixel p(x,y) are represented as N D (p). The 4-neighbourhood and ND
are collectively called the 8-neighbourhood. This refers to all the neighbours and pixels that
share a common corner with the reference pixel p(x,y). These pixels are called indirect
neighbours. This is represented as N8(p) and is shown graphically in Fig 4.11.

The set of pixels N8(x)=ND(x)  N4(x)

𝑋 𝑋 𝑋
[𝑋 𝑝(𝑥, 𝑦)
𝑋]
𝑋 𝑋 𝑋

Figure 4.11: 8-Neighbourhood N8(p)

Connectivity
The relationship between two or more pixels is defined by pixel connectivity. Connectivity
information is used to establish the boundaries of objects. The pixels p and q are said to be
connected if certain conditions on pixel brightness specified by the set V and spatial adjacency
are satisfied. For a binary image, this set V will be {0,1} and for grey scale images, V might be
any range of grey levels.

4-Connectivity : The pixels p and q are said to be in 4-connectivity when both have the same
values as specified by the set V and if q is said to be in the set N4(p). This implies any path from
p to q on which every other pixel is 4-connected to the next pixel.

8- Connectivity : It is assumed that the pixels p and q share a common grey scale value. The
pixels p and q are said to be in 8-connectivity if q is in the set N8(p)

Mixed Connectivity: Mixed connectivity is also known as m-connectivity. Two pixels p and q
are said to be in m-connectivity when
Page 18
 Computer Graphics and Fundamentals of Image Processing(21CS63)

1. q is in N4(p)
2. q is in ND(p) and the intersection of N4(p) and N4(q) is empty.

For example, Fig 4.12 shows 8-connectivity when V={0,1}

Figure 4.12 : 8-connectivity represented as lines

8- Connectivity is shown as lines. Here, a multiple path or loop is present. In m-connectivity,

there are no such multiple paths. The m-connectivity for the image in Fig 4.12 is as shown in Fig
4.13.

Figure 4.13 : m-Connectivity

Page 19
 Computer Graphics and Fundamentals of Image Processing(21CS63)

It can be observed that the multiple paths have been removed.

Relations

A binary relation between two pixels a and b, denoted as aRb, specifies a pair of elements of an
image.

For example, consider the image pattern given in Fig 4.14. The set is given as A ={x 1,x2,x3}. The
set based on the 4-connecivity relation is given as A={x 1,x2}. It can be observed that x3 is
ignored as it is not connected to any other element of the image by 4-connectivity.

Figure 4.14 : Image pattern

The following are the properties of the binary relations:

Reflexive: For any element a in the set A, if the relation aRa holds, this is known as a reflexive
relation.

Symmetric: If aRb implies that bRa also exists, this is known as a symmetric relation.

Transitive : If the relation aRb and bRc exist, it implies that the relationship aRc also exists.
This is called the transitivity property.

If all these three properties hold, the relationship is called an equivalence relation.

Distance Measures
The distance between the pixels p and q in an image can be given by distance measures such as
Euclidian distance, D4 distance and D8 distance. Consider three pixels p,q, and z. If the

Page 20
 Computer Graphics and Fundamentals of Image Processing(21CS63)

coordinates of the pixels are P(x,y),Q(s,t) and Z(u,w) as shown in Fig.4.15, the distances
between the pixels can be calculated.

Figure 4.15 : Sample image

The distance function can be called metric if the following properties are satisfied:

1. D(p,q) is well-defined and finite for all p and q.

2. D(p,q) 0 if p=q, then D(p,q)=0
3. The distance D(p,q)=D(q,p)
4. D(p,q)+D(q,z)  D(p,z). This is called the property of triangular inequality.

The Euclidean distance between the pixels p and q, with coordinates (x,y) and (s,t) respectively
can be defined as

The advantage of the Euclidean distance is its simplicity. However, since its calculation involves
a square root operation, it is computationally costly.

The D4 distance or city block distance can be simply calculated as

D4(p,q)=|x-s|+ |y-t|

The D8 distance or chessboard distance can be calculated as

D8(p,q)=max(|x-s|,|y-t|)

Page 21
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Important Image Characteristics

Some important characteristics of images are as follows:

1. The set of pixels that has connectivity in a binary image is said to be characterized by the
connected set.
2. A digital path or curve from pixel p to another pixel q is a set of points p 1,p2,….,pn. If the
coordinates of those points are (x0,y0) ,(x1,y1),…….(xn,yn), then p=(x0,y0) and q=(xn,yn).
The number of pixels is called the length. If x 0=xn and y0=yn , then the path is called a
closed path.
3. R is called a region if it is a connected component.
4. If a path between any two pixels p and q lies within the connected set S, it is called a
connected component of S. If the set has only one connected component, then the set S is
called a connected set. A connected set is called a region.
5. Two Regions R1 and R2 are called adjacent if the union of these sets also forms a
connected component. If the regions are not adjacent, it is called disjoint set. In Fig 4.16,
two regions R1 and R2 are shown. These regions are 8-connected because the pixels
(underlined pixel ‘1’) have 8-connectivity. If the regions are not adjacent, they are called
disjoint.
6. The border of the image is called contour or boundary. A boundary is a set of pixels
covering a region that has one or more neighbours outside the region. Typically, in a
binary image, there is a foreground object and a background object. The border of the
foreground object may have at least one neighbor in the background. If the border pixels
are within the region itself, it is called inner boundary. This need not be closed.
7. Edges are present whenever there is an abrupt intensity change among pixels. Edges are
similar to boundaries, but may or may not be connected. If edges are disjoint, they have
to be linked together by edge linking algorithms. However boundaries are global and
have a closed path. Figure 4.17 illustrates two regions and an edge. It can be observed
that edges provide an outline of the object. The pixels that are covered by the edges lead
to regions.

Page 22
Computer Graphics and Fundamentals of Image Processing(21CS63)

Figure 4.16 : Neighbouring regions

Figure 4.17 : Edge and regions

Page 23
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Classification of Image Processing Operations

There are various ways to classify image operations. The reason for categorizing the operations
is to gain an insight into the nature of the operations, the expected results, and the kind of
computation burden that is associated with them.

One way of categorizing the operations based on neighbourhood is as follows:

1. Point Operations
2. Local Operations
3. Global Operations

Point operations are those whose output value at a specific coordinate is dependent only on the
input value. A local operation is one whose output value at a specific coordinate is dependent on
the input values in the neighbourhood of that pixel. Global operations are those whose output
value at a specific coordinate is dependent on all the values in the input image.

Another way of categorizing operations is as follows:

1. Linear Operations
2. Non-linear Operations

An operator is called a linear operator if it obeys the following rules of additivity and
homogeneity.

1. Property of additivity

H(a1f1(x,y)+a2f2(x,y))= H(a1f1(x,y)) + H(a2f2(x,y))

= a1H(f1(x,y)) + a2H(f2(x,y))

=a1 x g1(x,y)+a2 x g2(x,y)

2. Property of homogeneity

H(kf1(x,y))=kH(f1(x,y))=kg1(x,y)

A non-linear operator does not follow these rules.

Page 24
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Image operations are array operations. These operations are done on a pixel-by-pixel basis.
Array operations are different from matrix operations. For example, consider two images

𝐴 𝐵
F1 = [ ]
𝐶 𝐷
𝐸 𝐹
]
𝐺 𝐻
F2=[

The multiplication of F1 and F2 is element-wise, as follows:

𝐴𝐸 𝐵𝐹
]
𝐶𝐺 𝐻𝐷
F1 X F2 =[

By default, image operations are array operations only.

Arithmetic Operations
Arithmetic operations include image addition, subtraction, multiplication, division and blending.

Image Addition

Two images can be added in a direct manner, as given by

g(x,y)=f1(x,y)+f2(x,y)

The pixels of the input images f1(x,y) and f2(x,y) are added to obtain the resultant image g(x,y).

Figure 4.18 shows the effect of adding a noise pattern to an image. However, during the image
addition process, care should be taken to ensure that the sum does not cross the allowed range.
For example, in a grey scale image, the allowed range is 0-255, using eight bits. If the sum is
above the allowed range, the pixel value is set to the maximum allowed value. Similarly, it is
possible to add a constant value to a single image, as follows:

g(x,y)=f1(x,y)+k

If the value of k is larger than 0, the overall brightness is increased. Figure 4.18(d) illustrates the
addition of the constant 50 increases the brightness of the image.

Page 25
 Computer Graphics and Fundamentals of Image Processing(21CS63)

The brightness of an image is the average pixel intensity of an image. If a positive or negative
constant is added to all the pixels of an image, the average pixel intensity of the image increases
or decreases respectively. The practical application of image addition is as follows:

1. To create double exposure. Double exposure is the technique of superimposing an image

on another image to produce the resultant. This gives a scenario equivalent to exposing a
film to two pictures.
2. To increase the brightness of an image

Figure 4.18: Results of the image addition operations (a) Image1 (b) Image 2 (c) Addition of
images 1 and 2 (d) Addition of image 1 and constant 50

Image Subtraction
The subtraction of two images can be done as follows. Consider

g(x,y)=f1(x,y)-f2(x,y)

where f1(x,y) and f2(x,y) are two input images and g(x,y) is the output image. To avoid negative
values, it is desirable to find the modulus of the difference as

Page 26
 Computer Graphics and Fundamentals of Image Processing(21CS63)

g(x,y)=| f1(x,y)-f2(x,y)|

It is also possible to subtract a constant value k from the image

i.e g(x,y)= | f1(x,y)-k|, as k is constant. The decrease in the average intensity reduces the
brightness of the image. Some of the practical applications of image subtraction are as follows:

1. Background elimination
2. Brightness reduction
3. Change detection

If there is no difference between the frames, the subtraction process yields zero, and if there is
any difference, it indicates the change. Figure 4.19 (a) -4.19(d) show the difference between the
images. In addition, it illustrates that the subtraction of a constant results in a decrease of the
brightness.

Figure 4.19: Results of the image subtraction operation (a) Image 1 (b) Image 2 (c)
Subtraction of images 1 and 2 (d) Subtraction of constant 50 from image 1

Page 27
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Image Multiplication

Image multiplication can be done in the following

manner: Consider

g(x,y)=f1(x,y) x f2(x,y)

f1(x,y) and f2(x,y) are two input images and g(x,y) is the output image. If the multiplied value
crosses the maximum value of the data type of the images, the value of the pixel is reset to the
maximum allowed value. Similarly, scaling by a constant can be performed as

g(x,y)=f(x,y)x k

where k is a constant.

If k is greater than 1, the overall contrast increases. If k is less than 1, the contrast decreases. The
brightness and contrast can be manipulated together as

g(x,y)=af(x,y)+k

Parameters a and k are used to manipulate the brightness and contrast of the input image. g(x,y)
is the output image. Some of the practical applications of image multiplication as follows:

1. It increases contrast. If a fraction less than 1 is multiplied with the image, it results in
decrease of contrast. Figure 4.20 shows that by multiplying a factor of 1.25 with the
original image, the contrast of the image increases.
2. It is useful for designing filter masks.
3. It is useful for creating a mask to highlight the area of interest.

Page 28
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Figure 4.20: Result of multiplication operation (image x 1.25) resulting in good contrast

Image Division
Division can be performed as

g(x,y) = f1(x,y)/f2(x,y)

where f1(x,y) and f2(x,y) are two input images and g(x,y) is the output image.

The division process may result in floating-point numbers. Hence, the float data type should be
used in programming. Improper data type specification of the image may result in loss of
information. Division using a constant can also be performed as

g(x,y) = f(x,y)/k where k is a constant

Some of the practical applications of image division is as follows:

1. Change detection
2. Separation of luminance and reflectance components
3. Contrast reduction

Figure 4.21 (a) shows such an effect when the original image is divided by 1.25.

Page 29
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Figure 4.21 (b)-4.21(e) show the multiplication and division operations used to create a mask. It
can be observed that image 2 is used as a mask. The multiplication of image 1 with image 2
results in highlighting certain portions of image 1 while suppressing the other portions. It can be
observed that division yields back the original image.

Figure 4.21: Image division operation (a) Result of the image operation (image/1.25) (b)
Image 1 (c) Image 2 used as a mask (d) Image 3=image 1 x image 2 (e) Image 4=image
3/image 1

Applications of Arithmetic Operations

Arithmetic operations can be combined and put to effective use. For example, the image
averaging process can be used to remove noise. Noise is a random fluctuation of pixel values,
which affects the quality of the image. A noisy image can be considered as an image plus noise:

g(x,y)=f(x,y)+(x,y)

where f(x,y) is the input image and g(x,y) is the output image. Several instances of noisy
images can be averaged as

Page 30
 Computer Graphics and Fundamentals of Image Processing(21CS63)

where M is the number of noisy images. As M increases the averaging process reduces the
intensity of the noise and it becomes so low that it can automatically be removed. As M becomes
large, the expectation

Logical Operations
Bitwise operations can be applied to image pixels. The resultant pixel is defined by the rules of
the particular operation. Some of the logical operations that are widely used in image processing
are as follows:

1. AND/NAND

2. OR/NOR

3. EXOR/EXNOR

4. Invert/ Logical NOT

1. AND/NAND

The truth table of the AND and NAND operators is given in Table 4.2

Table 4.2: Truth table of the AND and NAND operators

A B C(AND) C(NAND)
0 0 0 1
0 1 0 1
1 0 0 1
1 1 1 0

Page 31
 Computer Graphics and Fundamentals of Image Processing(21CS63)

The operators AND and NAND take two images as input and produce one output image. The
output image pixels are the output of the logical AND/NAND of the individual pixels. Some of
the practical applications of the AND and NAND operators are as follows:

1. Computation of the intersection of images

2. Design of filter masks
3. Slicing of grey scale images; for example, the pixel value of the grey scale image may be
1100 0000. The first bits of the pixels of an image constitute one slice. To extract the first
slice, a mask of value 1000 0000 can be designed. The AND operation of the image pixel
and the mask can extract the first bit and first slice of the image.

Figure 4.22 (a) -4.22(d) shows the effect of the AND and OR logical operators. It illustrates that
the AND operator shows overlapping regions of the two input images and the OR operator
shows all the input images with their overlapping.

Figure 4.22: Results of the AND and OR logical operators (a) Image 1 (b) Image 2 (c)
Result of image 1 OR image 2 (d) Result of image 1 and AND image 2

2. OR/NOR

The truth table of the OR and NOR operators is given in Table 4.3

Table 4.3: Truth table of the OR and NOR operators

A B C(OR) C(NOR)
0 0 0 1
0 1 1 0
1 0 1 0
1 1 1 0

Page 32
 Computer Graphics and Fundamentals of Image Processing(21CS63)

The practical applications of the OR and NOR operators are as follows:

1. OR is used as the union operator of two images.

2. OR can be used as a merging operator

3. XOR/NOR

The truth table of the XOR and XNOR operators is given in Table 4.4.

A B C(XOR) C(XNOR)
0 0 0 1
0 1 1 0
1 0 1 0
1 1 0 1

The practical applications of the XOR and XNOR operations are as follows:

1. Change detection
2. Use as a subcomponent of a complex imaging operation. XOR for identical inputs is zero.
Hence it can be observed that the common region of image 1 and image 2 in Figures
4.22(a) and 4.22 (b), respectively, is zero and hence dark. This is illustrated in Fig. 4.23.

Figure 4.23: Result of the XOR operation

Page 33
 Computer Graphics and Fundamentals of Image Processing(21CS63)

4. Invert/Logical NOT
The truth table of the NOT operator is given in Table 4.5.

Table 4.5: Truth table of the OR and NOR operators

A C(NOT)
0 1
1 0

For grey scale values, the inversion operation is described as

g(x,y)= 255- f(x,y)

The practical applications of the inversion operator are as follows:

1. Obtaining the negative of an image. Figure 4.24 shows the negative of the original image
shown in Fig. 4.22(a)
2. Making features clear to the observer
3. Morphological processing

(a) (b)

Figure 4.24: Effect of the NOT operator (a) Original Image (b) NOT of original Image

Similarly, two images can be compared using operators such as

 = Equal to
 >Greater than
 >= Greater than or equal to
 < Less than

Page 34
 Computer Graphics and Fundamentals of Image Processing(21CS63)

 <= Less than or equal to

  Not equal to

The resultant image pixel represents the truth or falsehood of the comparisons. Similarly shifting
operations are also very useful. Shifting of I bits of the image pixel to the right results in division
by 2I. Similarly, shifting of I bits of the image pixel to the left results in multiplication by 2I.

Shifting operators are helpful in dividing and multiplying an image by a power of two. In
addition, this operation is computationally less expensive.

Geometrical Operations
Translation

Translation is the movement of an image to a new position. Let us assume that the point at the
coordinate position X=(x,y) of the matrix F is moved to the new position X whose coordinate
position is (x’,y’). Mathematically, this can be stated as a translation of a point X to the new
position X. The translation is represented as

x

=x+x

y=y+y

The translation of the floor image by (25,25) is shown in Figure 4.25.

Figure 4.25 : Result of translation by 50 units

Page 35
 Computer Graphics and Fundamentals of Image Processing(21CS63)

In vector notation, this is represented as F=F+T, where x and y are translations parallel to the
x and y axes. F and F are the original and the translated images respectively. However, other
transformations such as scaling and rotation are multiplicative in nature. The transformation
process for rotation is given as F=RF, where R is the transform matrix for performing rotation,
and the transformation process for scaling is given as F=SF. Here, S is the scaling
transformation matrix.

To create uniformity and consistency, it is necessary to use a homogeneous coordinate system

where all transformations are treated as multiplications. A point (x,y) in 2D space is expressed as
(wx,wy,w) for w0. The properties of homogeneous coordinates are as follows:

1. In homogeneous coordinates, at least one point should be non-zero. Thus (0,0,0) does not
exist in the homogeneous coordinate system.
2. If one point is multiplicative of the other point, they are same. Thus, the points(1,3,5) and
(3,9,15) are same as the second point is 3 (1,3,5)
3. The point (x,y,w) in the homogeneous coordinate system corresponds to the point
(x/w,y/w) in 2D space.

In the homogeneous coordinate system, the translation process of point(x,y) of image F to the
new point (x,y) of the image F is described as

x=x+

y=y+

In matrix form, this can be stated as

1 0 𝑥
[x,y,1]= 0 1 𝑥] [x,y,1]T
[
0 0 1

Sometimes, the image may not be present at the origin. In that case, a suitable negative
translation value can be used to bring the image to align with the origin.

Page 36
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Scaling
Depending on the requirement, the object can be scaled. Scaling means enlarging and shrinking.
In the homogeneous coordinate system, the scaling of the point (x,y) of the image F to the new
point (x,y) of the image F is described as

x=x xSX

y=y xSy

𝑆𝑥 0
[x,y]=[ 0 ]
[x,y]

Sx and Sy are called scaling factors along the x and y axes respectively. If the scale factor is 1, the
object would appear larger. If the scaling factors are fractions, the object would shrink. Similarly,
if Sx and SY are equal, scaling is uniform. This is known as isotropic scaling. Otherwise, it is
called differential scaling. In the homogeneous coordinate system, it is represented as

𝑆𝑥 0 0
[x’,y’,1] =[ 0 0] [x,y,1]T

𝑆𝑦
0 0 1

𝑆𝑥 0 0
The matrix S=[ 0 𝑆𝑦 0] is called scaling matrix.
0 0 1

Mirror or Reflection Operation

This function creates the reflection of the object in a plane mirror. The function returns an image
in which the pixels are reversed. This operation is useful in creating an image in the desired order
and for making comparisons. The reflected object is of the same size as the original object, but
the object is in the opposite quadrant. Reflection is also described as rotation by 180 .. The
reflection along the x-axis is given by

1 0
] x [x,y]T
0 −1
F’=[-x,y]=[

Similarly, the reflection along the y-axis is given by

Page 37
 −1 0
Computer Graphics and Fundamentals of Image Processing(21CS63)
] x [x,y]T
0 1
F’=[x,-y]=[

Page 38
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Similarly, the reflection about the line y=x is given as

0 1
] x [x,y]T
1 0
F’=[x,-y]=[

The reflection about the line y=-x is given as

0 −1
] x [x,y]T
−1 0
F’=[x,-y]=[

The reflection operation is illustrated in Fig 3.22(a) and 3.22(b). In the homogeneous coordinate
system, the matrices for reflection can be given as

−1 0 0
Ry-axis=[ 0 1 0]
0 0 1
1 0 0
Rx-axis=[0 −1 0]
0 0 1
−1 0 0
Rorigin=[ 0 −1 0]
0 0 1

The reflection along a line can be given as

0 1 0
Ry=x =[1 0 0]
0 0 1
0 −1 0
Ry=-x =[−1 0 0]
0 0 1

Shearing
Shearing is a transformation that produces a distortion of shape. This can be applied either in the
x-direction or the y-direction. In this transformation, the parallel and opposite layers of the object
are simply sided with respect to each other.

Shearing can be done using the following calculation and can be represented in the matrix form
as

x’=x+ay

y’=y

Page 39
 Computer Graphics and Fundamentals of Image Processing(21CS63)

1 𝑎 0
Xshear=[0 1 0]
0 0 1
(where a=shx)

Yshear can be given as

x’=x

y’=y+bx
1 0 0
Yshear=[𝑏 1 0]
0 0 1
(where b=shy)

where shx and shy are shear factors in the x an y directions, respectively.

Rotation
An image can be rotated by various degrees such as 90 0,1800 or 2700. In the matrix form it is
given as
𝑐𝑜𝑠 −𝑠𝑖𝑛
[x’,y’]= [ ] [x,y]T
𝑠𝑖𝑛 𝑐𝑜𝑠

This can be represented as F=RA. The parameter  is the angle of rotation with respect to the x-
axis. The value of  can be positive or negative. A positive angle represents counter clockwise
rotation and a negative angle represents clockwise rotation. In homogeneous coordinate system,
rotation can be expressed as

𝑐𝑜𝑠 −𝑠𝑖𝑛 0
[x’,y’,1]= 𝑠𝑖𝑛 𝑐𝑜𝑠 0] [x,y,1]
[
0 0 1
If  is substituted with -, this matrix rotates the image in the clockwise direction.

Affine Transform
The transformation that maps the pixel at the coordinates(x,y) to a new coordinate position is
given as a pair of transformation equations. In this transform, straight lines are preserved and
parallel lines remain unchanged. It is described mathematically as

x =Tx(x,y)

Page 40
 Computer Graphics and Fundamentals of Image Processing(21CS63)

y =Ty(x,y)

Tx and Ty are expressed as polynomials. The linear equation gives an affine transform.

x =a0x+a1y+a2

y =b0x+b1y+b2

This is expressed in matrix form as

𝑥′ 𝑎0 𝑎1 𝑎2 𝑥
[𝑦′] = [𝑏0 𝑏1 𝑏2] [𝑦]
1 0 0 1 1

The affine transform is a compact way of representing all transformations. The given equation
represents all transformations.

 Translation is the situation where a0 =1, a1=0, and a2=x.

 Scaling transformation is a situation where a0=sx and b1=sy and a1=0, a2=0, b0=0 and
b2=0.
 Rotation is a situation where a0=cos, a1=-sin, b0=sin, b1=cos, a2=0 and b2=0.
 Horizon shear is performed with a0=1, b0=1,a1=Shx, a2=0, b1=Shy and b2=0

Inverse Transformation

The purpose of inverse transformation is to restore the transformed object to its

original form and position. The inverse or backward transformation matrices are
given as follows:

1 0 −𝑥
Inverse transform for translation= [0 1 −𝑦]
0 0 1

1 0 −𝑥
Inverse transform for scaling = [0 1 −𝑦]
0 0 1

Inverse transform for rotation can be obtained by changing the sign of the transform
term. For example, the following matrix performs inverse transform.
Page 41
 Computer Graphics and Fundamentals of Image Processing(21CS63)

𝑐𝑜𝑠
𝑐𝑜𝑠
+𝑠𝑖𝑛 0
[−𝑠𝑖𝑛 0]
0 0 1

3D Transforms

Some medical images such as computerized tomography(CT) and magnetic

resonance imaging(MRI) images are three-dimensional images. To apply
translation, rotation, and scaling on 3D images, 3D transformations are required.
3D transformations are logical extensions of 2D transformations. These are
summarized and described as follows:

Translation = [ ]

Image Interpolation Techniques

Transforms can be of two types. Affine transforms often produce pixels of the
resultant image that cannot be fit as some of the pixel values are non-integers and
often go beyond the acceptable range. This results in gaps(or holes) and issues
related to number of pixels and range. So interpolation techniques are required to
solve these issues.

Forward Mapping

Forward mapping is the process of applying transformations iteratively to

every pixel in the image, yielding a new coordinate position and copying the
values of the pixel to a new position.

Backward Mapping

Backward mapping is the process of checking the pixels of the output image to
determine the position of the pixels in the input image. This is used to guarantee
that all the pixels of the input image are processed.

During the process of both forward and backward mapping, it may happen that
pixels cannot be fitted into the new coordinates. For example, consider the process
Page 42
 Computer Graphics and Fundamentals of Image Processing(21CS63)
of rotation of a point(10,5) by 450. This yields

x’ =xcos-ysin

=10 cos(450)-5sin(450)

=10(0.707)-5(0.707)

=3.535

y’ = xsin+ycos

Page 43
 Computer Graphics and Fundamentals of Image Processing(21CS63)

= 10 sin(450)+5cos(450)

= 10(0.707)+5(0.707)

=10.605

Since these new coordinate positions are not integers, the rotation process cannot be
carried out. Thus, the process may leave a gap in the new coordinate position, which
creates poor quality output.
Therefore, whenever a geometric transformation is performed, a resampling
process should be carried out so that the desirable quality is achieved in the
resultant image. The resampling process creates new pixels so that the quality of
the output is maintained. In addition, the rounding off of the new coordinate
position (3.535,10.605) should be carried out as (4,11). This process of fitting the
output to the new coordinates is called interpolation.

Interpolation is the method of calculating the expected values for a function with
known pixels. Some of the popular interpolation techniques are:

Nearest neighbor

technique Bilinear

technique

Bicubic technique

The most elementary form of interpolation is nearest neighbor interpolation or

zero-order interpolation. This technique determines the closest pixel and assigns it
to every pixel in the new image matrix, that is, the brightness of the pixels is equal
to the closest neighbor. Sometimes, this may result in pixel blocking and can
degrade the resulting image, which may appear spatially disordered. These
distortions are called aliasing.

A more accurate interpolation scheme is bilinear interpolation. This is called first-

order interpolation. Four neighbours of the transformed original pixels that
surround the new pixel are obtained and are used to calculate the new pixel value.

Page 44
 Computer Graphics and Fundamentals of Image Processing(21CS63)
Linear interpolation is used in both the directions. Weights are assigned based on
the proximity. Then the process takes the weighted average of the brightness of
the four pixels that surround the pixels of interest.

g(x,y)=(1-a)(1-b)f(x’,y’)+(1-a)bf(x’,y’+1)+a(1-b)f(x’+1,y’)+abf(x’+1,y’+1)

Here g(x,y) is the output image and f(x,y) is the image that undergoes the
interpolation operation. If the desired pixel is very close to one of the four nearest
neighbor pixels, its weight will be much higher. This technique leads to blurring
of the edges.However, it reduces aliasing artefacts.

High-order interpolation schemes takes more pixels into account. Second-order

interpolation is known as cubic interpolation. It uses a neighbourhood of 16
pixels. Then it fits two polynomials to the 16 pixels of the transformed original
matrix and the new image pixel. This technique is very effective and produces
images that are very close to the original. In extreme cases, more than 64
neighbouring pixels can be used. However, as the number of pixel increases, the
computational complexity also increases.

Page 45
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Set Operations

An image can be visualized as a set. For example, the following binary

image(Fig3 3.25) can be visualizd as a set A={(0,0),(0,2),(2,2)}. The coordinates
values represent the value of 1. Set operators can then be applied to the set to get
the resultant, which is useful for image analysis.

The complement of set A can be defined as the set of pixels that does not

belong to the set A. Ac={c/cA}

The reflection of the set is

defined as A={c=-a, aA}

The union of two sets, A and B can be

represented as AUB={c/(cA)(cB)}

Where the pixel c belongs to A,B or both.

The intersection of two sets is given as AB={c/(cA)(cB)}. The pixel c

belongs to A,B or both. The difference can be expressed as

A-B={c/(cA)(cB)}
Which is equivalent to ABC

Morphology is a collection of operations based on set theory, to accomplish

various tasks such as extracting boundaries, filling small holes present in the
image, and removing noise present in the image.

Mathematical morphology is a very powerful tool for analyzing the shapes of the
objects hat are present in the images. The theory of mathematical morphology is
based on set theory. One can visualize a binary object as a set. Set theory can then
be applied to the sample set. Morphological operators often take a binary image
and a mask known as structuring element as input. The set operators such as
intersection, union, inclusion and complement can then be applied to images.
Dilation is one of the two basic operators. It can be applied to binary as well as
Page 46
 Computer Graphics and Fundamentals of Image Processing(21CS63)
grey scale images. The basic effect of this operator on a binary image is that it
gradually increases the boundaries of the region, while the small holes that are
present in the images become smaller.

Let us assume that A and B are a set of pixel coordinates. The dilation of A by B

can be denoted as AB={(x,y)+(u,v): (x,y)A,(u,v)B}

Where x and y corresponds to the set A, and u and v corresponds to the set B. The
coordinates are added and the union is carried out to create the resultant set. These
kinds of operations are based on Minkowski algebra.

Page 47
 Computer Graphics and Fundamentals of Image Processing(21CS63)

Statistical Operations

Statistics plays an important role in image processing. An image can be assumed

to be a set of discrete points. Statistical operations can be applied to the image to
get the desired results such as manipulation of brightness and contrast.Some of the
very useful statistical operations include mean,median, mode and mid-range.
These measures are useful in image processing. The measures of data dispersion
also includes quartiles,inter-quartile range and variance.

Some of the frequently used statistical measures are

the following Mean

Mean is the average of all the values in the sample(population) and is denoted as

The overall brightness of the grey scale image is measured using the mean. This is
calculated by summing all the values of the pixels of an image and dividing it by

μ = 𝑛−1 𝐼
the number of pixels in the image.

𝑛 𝑖=0 i
1

Sometimes the data is associated with a weight. This is called weighted mean.
The problem of mean is its extreme sensitivity to noise. Even small changes in
the input affect the mean drastically.

Median

Median is the value where the given X i is divided into two equal halves, with half
of the values being lower than the median and the other half higher. The procedure
for obtaining the median is to sort the values of the given Xi in ascending order. If
the given sequence has an odd number of values, the middle value is the median.
Otherwise, the median is the arithmetic mean of the two middle values.

Mode

Mode is the value that occurs most frequently in the dataset. The procedure for
finding the mode is to calculate the frequencies for all of the values in the data. The
mode is the value(or values) with the highest frequency. Normally, based on the
Page 48
 Computer Graphics and Fundamentals of Image Processing(21CS63)
mode, the dataset is classified as unimodal, bimodal,and trimodal. Any dataset that
has two modes is called bimodal.

Percentile

Percentiles are data that are less than the coordinate by some percentage of the
total value. For example, the median is the 50 th percentile and can be denoted as
Q0.50.The 25th percentile is called the first quartile and the 75th percentile is called
third quartile. Another measure that is useful to measure dispersion is the inter-
quartile range. The inter-quartile is defined as Q0.75 Q0.25.Semi quartile range is
=0.5 Xiqr
Unimodal curves are slightly skewed and the
empirical relation is Mean-Mode=3x(Mean-
Median)

Page 49
 Computer Graphics and Fundamentals of Image Processing(21CS63)

The interpretation of the formula is that the mode for the unimodal frequency
curve is moderately skewed. The mid-range is also used to assess the central
tendency of the dataset. In a normal distribution, the mean, median, and mode
are the same. In symmetrical distributions, it is possible for the mean and median
to be the same even though there may be several modes. By contrast, in
asymmetrical distributions, the mean and median are not the same. These
distributions are said to be skewed data where more than half the cases are either
above or below the mean.

Standard Deviation and Variance

The most commonly used measures of dispersion are variance and standard
deviation. The mean does not convey much more than a middle point. For
example, the following datasets {10,20,30} and
{10,50,0}, both have a mean of 20. The difference between these two sets is the
spread of data. Standard deviation is the average distance from the mean of the
dataset to each point. The formula for standard deviation is given by

Sometimes, we divide the value by N-1 instead of N. The reason is that in a larger,
real-world scenario, division by N-1 gives an answer that is closer to the actual
value. In image processing, it is a measure of how much a pixel varies from the
mean value of the image. The mean value and the standard deviation characterize
the perceived brightness and contrast of the image. Variance is another measure of
the spread of the data. It is the square of standard deviation. While standard
deviation is a more common measure, variance also indicates the spread of the data
effectively.

Entropy

This is the measure of the amount of orderliness that is present in the image.
The entropy can be calculated by assuming that the pixels are totally
uncorrelated. An organized system has low entropy and a complex system has a
very high entropy. Entropy also indicates the average global information
content. Its unit is bits per pixel. It can be computed using the formula

Page 50
 Computer Graphics and Fundamentals of Image Processing(21CS63)

𝑖= 𝑙𝑜𝑔2𝑝𝑖
Entropy H=-
∑𝑛 1

Where pi is the prior probability of the occurrence of the message. Let us

consider a binary image, where the pixel assumes only two possible states,0 or
1 and the occurrence of each state is equally likely. Hence, the probability is
½. Therefore, the entropy H=-[1/2log(1/2)+1/2log(1/2)]=1 bit

Therefore, 1 bit is sufficient to store the intensity of the pixel.Therefore, binary

images are less complex.

Thus, entropy indicates the richness of the image. This can be seen visually using a
surface plot where pixel values are plotted as a function of pixel position.

Convolution and Correlation Operations

The imaging system can be modeled as a 2D linear system.Let f(x,y) and g(x,y)
represent the input and output images, respectively. Then, they can be written as
g(x,y)=t*(f(x,y)).Convolution is a group

Page 51
 Computer Graphics and Fundamentals of Image Processing(21CS63)

process, that is, unlike point operations, group processes operate on a group of
input pixels to yield the result. Spatial convolution is a method of taking a group of
pixels in the input image and computing the resultant output image. This is also
known as a finite impulse response(FIR) filter. Spatial convolution moves across
pixel by pixel and produces the output image. Each pixel of the resultant image is
dependent on a group of pixels(called kernel).

The one-dimensional convolution formula is as

follows: g(x)=t*f(x)
= 𝑛
∑ 𝑖= 𝑡(𝑖)𝑓(𝑥 − 𝑖)
−𝑛

The convolution window is a sliding window that centres on each pixel of the
image to generate the resultant image. The resultant pixel is thus calculated by
multiplying the weight of the convolution mask by pixel values and summing these
values. Thus, the sliding window is moved for every corresponding pixel in the
image in both direction. Therefore, convolution is called ‘shift-add-multiply’
operation.

To carry out the process of convolution, the template or mask is first rotated by
1800. Then the convolution process is carried out. Consider the process of
convolution of two sequences – F, whose dimension is 1 x 5 and a kernel or
template T, whose dimension is 1x3.

Let F={0,0,2,0,0} and the kernel be {7 5 1}. The template has to be rotated by
1800. The rotated mask of this original mask [7 5 1] is a convolution template
whose dimensions is 1 x 3 with value {1,5,7}

To carry out the convolution process first, the process of zero padding should
be carried out. Zero padding is the process of creating more zeros and is done
as shown in Table 3.7.

Convolution is the process of shifting and adding the sum of the product of mask
coefficients and the image to give the centre value.

Correlation is similar to the convolution operation and it is very useful in

Page 52
 Computer Graphics and Fundamentals of Image Processing(21CS63)
recognizing the basic shapes in the image. Correlation reduces to convolution if
the kernels are symmetric. The difference between the correlation and convolution
processes is that the mask or template is applied directly without any prior
rotation., as in the convolution process.

The correlation of these sequences is carried out to observe the difference between
these processes. The correlation process also involves the zero padding process.

Page 53