FUNDAMENTALS OF IMAGING TECHNIQUES, IMAGE REGISTRATION AND DIGITAL

IMAGE PROCESSING
ELEN3008: Biomedical Measurement, Instrumentation and Imaging
Basheq Tarifi (1696842)

School of Electrical & Information Engineering, University of the Witwatersrand, Private Bag 3, 2050, Johannesburg, South Africa

Abstract: Fundamental concepts regarding imaging is presented, including functional and structural biomedical imaging techniques
such as x-ray, computed tomography, magnetic resonance imaging and nuclear medicine. Image registration and its importance
in the field of biomedical imaging is also discussed. Some basic digital image processing is carried out using MATLAB, including
processing of image format, conversion between different representations and manipulation of image histograms.

Key words: ELEN3008, biomedical imaging, image registration, histogram, image processing, MATLAB

1. INTRODUCTION 2.1 X-ray Imaging

It is now the age of healthcare where medical imag-
ing is integral to the diagnosis and treatment of dis-
ease - making it pertinent and essential to the field
of biomedical engineering. The ability to assess both
the physiological and anatomical condition of the
body without directly seeing the internal structure
has made the job of health practitioners much easier,
has advanced medical research and has saved countless
lives[1]. In short, it has dramatically altered medical
practice and healthcare as a whole.
There are many imaging techniques which use differ-
ent methods and mediums, depending on what it is
that the medical practitioner wants to image and what
the suspected diagnosis is. Each of these techniques
have certain uses and limitations, more of which will
be discussed in this paper.
Where in the past, medical images may have been pro-
duced on film, the future of imaging lies in the digital
world[1]. As such, it is important to understand how
digital images work, the composition and structure of
these images, as well as the different techniques of dig-
ital image processing. Since digital images can be rep-
resented by matrices and the processing may involve
matrix manipulation, computers and software such as
MATLAB or Octave lend themselves well to the field
of digital image processing[2].
X-rays are produced by a source and then pass
through a person to be detected by specific
equipment.”X-rays” are actually gamma rays with en-
ergies in the range of 20 − 150keV [4]. The only dif-
ference between x-rays and normal gamma rays is the
origin – gamma rays are usually produced by natural
decay whereas x-rays are produced in an x-ray tube.
Therefore, x-ray imaging is a transmission based imag-
ing technique. X-rays are produced by accelerating a
beam of electrons to hit a metal surface. The x-ray
tube consists of a cathode, which serves as the source
of the electrons, and an anode. There is a large po-
tential difference between 15kVp and 150kVp that is
applied from the anode to the cathode, also known
as the kVp . The cathode is heated and at a suffi-
cient temperature, electrons have enough energy to
break away. These electrons experience an attractive
force towards the anode due to the potential differ-
ence. At the anode, the accelerated electrons pene-
trate the metal surface and lose their kinetic energy,
which is emitted as x-rays. The intensity of the x-ray,
I, is proportional to the potential difference that was
applied across the anode and cathode, i.e. I ∝ (kVp )2 .
This beam of x-rays is then passed through a collima-
tor which shapes it and ensures that it is directed to
the area which needs to be imaged by only allowing
beams with the correct direction to pass through it
[4].

2. IMAGING TECHNIQUES
Imaging techniques can be categorised in multiple
ways – either as structural imaging techniques, which
image the anatomical structure of the body, or as func-
tional imaging techniques, which image the physiolog-
ical function of the body. When dealing with radia-
tion, imaging techniques can also be classified accord-
ing to the source of energy that is detected. Trans-
mission based imaging involves externally produced
energy passing through the body and detected. Emis-
sion based imaging, the energy or radiation that is
detected is produced inside the body [3].
As the x-rays pass through the body, they interact
with tissues and attenuate. The differential attenua-
tions due to unlike tissue is what causes the contrast
that is present on x-ray images [4]. X-rays that pass
through the body can have different levels of interac-
tion. There are some that are transmitted with no
interactions; these are known as primary radiation.
Other x-rays undergo compton scattering and coher-
ent scattering, which are mechanisms of interaction
with tissue – these may change trajectories as they
pass through the body and are called secondary ra-
diation. Those x-rays that are absorbed completely
do not reach the detector constitute absorbed radia-
tion [4]. X-rays are absorbed due to the photoelectric
effect. These phenomena result in an effective atten-
uation that differs between tissues. The intensity of
the x-ray after it passes through the body is given by
I = I0 e−µx (1)
where I, I0 and x represent the intensity of the beam
at position x, the initial intensity and the distance
the beam has travelled, respectively. µ represents the
attenuation coefficient, which is the combined effect of
the aforementioned phenomena and is a constant for a
specific tissue [4]. The x-rays are finally detected by a
detecting film or ionising chamber, and x-ray images
are produced [3].
The high contrast and high resolution images that are
produced by x-rays, as well as the relatively small ra-
diation dose required are some advantages of x-ray
imaging [3]. There is also permanent record of the
image. However, there are also some disadvantages;
x-rays are unable to discern depth and real-time imag-
ing is not possible. These qualities make x-rays ideal
for dental, chest and bone imagery but not suitable for
the imaging of soft tissue and areas where there are
tissues with similar attenuation coefficients [3]. Note
that the spatial resolution of x-rays is proportional to
the width of the x-ray beam, as well as the distance
between the source and the detector. The signal to
noise ratio is proportional to the kVp , the size of the
patient and a number of other factors [4].
2.2 Computed Tomography
Computed Tomography, commonly known as CT
scanning, was invented to address the shortfalls of
conventional x-ray imaging and to allow us to con-
struct 3D images from 2D scans. Computed tomogra-
phy combines x-ray detection with image reconstruc-
tion techniques in order to create planar “slices” of
the body [4]. If multiple scans are done, they can be
layered to generate a 3D reconstruction. Since CT
is an extension of x-ray imaging, it is also a struc-
tural imaging transmission technique. With regards
to the method, a collimated beam of x-rays is directed
through the section (or “slice”) of the body that needs
to be imaged. This beam is very narrow and is sent to
a specific detection site on the detector, which mea-
sures the absorption. The collimator and detector sys-
tem is then incrementally rotated around the body
and at each increment, an x-ray beam is transmitted
and an absorption measurement is attained. This pro-
cess is repeated until the system is rotated 180◦ . Af-
ter all the absorption measurements are collected, the
computer then uses the information to reconstruct the
planar cross-section using methods such as backpro-
jection and iterative techniques [3]. Post-processing is
also performed to colour code and highlight features
of the cross section.
Computed tomography has a vast number of medi-
cal applications. Tumours, cysts, injured tissue and
blood clots are easily viewed using CT. Different soft
tissues are also distinguishable when using CT and
this eliminates the need for many invasive procedures
which are often costly and require hospitalisation. CT
finds uses in radiation therapy as well [3]. There are
not many disadvantages to CT scanning today – in
the past they may have taken long but with modern
computers, the process is quick. The only downside
to computed tomography is that the radiation dose is
much higher than that of a conventional x-ray [4].
2.3 Magnetic Resonance Imaging
Magnetic resonance imaging, or MRI, is a structural
imaging technique which also has functional applica-
tions. It exploits the fact that roughly 60% of a per-
son’s body mass is hydrogen. Since a hydrogen nu-
cleus consists of just one proton, a tiny magnetic field
is produced when the nucleus spins. If hydrogen is
placed in a large static magnetic field it will spin like a
gyroscope at a specific frequency. The magnetic fields
used in MRI are typically more than 10 000 times more
intense than the Earth’s magnetic field [4]. The spin-
ning of the nucleus is also known as precession and
the frequency at which it precesses is called the Lar-
mor frequency, which is unique to every material and
proportional to the magnetic field. Although every hy-
drogen atom precesses at the Larmor frequency, they
are not precessing in phase. For imaging purposes, a
weak radiofrequency (RF) rotating pulse is applied in
a direction perpendicular to the static magnetic field.
This radiofrequency pulse is of the Larmor frequency
and excites the protons, which aligns the phases so
that the majority have coherent precession. This co-
herent precession produces a measurable voltage [4].
When the RF pulse is removed, the hydrogen protons
revert back to their initial equilibrium state. How-
ever, in order to do so, they need to release the energy
which they absorbed. This energy emission is the volt-
age signal that is measured in MRI. The time it takes
to revert back to its original magnetic moment is de-
pendent on two time constants – T1 , the longitudinal
relaxation constant and T2 , the transversal relaxation
constant [3]. These constants are used to discrimi-
nate different types of tissue and in the process of
image construction. In addition to the strong mag-
netic field, weaker graded magnetic fields are applied
such that the magnetic field the body experiences has
a gradient in the three spatial directions. This results
in spatial information being encoded into the signals
that are measured – for example, since the frequency
of precession is dependent on the magnetic field, every
point will have a different Larmor frequency and the
signal measured will be different. In order to create
the image, the signal that is measured is digitised and
an inverse Fourier transform is performed in order to
convert the signal to the spatial domain [4]. Once the
image is formed, computers can be used to introduce
highlights between different types of tissue. It takes
considerable skill to adjust the equipment such that
different tissues are distinguishable [3].
MRI is used extensively in detecting disease and plan-
ning for surgical procedures. It is used to image most
of the body, including the brain, abdomen, heart,
large vessels, breast, bones, soft tissue, joints, carti-
lage, muscle, the head and the neck. It can be used
for both children and adults and can also detect can-
cer pathologies, tumours, and haemorrhages[3]. MRI
has high spatial resolution (approximately 1 mm) and
three-dimensional imaging capabilities, which make it
a favourable imaging technique. However, MRI is
much slower compared to other imaging techniques
and this makes it susceptible to movements of the pa-
tient. The equipment is also extremely expensive and
hence less common when compared to other imaging
systems. The signal to noise ratio is affected by the
strength of the magnetic field as well as the relaxation
constants. The spatial resolution is inversely propor-
tional to the number of data points acquired, as well
the width of the RF pulse applied [4].
Functional MRI (fMRI) is an application of MRI
which allows the detection of brain activity,
which manifest as changes in blood flow, cerebral
metabolism, oxygenation or volume. Since oxy-
genated and deoxygenated blood react differently to
magnetic fields, measuring the change of magnetic sus-
ceptibility of blood is a common means of detection in
fMRI. fMRI has been used to study and map areas of
the brain linked with stimuli and high level cognition,
and can be used to image a brain in real-time [3].
2.4 Nuclear Medicine
Unlike the methods that have been discussed up to
this point, nuclear medicine does not produce a map
of the anatomical structure of the body. Instead, it
images the distribution of radioactive substances that
were purposely introduced into the body. The use of
nuclear medicine lies in the fact that many diseases are
initiated by or affect the physiological and biochemical
processes that occur within the body. These changes
may damage tissue and lead to decreased organ func-
tion with time – examples include oedema (swelling
due to excess fluid), and tumours (both growth and
metastasis). As such, nuclear medicine forms an im-
portant part of clinical diagnosis since it can detect
these early changes in the body before they do dam-
age. It does this by imaging the uptake and biodis-
tribution of radiotracers that are administered to the
patient. These radiotracers are actually compounds
which consist of chemical substrates which are linked
to a small amount of a radioactive element [3]. The
choice of chemical substrate depends on the part of
the body which needs to be imaged – for example, if
it is the kidney then a chemical which passes through
the kidney will be chosen. The radioactive element
used depends on which element bonds well with the
chemical substrate needed. These radiotracers are of-
ten inhaled, injected into the bloodstream or ingested
orally by the patient (once again, this is dependent
on the area of the body in question). An uptake of
chemical substrate above normal levels is often an in-
dicator of disease. For example, cancer cells require
a much higher level of glucose – this shows up in a
nuclear medicine image [4].
Once the radiotracer is introduced into the body, the
radioactive element decays. The decay is usually in
the form of γ (gamma) rays which are emitted in all
directions [4]. Note that these γ rays are no different
in form to the γ rays produced by an X-ray source, and
so they undergo the same attenuation in the body due
to Compton scattering, coherent scattering and the
photoelectric effect. The detection is done by what is
called a “gamma camera”, which is placed around the
area to be imaged. This gamma camera consists of
a collimator, which only allows γ rays close to 90◦ to
pass through it. This allows for the position of the γ
ray to be determined (since the γ rays from other parts
which had angled trajectories will not pass through).
The now parallel γ rays then come into contact with
scintillation crystals, which are special substances that
emit photons when energised. The photons are in
turn converted to an electrical signal by photomul-
tiplier tubes. The image is then formed by analysing
the spatial distributions as well as the magnitudes of
the signals formed by the photomultiplier tubes. The
spatial resolution of nuclear medicine images depends
on the gamma camera as well as the interactions of γ
rays in the body [4]. The geometry of the collimator
and the thickness of scintillation crystals used can de-
crease spatial resolution. The signal to noise ratio is
affected by the amount of of γ rays produced, which in
turn is depends on the amount of radioactive material
introduced into the body. However, there is an upper
limit to this as the radiation dose can not be exceeded
[4].
Nuclear medicine also has the ability to produce three
dimensional images by using the principle of tomogra-
phy [4]. This technique combines nuclear medicine
and certain procedures from CT scanning and is
called Single Photon Emission Computed Tomogra-
phy (SPECT). The basic difference between SPECT
and normal CT is that the γ rays are emitted from
within the body [1]. A rotating gamma camera is
used to obtain data from around the patient and then
image reconstruction methods are used to generate
cross sectional images of the internal distribution of
radiotracers. SPECT provides much higher contrast
and resolution when compared with the images pro-
duced by conventional nuclear medicine [1]. SPECT
and nuclear medicine find major clinical applications
in the measurement of blood perfusion in the brain,
the diagnosis of tumors, and the assessment of cardiac
function [4].
Another application of nuclear medicine is known as
Positron Emission Tomography (PET). PET scan-
ning is similar to SPECT in that tomography is
used to image physiological function, not anatom-
ical structure [4]. The main difference is that in
PET, the radiotracer used emits positrons, not γ rays
(as with SPECT and conventional nuclear medicine).
Positrons are anti-matter versions of electrons, i.e.
they have the structure of electrons but they have a
positive charge instead of negative. When this emitted
positron collides with an electron from the body, there
is an annihilation reaction. Annihilation reactions re-
sult from the collision of matter with anti-matter. 2
γ rays are produced from this reaction, each with an
energy of 511 keV [4]. The reaction can be written as
follows:
e− + e+ → 2γ (2)
The two γ rays travel at 180◦ to each other and both
need to be detected. As such, a PET detection sys-
tem consists of a complete ring of scintillation crystals
around the patient [4]. Since the γ rays will be emitted
at the same time and they travel at the speed of light,
they will be both be detected by the camera within a
very small time window. The location of the two scin-
tillation crystals which detect the γ rays act as the end
points of the line on which the annihilation reaction
occured – this forms the basis for localisation of the
signal (as opposed to using a collimator as in SPECT
and conventional nuclear medicine). This makes PET
more accurate and efficient than SPECT. PET finds
uses mainly in oncology, cardiology and neurology [3].
It produces high contrast and high resolution images
but has disadvantages in the form of cost and the dif-
ficulty in producing positron-emitting radiotracers [4].
3. IMAGE REGISTRATION
3.1 Definition
One of the most important image processing tech-
niques in the context of biomedical engineering is im-
age registration. Image registration is a method of
aligning or fusing two or more images from the same
scene [2]. These images may have been taken at differ-
ent times or in different conditions. Image registration
finds many uses in medical imaging – for example, to
combine images taken by different instruments such as
PET and CT. The images may have also been taken
at different points in time, an example of which is
to monitor the growth of a tumour. A more tech-
nical definition of image registration is that it is the
determination of geometric transformations of images
that would align the point of view of an object with
corresponding points of view of the same object in a
different image, to provide an enhanced visualisation
of the scene [2].
3.2 Uses in Medical Imaging
In many imaging techniques such as CT, PET,
SPECT and fMRI, two dimensional “slices” or cross
sectional images of the body are taken. They are taken
at multiple positions such that many two dimensional
images can be layered to produce a three dimensional
recreation. Image registration is required for this, in
order to map every image together and form a coher-
ent three dimensional image. Since nuclear medicine
(PET or SPECT) just images the distribution of ra-
diotracers within the body, a CT scan is often done be-
forehand or simultaneously and used as a background
to the PET/SPECT image that is produced. This pro-
vides anatomical context for the physiological function
that is imaged [5].
As explained in Section 2.3, fMRI is often used to as-
sess the activity of the brain. For such activities, im-
age registration is critical. During the fMRI process,
the patient is asked to perform an activity over and
over and the brain function is assessed by taking 3D
MR images of the brain. The brain activity manifests
as signals on the fMR image and in order to maximise
this signal and produce accurate results, many images
need to be taken and registered [6].
3.3 Methods of Image Registration
There are many methods of image registration, some
of which are listed here.
3.3.1 Point based registration: This form of regis-
tration relies on pairs of corresponding points that can
be detected either by the algorithm or by a person be-
forehand. The registration algorithm then attempts to
perform a geometric transformation that would map
the corresponding points together [5].
3.3.2 Surface based registration: The surfaces and
boundaries of an anatomical object or structure is a
easily characterized geometrical feature that can be
used for the mapping of images in medical image reg-
istration. Surface-based image registration methods
involve determining corresponding surfaces in differ-
ent images and their surroundings and then comput-
ing the transformation that would align the surfaces
in the best way [5].
3.3.3 Intensity based registration: This method of
registration attempts to fuse images based on the in-
tensity values of pixels alone, without looking at the
image itself. These algorithms measure the similarity
of intensity distributions and compute a transforma-
tion based off this. A major advantage of intensity
based registration is that it requires much less human
interaction beforehand than is required for surface and
point based registration [5].

(a) FluorescentCells.jpg (b) CellDebris.JPG
Figure 1: Sample images used
4. DIGITAL IMAGE PROCESSING
All image processing was carried out on MATLAB
and performed on sample images that were provided
by Professor Aharonson, the course co-ordinator for
ELEN3008 (Biomedical Measurement, Instrumenta-
tion and Imaging) at the University of the Witwa-
tersrand, Johannesburg.
The images in Figure 1 were used for the image pro-
cessing in Section 4.1 and Section 4.2, the code of
which can be found in Appendix A.
4.1 Image Formats
There are many different digital formats and each one
was invented for a particular reason – some formats
take up less storage space than others, whereas others
might be able to display more colours[7]. The follow-
ing image manipulation processes were carried out on
Figure 1a and Figure 1b in MATLAB. The images
were first opened and displayed as figures. They are
JPEG (Joint Photographic Experts Group) images,
which is the most commonly used image format due
to its high compression rates and the ability to dis-
play many colours (approximately 16 million). When
it comes to compression, there are two modes; lossless
compression and lossy compression. Lossless compres-
sion maintains all image information - the original
image can be reconstructed exactly. The algorithm
looks for efficient ways to represent the image while
maintaining accuracy and quality - one way of achiev-
ing this is to search for recurring patterns and replace
every occurrence with an abbreviation, which will in
turn reduce the size of the file[7]. Lossy compression,
on the other hand, will accept compromises in accu-
racy in order to achieve even smaller file sizes. For
example, it might look for similar colours in the image
and store colour information at a lower resolution than
the image itself[7]. JPEG images use lossy compres-
sion and are able to be compressed at varying levels
[2].
In order to investigate the differences between im-
age file formats, the JPEG images in 1a and 1b
were saved as GIF (Graphics Interchange Format) and
BMP (bitmap) images. GIF images use lossless com-
Table 1: Comparison of different image formats
File Size (kB)
Image
JPEG GIF BMP
FlourescentCells 56.1 170 768
Cell Debris 455 1587.2 4700.16
(a) Binary (b) Greyscale
Figure 2: Different representations of CellDebris.jpg
pression but are limited to 8 bit (256) colours. BMP
images are not commonplace - they are actually un-
compressed images (which obviously implies no loss of
information)[7]. A comparison of the file sizes is shown
in Table 1. It can be seen that the BMP images have
the largest file size due to them being uncompressed,
whereas the JPEG files are the smallest due to them
using lossy compression.
There are also other formats such as TIFF and PNG
that may be used instead of JPEG, GIF and BMP[7].
4.2 Image Representation
Digital images have multiple methods of representa-
tion. Binary images, also known as black and white
images, are represented by an m × n array of 0’s and
1’s (0 and 1 represent black and white, respectively)
where m × n is the resolution of the image and each
cell represents one pixel. Binary images are commonly
used as masks in photo editing or when analysis is
done using basic shapes. They are very low on stor-
age and are ideal for representing simple shapes. Opti-
cal recognition and text interpreting applications also
utilise binary images[8]. Greyscale images are repre-
sented by an ×n array of integers which correspond
to different intensities of grey (from black to white).
Greyscale images are used when there is no need for
colour, such as a newspaper article or a manuscript
(colour is much more expensive)[2]. Medical imaging
such as MRI or CT scanning often produces greyscale
images since they provide a useful middle ground be-
tween binary images, which cannot represent complex
image characteristics, and colour images, which may
add an unwanted level of complexity. MATLAB can
be used for interconversion between image types – Bi-
nary and Greyscale versions of CellDebris.JPG (Fig-
ure 1b) can be seen in Figure 2.
Colour images have a number of representations –
RGB, Indexed and HSV images[2]. All three of these
can produce images that are identical. An RGB (Red,
Table 2: Staining of endothelial cell in Fluorescent-
Cells.jpg

Stain Structure Fluorophore Used

Red Actin Filaments TRITC
Green Microtubules FITC
Blue Nuclei DAPI

(a) Red (b) Green (c) Blue

Figure 3: RGB components of FluorescentCells.jpg Figure 4: HSV Colour Cone. Image adapted from
‘Introduction to Color Theory’ [10]

Green, Blue) image is an m × n × 3 matrix of colour

pixels, where each colour pixel consists of elements
from a red, a green and a blue colour channel (hence
a “depth” of 3 in the matrix). The red, green and
blue channels can each be thought of as greyscale im-
ages, which will produce the colour image when fed
to the red, green and blue inputs of a screen and
layered[2]. The number of colours depend on the num-
ber of bits used to represent the values in each element (a) Hue (b) Saturation
of the matrix. If it is 8 bits, then the RGB image will
have (28 )3 colours (approximately 16 million). The
RGB components of an image can be separated using
MATLAB – an example of this is shown in Figure 3,
where the components of FluorescentCells.jpg (Figure
1a) are shown. According to the webpage that the
image was found on, the image is showing endothe-
lial cells – it was taken using an imaging technique (c) Value
called fluorescence microscopy where specific biologi-
cal structures are highlighted using special compounds Figure 5: HSV components of CellDebris.JPG
called fluorophores[9]. The structures which are high-
lighted in the image and the fluorophores used to do
so can be seen in Table 2. This image is a Wikimedia
Commons image and is in the public domain, which HSV (Hue, Saturation, Value) images describe colours
means that anyone may use it without crediting the in a way that is akin to how people perceive colour[10].
creator. The hue of the colour refers to which pure colour it
consists of. It is a number between 0 and 1 and repre-
Indexed images consist of two components – an m × n sents how far around the colour wheel in Figure 4 the
data matrix and a colour map. The colour map is a pure colour is. The saturation refers to how ‘pure’ the
c×3 matrix, which each row representing a colour (the colour is, and is a number from 0 to 1 representing how
three columns correspond to the red, green and blue close to the centre of the wheel in Figure 4 the colour
components). As such, the indexed image will have c is – 0 being the pure hue and 1 being pure white. Fi-
colours. The data matrix contains a integer between nally, the value refers to how dark the colour is, and
0 an (c − 1)) which points to a corresponding row represents a number on the long axis of the cone. A
of the colour map matrix (it acts as an index, hence value of 0 is completely black, with the darkness de-
the name)[2]. Indexed images are useful when there creasing as the value decreases. The value property
is a need for a limited amount of colours since the is also called the lightness[10]. HSV images are rep-
colour map can be tailor-made to include a certain resented digitally by a 3D matrix, with each element
range of colours – MATLAB comes with predefined having a hue, a saturation and a value. The separated
colour maps, for example, ‘autumn’, ‘prism’, ‘jet’ and HSV components of CellDebris.JPG (Figure 1b) can
more[2]. be seen in Figure 5
 (a) Original Image (b) Image after equalisa-
tion

Figure 6: Sample image pollen.jpg

(c) Original histogram

Figure 7: An image histogram for a 4 bit image
(K=16). The image in question has 10 pixels with
an intensity of 2. Adapted from ‘Principles of Digital
Image Processing’ [11].

4.3 Image Histograms

In order to understand image histograms, the image in

Figure 6 was used to perform image processing related
to histograms. All processing was done in MATLAB,
the code of which can be found in Appendix B.

Image histograms are used to graphically show sta-

tistical data of an image so that inferences about the
image can be made easily. An image histogram de-
scribes the frequency distribution of the intensity val-
ues that occur within an image. At each intensity,
the value is defined as the number of pixels in the im-
age that have that intensity. The intensity values are
in the range [0, K − 1]Z, where K is the number of
different intensity values (for a typical 8 bit greyscale
image is K = 28 = 256)[11]. The concept of an im-
age histogram can be understood from Figure 7 which
shows an image histogram for a 4 bit image. Image
histograms allow conclusions about qualities such as
exposure and contrast of the image to be made but
since they contain no information about the spatial
arrangement of pixels, the original image can not be
reconstructed from the histogram itself[11]. In fact,
multiple images may have the same image histogram.
Histogram equalisation is a process whereby an im-
age is modified so that the image histogram has an
approximately equal distribution of intensity values,
(d) Histogram after equalisation
Figure 8: Histogram equalisation – the sample image
pollen.jpg is shown before and after equalisation, with
the corresponding histograms
which leads to a picture with better contrast and ex-
posure. Thus, histogram equalisation improves the
overall quality of an image. However, since we are
dealing with discrete frequencies and intensity values,
it is sometimes not possible to obtain a truly uniform
histogram (especially when there are individual peaks
in the histogram)[11]. Histogram equalisation can be
performed on MATLAB and is demonstrated in Fig-
ure 8 – it is clear that the equalised image has much
better contrast and image quality. The algorithm used
for the equalisation can be found in Appendix B as
well. Histogram equalisation on an RGB image would
be done by first separating it into the three colour
components, then performing histogram equalisation
on each component and finally combining the compo-
nents into one RGB image again. [1]
The matrix of CellDebris.jpg can be plotted as a 3D
image, where the x and y indexes act as the x and [2]
y axes and the intensity values act as the function.
Rotating the 3D graph in MATLAB gives a sense of
the top surface of the structure but not the rest of
it. If the equalised image is plotted as a 3D graph, [3]
it gives a much clearer sense of the structure because
the intensities are more evenly spread. It has better
contrast and the structure itself is moree visible.
[4]
Thresholding an image is an operation that maps all
pixels with intensity values below a certain threshold
value to a fixed intensity value, and all pixels with in- [5]
tensity values above or equal to the threshold value
to a different fixed intensity value[11]. This is demon-
strated in Figure 9. A common application of thresh-
olding is to convert an image to a binary image, which [6]
was done in Section 4.1. When binarizing an 8 bit
greyscale image, the threshold value would be 128 and
the two fixed intensity values would be 0 and 255 [11]. [7]
[8]
[9]
Figure 9: Thresholding of an image. The image his-
tograms are shown with the threshold value and the
two fixed intensity values. Image adapted from ‘Prin-
ciples of Digital Image Processing’ [11]. [10]
5. CONCLUSION
There are many different imaging techniques that are [11]
used today – some of them image the anatomical
structure and others image the physiological function
of the body. These imaging techniques are used in
many fields of medicine and may even be used in con-
juction in order to achieve better results. Image regis-
tration has multiple uses and is often implemented in
imaging or when linking the results of different imag-
ing techniques.
Digital image processing is very important since the
outputs of imaging systems are increasingly digital.
As such, factors such as the image format and the
image representation used are important to know and
understand. MATLAB can be used to process and
change these factors. It can also edit as other qualities
of the image such as the image histogram, performing
histogram equalisation and other functions.
REFERENCES
J. D. Bronzino. The Biomedical Engineering
Handbook: Medical Devices and Systems. USA:
CRC Press, third ed., 2006.
R. Gonzalez, R. E. Woods, and S. L. Eddins. Dig-
ital Image Processing Using MATLAB , pp. 27–
28,305–308,318–335. USA: Gatesmark Publish-
ing, second ed., 2009.
J. D. Enderle and J. D. Bronzino. Introduction
to Biomedical Engineering, pp. 997–998,1019–
1030,1071–1074,1098–1099. USA: Elsevier, third
ed., 2012.
A. Webb. Introduction to Biomedical Imaging,
pp. 1–15,23–25,57–60,73–100. USA: John Wiley
and Sons, first ed., 2003.
C. Maurer, J. Fitzpatrick, and D. L. G. Hill.
Handbook of Medical Imaging, Volume 2. Medi-
cal Image Processing and Analysis, chap. 8. UK,
2000.
P. J. Kostelec and S. Periaswamy. “Image Reg-
istration for MRI.” Modern Signal Processing,
vol. 46.
R. Matthews. “Digital Image File Types
Explained.” URL http://users.wfu.edu/
matthews/misc/graphics/formats/formats.
html. Department of Physics, Wake Forest
University. [Accessed: 2019-02-04].
R. Owens. “Binary Images.” URL http:
//homepages.inf.ed.ac.uk/rbf/CVonline/
LOCAL_COPIES/OWENS/LECT2/node3.html.
University of Edinburgh. [Accessed: 2019-02-04].
Pinterest. “Discover ideas about Fluorescence
Microscopy.” URL https://za.pinterest.
com/pin/29273466297770010/. [Accessed:
2019-02-04].
J. W. Shipman. “Introduction to Color The-
ory.” URL http://infohost.nmt.edu/tcc/
help/pubs/colortheory/web/hsv.html. New
Mexico Tech. [Accessed: 2019-02-04].
W. Burger and M. J. Burge. Principles of Digital
Image Processing, pp. 37–50,57–59,63–66. UK:
Springer, first ed., 2009.
 Appendix

A General Image Processing Code

The following code performs basic image processing such as changing of file formats, interconversion between
different forms of image representation, separation of RGB into colour channels and more.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% BASHEQ TARIFI ( 1 6 9 6 8 4 2 ) %
% ELEN3008 Assignment 1 %
% Q u e s t i o n 1 Code %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%read i n images
i m g F l u o C e l l s = imread ( ’ F l u o r e s c e n t C e l l s . j p g ’ ) ;
i m g C e l l D e b r i s = imread ( ’ C e l l D e b r i s . j p g ’ ) ;

%d i s p l a y images
figure ;
imshow ( i m g F l u o C e l l s )
t i t l e ( ’ Fluore scentC ells . jpg ’ ) ;

figure ;
imshow ( i m g C e l l D e b r i s )
t i t l e ( ’ CellDebris . jpg ’ ) ;

%c o n v e r s i o n o f images t o bmp
i m w r i t e ( i m g F l u o C e l l s , ’ F l u o r e s c e n t C e l l s b m p . bmp ’ ) ;
i m w r i t e ( i m g C e l l D e b r i s , ’ C e l l D e b r i s b m p . bmp ’ ) ;
%c o n v e r s i o n o f images t o g i f
[ indFC , map ] = r g b 2 i n d ( i m g F l u o C e l l s , 2 5 6 ) ;
i m w r i t e ( indFC , map , ’ F l u o r e s c e n t C e l l s g i f . g i f ’ ) ;
[ indCD , map ] = r g b 2 i n d ( i m g C e l l D e b r i s , 2 5 6 ) ;
i m w r i t e ( indCD , map , ’ C e l l D e b r i s g i f . g i f ’ ) ;

%d i s p l a y i n g RGB components o f F l u o r e s c e n t C e l l s . j p g
b l a c k = zeros ( s i z e ( i m g F l u o C e l l s , 1 ) , s i z e ( i m g F l u o C e l l s , 2 ) , ’ u i n t 8 ’ ) ;
i m g F l u o C e l l s r e d = c a t ( 3 , i m g F l u o C e l l s ( : , : , 1 ) , black , b l a c k ) ;
i m g F l u o C e l l s g r e e n = c a t ( 3 , black , i m g F l u o C e l l s ( : , : , 2 ) , b l a c k ) ;
i m g F l u o C e l l s b l u e = c a t ( 3 , bla ck , black , i m g F l u o C e l l s ( : , : , 3 ) ) ;
figure ;
subplot ( 2 , 3 , 1 ) , imshow ( i m g F l u o C e l l s ( : , : , 1 ) )
t i t l e ( ’ Red component ’ ) ;
subplot ( 2 , 3 , 2 ) , imshow ( i m g F l u o C e l l s ( : , : , 2 ) )
t i t l e ( ’ Green component ’ ) ;
subplot ( 2 , 3 , 3 ) , imshow ( i m g F l u o C e l l s ( : , : , 3 ) )
t i t l e ( ’ Blue component ’ ) ;
subplot ( 2 , 3 , 4 ) , imshow ( i m g F l u o C e l l s r e d )
t i t l e ( ’ Red , c o l o r i s e d ’ ) ;
subplot ( 2 , 3 , 5 ) , imshow ( i m g F l u o C e l l s g r e e n )
t i t l e ( ’ Green , c o l o r i s e d ’ ) ;
subplot ( 2 , 3 , 6 ) , imshow ( i m g F l u o C e l l s b l u e )
t i t l e ( ’ Blue , c o l o r i s e d ’ ) ;

%d i s p l a y i n g HSV components o f C e l l D e b r i s . j p g

h s v C e l l D e b r i s = rgb2hsv ( i m g C e l l D e b r i s ) ;
b l a c k h s v = zeros ( s i z e ( h s v C e l l D e b r i s , 1 ) , s i z e ( h s v C e l l D e b r i s , 2 ) , ’ d o u b l e ’ ) ;
figure ;
subplot ( 1 , 3 , 1 ) , imshow ( c a t ( 3 , h s v C e l l D e b r i s ( : , : , 1 ) , b l a c k h s v , b l a c k h s v ) )
t i t l e ( ’ Hue ’ ) ;
subplot ( 1 , 3 , 2 ) , imshow ( c a t ( 3 , b l a c k h s v , h s v C e l l D e b r i s ( : , : , 2 ) , b l a c k h s v ) )
title ( ’ Saturation ’ ) ;
subplot ( 1 , 3 , 3 ) , imshow ( c a t ( 3 , b l a c k h s v , b l a c k h s v , h s v C e l l D e b r i s ( : , : , 3 ) ) )
t i t l e ( ’ I n t e n s i t y Value ’ ) ;

%c o n v e r t t o b i n a r y g r a y s c a l e and d i s p l a y
grayCellDebris = rgb2gray ( imgCellDebris ) ;
figure ;
imshow ( g r a y C e l l D e b r i s )
t i t l e ( ’ Cell Debris − Grayscale ’ ) ;

%c o n v e r t t o b i n a r y and d i s p l a y
bwCellDebris = imbinarize ( grayCellDebris ) ;
figure ;
imshow ( b w C e l l D e b r i s )
t i t l e ( ’ C e l l D e b r i s − Binary ( Black and White ) ’ ) ;

%c o n v e r t t o i n d e x e d and d i s p l a y
[ i n d C e l l D e b r i s , cdmap ] = r g b 2 i n d ( i m g C e l l D e b r i s , 2 5 6 ) ;
figure ;
imshow ( i n d C e l l D e b r i s , cdmap ) ;
t i t l e ( ’ C e l l D e b r i s Indexed ’ ) ;
 B Image Histogram Processing Code

This piece of code diaplays image histograms and performs histogram equalisation without the built in function.
It also displays 3D graphs of the images.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% BASHEQ TARIFI ( 1 6 9 6 8 4 2 ) %
% ELEN3008 Assignment 1 %
% Q u e s t i o n 2 Code %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%read image
i m g P o l l e n = imread ( ’ P o l l e n . j p g ’ ) ;
figure ;

%d i s p l a y h i s t o g r a m o f o r i g i n a l image
imhist ( imgPollen ) ;
xlabel ( ’ Gray L e v e l ’ ) ;
ylabel ( ’ Number o f p i x e l s ’ ) ;
ylim ( [ 0 , 9 ∗ 1 0 ˆ 4 ] ) ;
t i t l e ( ’ Image Histogram o f P o l l e n . j p g ’ ) ;

%HISTOGRAM EQUALISATION [ 1 ]
%g e t w i d t h and h e i g h t
width = s i z e ( imgPollen , 2 ) ;
h e i g h t = s i z e ( imgPollen , 1 ) ;
r e s = width ∗ h e i g h t ; %t o t a l number o f p i x e l s
hist = imhist ( imgPollen ) ;
prob = h i s t / r e s ; %p r o b a b i l i t y d i s t r

c p r o b = prob ;
for x = 2 : s i z e ( c prob , 1 )
c p r o b ( x ) = c p r o b ( x−1) + c p r o b ( x ) ;
end %c a l c u l a t e c u m u l a t i v e p r o b a b i l i t y d i s t r

c p r o b = round ( 2 5 5 ∗ ( c p r o b ) ) ; %c o n v e r t t o 8 b i t

e q i m g P o l l e n = u i n t 8 ( zeros ( h e i g h t , width ) ) ; %e q u a l i s e d image

for m=1: h e i g h t
f o r n=1: width
temp = i m g P o l l e n (m, n ) ;
e q i m g P o l l e n (m, n)= c p r o b ( temp +1); %mapping
end
end

f i g u r e ; %d i s p l a y e q u a l i s e d image
imshow ( e q i m g P o l l e n ) ;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
f i g u r e ; %e q u a l i s e d h i s t o g r a m
imhist ( eq imgPollen ) ;
xlabel ( ’ Gray L e v e l ’ ) ;
ylabel ( ’ Number o f p i x e l s ’ ) ;
ylim ( [ 0 , 9 ∗ 1 0 ˆ 4 ] ) ;
t i t l e ( ’ E q u a l i z e d Image Histogram o f P o l l e n . j p g ’ ) ;

map img = i m g P o l l e n ;
map eq img = e q i m g P o l l e n ;
figure ;
mesh( map img ) ;
figure ;
mesh( map eq img ) ;

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% REFERENCES %
% [ 1 ] W. Burger and M. J . Burge . P r i n c i p l e s o f D i g i t a l %
% Image P r o c e s s i n g . UK: S p r i n g e r , f i r s t ed . , 2 0 0 9 . %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%