0% found this document useful (0 votes)

484 views

Digital Image Processing - Lectures

Digital Images and Image Processing, Sensor Signal Processing, Electromagnetic Radiation, Image Formation, Image Acquisition and Sensing, Use of Matlab, Images and Matlab, Image Display, Spatial Resolution. Point Operations, Thresholding, Spatial Filtering, Gaussian Filters, Noises and Noise Cleaning Methods, Cleaning Impulse Noise, Cleaning Gaussian Noise with Image Averaging and Wiener Filtering, LMS filtering, Edges, Second Differences, Edge Enhancement, Fourier Transforms, DFT, FFT, Convolution, Filtering in the Frequency Domain, High Pass Filtering, Butterworth Filtering, Gaussian Filtering, Removal of Periodic Noise, Inverse Filtering, Color Processing, Color models, Pseudocoloring, Processing of Color Images

Uploaded by

Jorma Kekalainen

Image processing involves changing the nature of an image
in order to either
1) improve its pictorial information for human interpretation,
2) render it more suitable for autonomous machine perception.

These two aspects represent two separate but equally

important aspects of image processing.
A procedure which satisfies the first condition - a
procedure which makes an image look better - may be the
very worst procedure for satisfying the second condition.
Humans like their images to be sharp, clear and detailed;
machines prefer their images to be simple and uncluttered.

Jorma Kekalainen

Digital Image Processing

Improving pictorial information

for human interpretation
Enhancing the edges of an image to make it
appear sharper.
Sharpening edges is a vital component of printing.
In order for an image to appear at its best on the
printed page some sharpening is usually
performed.

Removing noise from an image.

Noise being random errors in the image.

Removing motion blur from an image.

Jorma Kekalainen

Digital Image Processing

Image sharpening
Original image

Jorma Kekalainen

Enhanced original

Digital Image Processing

Processed images

Jorma Kekalainen

Digital Image Processing

Image sharpening

Jorma Kekalainen

Noise reduction with a
X-ray image of circuit
3 x3 median filter.
board corrupted by saltand-pepper noise.

Jorma Kekalainen

Digital Image Processing

Image scrambling: Same information?!

Jorma Kekalainen

Digital Image Processing

Rendering images more suitable for

autonomous machine perception
Examples may include:
Obtaining the edges of an image.
This may be necessary for the measurement of objects in an
image.
Once we have the edges we can measure their spread, and the
area contained within them.
We can also use edge detection algorithms as a first step in edge
enhancement.

Removing details from an image.

For measurement or counting purposes, we may not be
interested in all the detail in an image.
For example, a machine inspected items on an assembly line,
the only matters of interest may be shape, size or color.
For such cases, we might want to simplify the image.
Jorma Kekalainen

Digital Image Processing

Images and edges

Ultrasound
Transmitted and reflected ultrasound
Jorma Kekalainen

Digital Image Processing

Image
Suppose we take an image, e.g. a photo.
Suppose the photo is black and white (that is, lots of
shades of gray), so no color.
We may consider this image as being a two
dimensional function f(x,y), where the function values
give the brightness of the image at any given point
(x,y).
In other words, there is a function f that depends on
the spatial (room) coordinates x and y.
We may assume that in such an image brightness
values can be any real numbers in the range 0 (black)
to 1 (white).
Jorma Kekalainen

Digital Image Processing

Digital image
A digital image differs from a photo in that the x, y and f(x,y)
values are all discrete.
Usually they take on only integer values, so e.g. x and y
ranging from 1 to 256 each, and the) brightness values also
ranging from 0 (black) to 255 (white).
A digital image can be considered as a large array of discrete
dots, each of which has a brightness associated with it.
These dots are picture elements called pixels.
Surroundings of a given pixel form a neighborhood.
A neighborhood can be characterized by its shape in the same
way as a matrix.
E.g. we can speak of a 5*5 neighborhood, or of a 7*9 neighborhood.
Note: Except in very special circumstances, neighborhoods have odd numbers of rows and
Image Processing
22
columns, because this ensures thatDigital
the current
pixel is in the centre of the neighborhood.

Jorma Kekalainen

Example of a digital image

pixel =
picture element

Zoom

Jorma Kekalainen

Original
image: 500x340
Digital Image Processing
pixels

Digital image: Formal definition

A digital image, I, is a mapping from a 2D grid of uniformly
spaced discrete points, {p = (r,c)}, into a set of positive integer
values, {I( p)}, or a set of vector values, e.g., {[R G B]T( p)}.
At each column location in each row of I there is a value.
The pair ( p, I( p) ) is called a pixel (for picture element).
p = (r,c) is the pixel location indexed by row, r, and column, c.
I( p) = I(r,c) is the value of the pixel at location p.
If I( p) is a single number then I is monochrome.
If I( p) is a vector (ordered list of numbers) then I has multiple
bands (e.g., a color image).

Jorma Kekalainen

Digital Image Processing

Pixel: [ p, I(p)]
c
Value: I(p)=I(r,c)

Location: p=(r,c)

Pixel Location: p = (r , c)
Pixel Value: I(p) = I(r , c)
Jorma Kekalainen

Digital Image Processing

Pixel: [ p, I(p)]
c
Value: I(p)=I(r,c)

p=(r,c)
=(row#, col#)
Jorma Kekalainen

Location: p=(r,c)

red

I(p ) = green
blue

Digital Image Processing

Example
Pixels and a neighborhood
Current pixel

3*5 neighborhood
Note: If a neighborhood has an even number of rows or columns (or both), it may
Jorma Kekalainen
Processing
27
be necessary
to specify which pixel inDigital
theImage
neighborhood
is the current pixel.

Digital image processing

We shall be concerned with digital image
processing, which involves using a computer
to change the nature of a digital image.
DFT of the small circle

Jorma Kekalainen

Digital Image Processing

Some applications
Image processing has an enormous range of applications in almost
every area of science and technology e.g.,
Industry
Automatic inspection of items on a production line

Agriculture
Satellite/aerial views of land, for example to determine how much
land is being used for different purposes, or to investigate the
suitability of different regions for different crops, or inspection of fruit
and vegetables - distinguishing good and fresh produce from old.

Medicine
Inspection and interpretation of images obtained from X-rays, MRI
(magnetic resonance imaging) or CT scans.

Law enforcement:
Fingerprint analysis,
Sharpening or de-blurring of speed-camera images,
All kinds of surveillance camera applications
Jorma Kekalainen

Digital Image Processing

Aspects of image processing

It is convenient to subdivide different image processing
algorithms into broad subclasses.
There are different algorithms for different tasks and
problems, and often we would like to distinguish the nature
of the task at hand.
Image enhancement refers to processing an image so that the
result is more suitable for a particular application.
Image restoration may be considered as reversing the damage
done to an image by a known cause.
Image analysis or segmentation involves subdividing an image
into principal parts, or isolating certain aspects of an image.

These classes are not disjoint.

A given algorithm may be used for both image enhancement or
for image restoration.
Jorma Kekalainen

Digital Image Processing

Example: Image enhancement

Image enhancement refers to processing an image so that
the result is more suitable for a particular application:

sharpening or de-blurring an out of focus image,

highlighting edges,
improving image contrast, or brightening an image,
removing noise

Image enhancement techniques bring out the detail in an

image that is obscured or highlight certain features of
interest in an image.
Image enhancement techniques, such as contrast
adjustment and filtering, typically return a modified version
of the original image.
These techniques are frequently used as a preprocessing step to
improve the results of image analysis.
Jorma Kekalainen

Digital Image Processing

Example: Image restoration

Image restoration may be considered as
reversing the damage done to an image by a
known cause:
removing of blur caused by linear camera motion,
removal of optical distortions,
removing periodic interference.

Jorma Kekalainen

Digital Image Processing

Example: Image analysis or

segmentation
Image analysis is the process of extracting meaningful
information from images such as finding shapes, counting
objects, identifying colors, or measuring object properties.
Image segmentation is the process of partitioning an image
into parts or regions
finding lines, circles, or particular shapes in an image,
in an aerial photograph: identifying cars, trees, buildings, or
roads.

This division into parts is often based on the characteristics

of the pixels in the image.
E.g., one way to find regions in an image is to look for abrupt
discontinuities in pixel values, which typically indicate edges.
These edges can define regions.

Another method is to divide the image into regions based

on color values.
Jorma Kekalainen

Digital Image Processing

Types of digital images

We shall consider four basic types of images:
Binary
Grayscale
True color, or RGB
Indexed

Example: Grayscale image

C(150:170,210:235)

Jorma Kekalainen

Digital Image Processing

Color images
Are constructed from
three intensity maps.
Each intensity map is
projected through a color
filter (e.g., red, green, or
blue, or cyan, magenta, or
yellow) to create a
monochrome image.
The intensity maps are
overlaid to create a color
image.
Each pixel in a color image
is a three element vector.
Jorma Kekalainen

Digital Image Processing

Primary and secondary colors of

light (Additive color mixing)
Here, secondary colors are mixtures of two primary colors.
yellow = red + green
cyan = green + blue
magenta = red + blue
CRT
LCD
plasma

Jorma Kekalainen

Blue

Example: Numerical values of

some discrete pixels

Jorma Kekalainen

Digital Image Processing

Indexed image
Most color images only have a small subset of the
more than sixteen million possible colors.
For convenience of storage and file handling, the image
has an associated color map, or color palette, which is
simply a list of all the colors used in that image.
Each pixel has a value which does not give its color (as
for an RGB image), but an index to the color in the
map.
It is convenient if an image has 256 colors or less, for
then the index values will only require one byte each to
store.
Jorma Kekalainen

Digital Image Processing

Example: Indexed image

Indices

Color map

In this image the indices, rather then being the gray values
of the pixels, are simply indices into the color map.
Without the color map, the image would be very dark and
colorless.
Jorma Kekalainen

Digital Image Processing

Example: Indexed image

In this image the indices are simply indices into the color map.
Color map
0.1451
0.7451
0.7412
0.2392
0.8706
0.8706

Jorma Kekalainen

Digital Image Processing

0.1176
0.2627
0.5451
0.2588
0.6941
0.8431

0.1412
0.2314
0.6392
0.4000
0.1490
0.8196

Without the color map,

this image would be
nearly black, because
here the maximum index
48
value is 5.

Example
A small index sample from the previous image
>> ind(117:123,347:352)
ans =
0.1451
2 1 2 1 1 4
0.7451
1 4 4 1 1 1
0.7412
0.2392
2 1 4 4 1 4
0.8706
5 5 2 2 1 1
0.8706
5 5 5 5 5 1
5 5 5 5 5 5
5 5 5 5 5 5
Jorma Kekalainen

Digital Image Processing

Color map
0.1176
0.2627
0.5451
0.2588
0.6941
0.8431

0.1412
0.2314
0.6392
0.4000
0.1490
0.8196

Image file sizes

Image files tend to be large.
For example, suppose we consider a 512*512
binary image.
The number of bits used in this image
(assuming no compression, and neglecting any
header information) is
512*512*1 bit = 262 144 bits =32768 bytes
= 32.768 kB 0.033 MB
Note: Here we use the convention that a kilobyte is one thousand bytes, and a
Jorma Kekalainen
Digital Image Processing
megabyte
is one million bytes.

Image file sizes

A grayscale image of the same size requires:
512*512*1 byte = 262 144 bytes 262.14 kB
0.262MB
In color images each pixel is associated with
3 bytes of color information.
A 512*512 image thus requires
512*512*3 bytes = 786 432 bytes 786.43 kB
0.786 MB
Note: Many images are of course much larger than this. E.g. satellite images may be of
Jorma Kekalainen
Digital Image Processing
51
the order
of ten thousand pixels in each
direction.

Example: A picture is worth one

thousand words
Lets test the well-known proverb A picture is worth
one thousand words.
Assuming a word to contain 10 ASCII characters (on
average), and that each character requires 8 bits of
storage, then 1000 words contain
1000*10*8 = 80 000 bits of information
This is roughly equivalent to the information in a
283*283
binary image
100*100
grayscale image
58*58
RGB color image
Jorma Kekalainen

Digital Image Processing

Image perception
Much of image processing is concerned with
making an image appear better to human eyes.
We should therefore be aware of the limitations
of the human visual system.
Image perception consists of two basic steps:
capturing the image with the eye,
recognizing and interpreting the image with the visual
cortex in the brain.

The combination and immense variability of

these steps influences the ways in we perceive
the world around us.
Jorma Kekalainen

Digital Image Processing

Eye works rather much like a

camera
Reflected light from
the object

Lens

Retina is
the sensor

Focal length

upside down

But the brain performs some more processing that

gives rise to visual phenomena
Jorma Kekalainen

Digital Image Processing

Color vision
The human eye has cones which are sensitive to
different wavelength bands

Jorma Kekalainen

Digital Image Processing

Absorption of light by
the cones in the human eye

Jorma Kekalainen

Digital Image Processing

Example: Visual phenomenon

Observed intensities vary as to the
background.
A single block of gray will appear darker if
placed on a white background than if it were
placed on a black background.
That is, we don't perceive gray scales as they
are, but rather as they differ from their
surroundings.
Jorma Kekalainen

Digital Image Processing

Example: Visual phenomenon

A gray square on different backgrounds

Notice how much darker the square appears when it is

surrounded by a light gray. However, the three central squares
Jormaexactly
Kekalainen
Digital Image Processing
58
have
the same intensity.

Digital image sensor

A continuous image projected
onto the sensor array

Result after sampling

and quantization

For a color
image, there
are 3 different
types of sensor
elements:
red, green and
blue

Sensor element

Jorma Kekalainen

Digital Image Processing

Grayscale color map (256 colors)

Pixel value f(x,y)

Linear transformation

A D/A-converter
converts a digital
value to an
analog value, an
Jorma Kekalainen
electrical voltage

To D/A-converter and
further to the screen
Digital Image Processing

True color map

To D/A-converter and
further to the red
Jorma Kekalainen
channel of the screen.

To D/A-converter and
further to the green
Digital Image Processing
channel
of the screen.

Over 16
million
colors

To D/A-converter and
further to the blue
66
channel of the screen.

Example: Different colormaps

>> x=magic(6)
x=
35 1 6
3 32 7
31 9 2
8 28 33
30 5 34
4 36 29
>> imagesc(x)

Jorma Kekalainen

26
21
22
17
12
13

19
23
27
10
14
18

Digital Image Processing

24
25
20
15
16
11

Jorma Kekalainen

Digital Image Processing

Spectrum
In practice, light consists of
photons with a range of energies, or
waves with a range of frequencies

This mix of frequencies/wavelengths/energies is

called the spectrum of the light.
The spectrum gives the total amount of energy
for each frequency/wavelength/energy.
Monochromatic light consists of only one
frequency/wavelength
Can be produced by special light sources, e.g., lasers
Jorma Kekalainen

Digital Image Processing

Spectrum

Jorma Kekalainen

Digital Image Processing

Color spectrum
In 1666, Newton discovered that sunlight (white light) passing
through a glass prism split up into a color spectrum of wave
lengths in the interval 400-700nm.

Jorma Kekalainen

Digital Image Processing

Color wavelength

Jorma Kekalainen

Digital Image Processing

Classification of EM spectrum

Jorma Kekalainen

Digital Image Processing

Polarization
The electromagnetic field has a direction
Perpendicular to the direction of motion

The polarization of the light is defined as the

direction of the electric field.
Natural light is a mix waves with polarization
in all possible directions: it is unpolarized
light.
Special light sources or filters can produce
polarized light of well-defined polarization.
Jorma Kekalainen

Digital Image Processing

Polarization
Plane polarization
The electric field varies only in a single plane

Digital Image Processing

Radiation power
The power of the radiation, i.e., the energy
per unit time, is the radiant flux
Since the energy depends on the frequency, so
does the radiant flux
Unit: Watt (W=J/s= Joule per second)
Is usually not measured directly.

Note: Radiant flux is one of radiometric quantities. Radiometry is a set of techniques

for measuring electromagnetic radiation, including visible light. Radiometric
Jorma Kekalainen
Image Processing
techniques
in optics characterize theDigital
distribution
of the radiation power in space.85

Radiant flux density

The radiant flux per unit area is the flux density
Since the flux depends on the frequency, so does the flux
density
Unit: Watt per square meter
As the energy propagates
through a area during a
Can be measured directly!
specific time interval

Irradiance is the radiant flux received by a surface per unit

area
Excitance or emittance (old term): flux density emitted from a
surface
Note: Irradiance is often called "intensity" in branches of physics other than
Jorma Kekalainen
Digital Image Processing
86
radiometry, but in radiometry this usage leads to confusion with radiant intensity.

Radiant intensity
For point sources, or distant sources of small
extent, the flux density can also be measured per
unit solid angle

The radiant intensity is the radiant flux per unit

solid angle
Unit: W/sr= Watt per steradian
Jorma Kekalainen

Digital Image Processing

Basic principle
Preservation of energy A constant light
source must produce the same amount of
energy through a solid angle regardless of
distance to the source
The radiant intensity is constant
The radiant flux density decreases with the square
of the distance to the source

Jorma Kekalainen

Digital Image Processing

Radiometric chain

Jorma Kekalainen

Digital Image Processing

Interaction between light and

matter
Most types of light-matter interactions can be
represented by
n = the materials refractive index
= the materials absorption coefficient
Both parameters depend on .
More complex interactions include
polarization effects or non-linear effects.

Absorption spectrum
The spectrum of the reflected/transmitted
light is given by

s1 = incident spectrum
s2 = reflected/transmitted spectrum
a = absorption spectrum ( 0 a() 1)
Jorma Kekalainen

Digital Image Processing

Reflection
Highly dependent on the surface type

A real surface is often a mix between the two

cases
Jorma Kekalainen

Digital Image Processing

104

108

Object plane
The object plane consist of all points that appear
sharp when projected through the lens onto the
image plane.
The object plane is an ideal model of where the
sharp points are located
In practice the object plane may be non-planar: e.g.
described by the surface of a sphere
The shape of the object plane depends on the quality
of the lens (system)

For thin lenses the object plane can often be

approximated as a plane.
Jorma Kekalainen

Digital Image Processing

109

Focal length
The thin lens is characterized by a single parameter:
the focal length fL

To change a (distance to object plane), we need to

change b since fL is constant
a = for b = fL !

Jorma Kekalainen

Digital Image Processing

110

single 3D point

Jorma Kekalainen

Digital Image Processing

115

Airy disk
The smallest resolvable distance in the image
plane, x, is given by

fL/D is the F-number of the lens or lens system

Jorma Kekalainen

Digital Image Processing

116

Point spread function (Airy disk)

Conclusions:
The image cannot have a better resolution than
x!
No need to measure the image with higher
resolution than x!
Image resolution is not defined by number of
pixels in the cameras with high pixel resolution
and high diffraction!

Jorma Kekalainen

Digital Image Processing

117

Point spread function

The point spread function or Airy disk is also
called blur disk or circle of confusion or
sometimes modulation transfer function (MTF)
In general the point spread function can be
related to several effects that make the image of
a point appear blurred
Diffraction
Lens imperfections
Imperfections in the position of the image plane

Often modeled as constant over the image

Can be variable for poor optical systems
Jorma Kekalainen

Digital Image Processing

118

Lens distortion
A lens or a lens system can never map straight
lines in the 3D scene exactly to straight lines in
the image plane
Depending on the lens type, a square pattern
will typically appear like a barrel or a
pincushion

Jorma Kekalainen

Digital Image Processing

119

4
Cos

law

In general, there is an attenuation of the

image towards the edges of the image,
approximately according to cos4
This effect can be compensated in a digital
camera

Note: The flux density decreases with the square of the distance to the light source:
cos2 . The effective area of the detector relative to the aperture varies as cos . The
Jorma Kekalainen
Digital Image Processing
120
effective
area of the aperture relative
to the detector varies as cos

Chromatic aberration
The refraction index of matter (lenses) is
wavelength dependent
E.g., a prism can decompose the light into its
spectrum
A ray of white light is decomposed into rays of different
colors that intersect the image plane at different points

Jorma Kekalainen

Digital Image Processing

121

Image formation

Jorma Kekalainen

Digital Image Processing

122

Image formation

Jorma Kekalainen

Digital Image Processing

123

Image formation

projection
through lens
image of object

2. Ortographic projection
This is unnatural.
Objects appear the same size regardless of their distance to the
capture device.

Both types of projections can be represented via

mathematical formulas.
Ortographic projection is easier and is sometimes used as a
mathematical convenience
Jorma Kekalainen

Digital Image Processing

128

Perspective projection
Perspective projection: 1 = 2, l1 < l2 2 < 1.

Jorma Kekalainen

Digital Image Processing

129

Inside the camera - Sensitivity

Once we have the projection of the reflected light cp(x; y; )
the characteristics of the capture device take over.
V () is the sensitivity function of a capture device.
Each capture device has such a function which determines
how sensitive it is in capturing the range of wavelengths ()
present in cp(x; y; ).
The result is an image function which determines the
amount of reflected light that is captured at the camera
coordinates (x; y)
f(x; y) =cp(x; y; )V()d

Jorma Kekalainen

Digital Image Processing

130

Example: Sensitivity functions

Let us determine the image functions for the below sensitivity
functions V1() , V2() , V3() imaging the same scene:
1. This is the most realistic of the three. Sensitivity is
concentrated in a band around 0
f1(x; y) =cp(x; y; )V1()d
2. This is an unrealistic capture device which has sensitivity
only to a single wavelength 0 as determined by the delta
function. However there are devices that get close to such
selective behavior.
f2(x; y) =cp(x; y; )V2()d = cp(x; y; )(- 0)d
= cp(x; y; 0)
Jorma Kekalainen

Digital Image Processing

131

Example: Sensitivity functions

3. This is what happens if we take a picture without
taking the cap off the lens of our camera.
f3(x; y) =cp(x; y; )V3()d = cp(x; y; )0d=0

Jorma Kekalainen

Digital Image Processing

132

Image capturing

projection onto
discrete sensor
array
Jorma Kekalainen

digital camera
Digital Image Processing

133

Image capturing

sensors register
average color

Jorma Kekalainen

sampled image

Digital Image Processing

134

Image capturing

continuous colors
discrete locations

Jorma Kekalainen

discrete realvalued image

Digital Image Processing

135

Sampling and quantization

144

Losses
When a photon enters the semiconductor
material, it may not interact as intended
1+2: reflection before
entering the active material
3+4: absorption before
entering the active material
5: absorption too deep in the material
6: the photon doesnt interact with the
material and exits at the back
Jorma Kekalainen

voltage
Solar cells
Image sensors in the range near IR - UV
Jorma Kekalainen

Digital Image Processing

149

Thermal excitation
Because of heat in the material electrons are
always excited (moved from the valence band
to the conduction band) due to thermal
energy in the material.
This induces an electric current Ithermo

Jorma Kekalainen

Digital Image Processing

150

Thermal noise
Ithermo is not a constant current, it is rather a
noise current of a mean given by the
expression

We have to treat it as a random signal added

onto the wanted signal Iphoto
Ithermo is a type of noise
Jorma Kekalainen

Digital Image Processing

151

Two main types of photo

detectors
Most image sensors are based on either of
two distinct types of photo detectors
The photodiode (the photovoltaic effect)
The MOS capacitance (intrinsic absorption)

Both can be manufactured using standard

semiconductor processes.
Both can be seen as a capacitor which is
discharged by means of Iphoto and Ithermo .
Jorma Kekalainen

Digital Image Processing

152

Photodiode
Electrons move from the
n+ region to fill holes in
the p-region

Holes from the p-region

move to the n+ region to
be filled with free electrons
The net result is that
A space-charge region
develops, depleted of free
electrons or holes.
The space-charge region is
an electric insulator.
An electric field is
established in the space
charge region (from n+ to p).

Jorma Kekalainen

Digital Image Processing

Due to the electric field:

The electron sweeps to the
n+ region and cancels a hole.
The hole sweeps to the pregion and cancels an
electron.
This builds up a negative
voltage across the junction.
The electric field reduces
linearly to the number of
absorbed photons.
This voltage difference can be
measured.
156

Photodiode
The voltage difference
generated by the photons
occurs even if the diode
had not have been
precharged.
Caused by the photovoltaic
effect.
The diode can in principle
be used as a solar cell.
The pre-charging makes
the photovoltaic effect
stronger since it increases
the space-charge region.

160

MOS capacitor
An absorbed photon
creates an electron/hole
pair
The hole is swept into the
substrate.
The electron is drawn by
the electric field toward the
oxide barrier.
Due to the oxide
insulation, the electrons
accumulate below the
oxide barrier, no current
flows through the capacitor.
The amount of
accumulated electrons is
proportional to the number
of absorbed photons.
Jorma Kekalainen

Digital Image Processing

161

Basic mode of MOS capacitor

operation
1. Apply a voltage across the capacitor
2. Photon absorption creates an electron deposit under
the oxide layer
3. Allow this deposit to accumulate over a certain time
period (the exposure time)
4. The corresponding electron charge is proportional to
the incident flux density of the sensor
5. Move the charge deposit to somewhere where it can
be measured (typically CCD transport)
6. Go to 1.
We need to measure electric charge
Jorma Kekalainen

Digital Image Processing

162

Blooming
Both the photodiode and the MOS capacitor
collect electric charge in a small region
corresponding to the conductor region
When this region becomes saturated, the
charge spills over to neighboring elements
This is called blooming
Barriers between the detectors can reduce
this effect, but not eliminate it entirely
Jorma Kekalainen

Digital Image Processing

163

Fill factor
In practice, the light sensitive area of an image
sensor cannot fill the entire detector area.
Electronic components and wiring reduce the
light sensitive area
The fill factor is the percentage of the total
area which is light sensitive
Light sensitive area
Total pixel area
Jorma Kekalainen

Digital Image Processing

164

Micro-lenses
To overcome low fill factors, an array of microlenses in front of the sensor array can be used

At large incident angles, this spot may

miss the detector area
Jorma Kekalainen

Digital Image Processing

165

Pro and cons of micro-lenses

Micro-lenses enhance the fill factor
But
Due to the manufacturing process, the detector
area can often have an inhomogeneous sensitivity
When light is focused onto a smaller spot in the
sensor, the inhomogeneities become more
noticeable as measurement noise
At large incident angles, this spot may miss the
detector area
Jorma Kekalainen

Digital Image Processing

166

Transport problem
Light has caused a change in electric voltage or charge in a light
detector element (photodiode or MOS capacitor), and this change
needs to be measured to produce an image
Traditionally not measured per detector element
Would require many components per detector
Would give too small fill factor for 2D arrays

The transport problem:

The voltage/charge has to be transported out of the array and
measured outside
Often with a single measurement unit per sensor array

Two principles for solving the transport problem

The CCD array (MOS capacitor only)
Switches to a common signal/video line (photodiode or MOS
capacitor)
Regardless of whether the photo charge has been transported by a
CCD or a signal line, it needs to be converted to a voltage signal
Jorma Kekalainen

Digital Image Processing

Note: The fill factor is the percentage of the total area which is light sensitive.

167

CCD (Charge Coupled Device)

array
A chain of MOS capacitors where the voltages change in the
pattern shown below can move the charge
This transport can take place along an entire row/column of a
detector array
One pixel = 3 capacitors

Jorma Kekalainen

Digital Image Processing

168

Limiting factor of readout

A CCD array can have different ways of implementing
the readout of an entire image
Frame-transfer CCD
Interline-transfer CCD
Field-interline-transfer CCD

Limiting factors:
there is a maximal readout frequency from the entire array
this limits the readout speed from the individual pixel

the MOS-capacitors are sensitive to light exposure during

the transport

Charges should be moved to light insensitive areas as

quickly as possible
Jorma Kekalainen

Digital Image Processing

169

Example: Frame-transfer CCD

Noise sources
The space-charge region is not a perfect
isolator there is a small leakage current
Called dark current since it discharge the
capacitor even when no photons are absorbed

Thermal noise
Can be reduced by cooling

Design noise effect: blooming, after-effects

Note: Both the photodiode and the MOS capacitor collect electric charge in a small
region. When this region becomes saturated, the charge spills over to neighboring
Jorma Kekalainen
175
elements.
This is called blooming Digital Image Processing

Shot noise
Even if a constant number of photons hit the
photo detector, the absorption process is
probabilistic:
Each time we observe/measure the
voltage/charge difference at the detector, there
will a small variation in the result
This variation is the larger the shorter the
exposure time is, and vice versa
This noise has approximately a Poisson
distribution
Jorma Kekalainen

Digital Image Processing

176

SNR and dynamic range

V = The overall noise voltage measured at the output
V = the actual output voltage

V
SNR = 20 log

V
It means that darker images have a lower SNR than brighter
images (assuming constant average noise)
The dynamic range is the SNR of the largest detectable signal
Vmax

Vmax
DR = 20 log

Jorma Kekalainen

Digital Image Processing

Typical values for CMOS

and CCD: DR 40-60 dB
177

Saturation and noise

Saturation is the highest
value beyond which all
intensity levels are clipped

Noise appears often as

a grainy texture

Jorma Kekalainen

Digital Image Processing

178

Digitalization
The analogue voltage signal is normally transformed to a digital
representation by means of an analog-to-digital converter (ADC)
Two common principles:
Flash ADC (up to 8 bits)
Successive approximation (>8 bits)

Quantization noise is independent of method

If b bits are used to represent voltage up to Vmax:

DR
b=
20 log(2 )
gives a quantization noise of the same magnitude as the image
noise
Often, we want a few more bits than this to accurately represent
the image signal
Jorma Kekalainen

Digital Image Processing

179

Flash ADC
Also called the parallel A/D converter

It is formed of a series of
comparators, each one comparing
the input signal to a unique
reference voltage.
The comparator outputs connect to
the inputs of a priority encoder
circuit, which then produces a
binary output.
Vref is a stable reference voltage.
As the analog input voltage exceeds
the reference voltage at each
comparator, the comparator
outputs will sequentially saturate
to a high state.
The priority encoder generates a
binary number based on the
highest-order active input, ignoring
all other active inputs.

Jorma Kekalainen

Digital Image Processing

3-bit flash ADC circuit

180

Successive approximation ADC

The values of the resulting digital output

Instead of counting up in binary
sequence, this successiveare determined successively, from MSB to
approximation register register
LSB
counts by trying all values of bits
starting with the MSB and finishing at
the LSB.
Throughout the count process, the
register monitors the comparators
output to see if the binary count is
less than or greater than the analog
signal input, adjusting the bit values
accordingly.
The advantage to this counting
strategy is much faster results: the
DAC output converges on the analog
signal input in much larger steps than
with the 0-to-full count sequence of a
regular counter.

Jorma Kekalainen

Digital Image Processing

181

Example: Image acquisition using

CCD camera
CCD camera is one of the means for getting a picture into a
computer.
CCD camera has, in place of the usual film, an array of
photosensors, whose output is proportional to the
intensity of light falling on them.
For a camera attached to a computer, information from the
photosensors is then output to a suitable storage medium.
Generally this is done on hardware, as being much faster
and more efficient than software, using a frame-grabbing
card.
This allows a large number of images to be captured in a
very short time.

Jorma Kekalainen

Digital Image Processing

182

Example: Image acquisition using

flat bed scanner
Flat bed scanner works on a principle similar to
the CCD camera.
Instead of the entire image being captured at
once on a large array, a single row of
photosensors is moved across the image,
capturing it row-by-row as it moves.
Since this is a much slower process than taking a
picture with a camera, it is quite reasonable to
allow all capture and storage to be processed by
suitable software.
Jorma Kekalainen

Digital Image Processing

183

Image acquisition
Single imaging sensor

Line imaging sensor

Array imaging sensor

Jorma Kekalainen

Digital Image Processing

184

Image acquisition
Combining a single sensor with motion to
generate a 2-D image.

Jorma Kekalainen

Three identical standard chips

Two semi-transparent mirrors that
refract different wavelengths
Digital Image Processing

188

Three chip color cameras

Based on standard black-and-white sensor
chips (three identical sensor chips)
The three sensor arrays need to be aligned
with tolerances smaller than the inter-pixel
distance
Gives good performance
is expensive
used in professional cameras
Jorma Kekalainen

Digital Image Processing

189

One chip color cameras

To reduce cost:
use one sensor array
place a color filter on top of each detector
element
each detector area is now sensitive to only a
specific wavelength range
reduces the fill factor for each range
the colors are not measured at the same places

Jorma Kekalainen

Digital Image Processing

190

One chip color cameras

The color information may also be converted to some

other color space than RGB (e.g. HIS or XYZ)
Jorma Kekalainen

Digital Image Processing

196

Video camera
Basic idea: take one image after another in sequence
(temporal sampling)
Legacy television standards (PAL, NTCS,) require
interlaced video
Take one half-image with all odd rows and then another
half-image with all even rows, odd, even, etc.
Odd and even rows are exposed at different times
Motivation: better bandwidth usage in broadcasted TV

The effects described here relate to any type of light
measuring digital camera
Modern cameras (e.g., in mobile phones), however,
include increasingly more and more sophisticated
processing of the image and control of the camera

Automatic exposure time control

Automatic focus
Red-eye removal
Color balancing
Motion compensation

Jorma Kekalainen

Digital Image Processing

200

Rapid development
The technology related to image sensors is in
rapid development
The components are constantly becoming smaller
(Moores law)
New solutions to various problems appear at high
pace
More and more functionality is being integrated with
the image sensor
Image sensors are being integrated with other
functionalities (all kinds of supervision, control, and
surveillance anywhere and everywhere)
Jorma Kekalainen

Matlab's standard data type is the matrix -- all
data are considered to be matrices of some sort.
Images, of course, are matrices or stack of
matrices whose elements are the gray values (or
possibly the RGB values) of its pixels.
Single values are considered by Matlab to be
(1*1) matrices, while a string is merely a (1*n)
matrix of characters; n being the string's length.

Jorma Kekalainen

Digital Image Processing

206

Command window
When we start up Matlab, we have a blank
window called the Command window in
which we enter commands.
Command Window is shown in the following
figure.
The prompt consists of two right arrows:
>>

Jorma Kekalainen

Digital Image Processing

207

Command window ready for action

Jorma Kekalainen

Digital Image Processing

219

Example
We can now use these new variables in
further calculations:
>> log(a^2)/log(5)
ans =
7
>> atan(1/b)
ans =
-1.0321
Jorma Kekalainen

Example
Simplest way to achieve
some printing on the screen
is to leave out the semicolon used to end rows.
Function disp works in the
same manner with the
exception that it does not
print the variable names

Jorma Kekalainen

>> A=rand(3);
>> A
A=
0.4218 0.9595
0.9157 0.6557
0.7922 0.0357
>> disp(A)
0.4218 0.9595
0.9157 0.6557
0.7922 0.0357

Digital Image Processing

0.8491
0.9340
0.6787
0.8491
0.9340
0.6787

224

Workspace
The workspace contains variables that you create within or
import into MATLAB from data files or other programs.
Workspace lists all your currently defined variables, their
numeric data types, and their sizes in bytes.
The same information can be obtained using the whos
function:
>>whos
Name Size
a
1x1
ans 1x1
b
1x1
c
1x1
d
1x1
e Kekalainen
1x1
Jorma

Bytes
8
8
8
8
8
8

Class Attributes
double
double
double
double
double
double
Digital Image
Processing

225

who
A listing of the variable names only is obtained
using who:
>> who
Your variables are:
a ans b
c
d
e
The numeric data type double is Matlab's
standard for numbers; such numbers are
stored as doubleprecision 8-byte values.
Note also that ans is variable: it is automatically created by Matlab to store the result
Jorma Kekalainen
Digital Image Processing
226
of the
last calculation.

Complex numbers
Complex numbers have both real and imaginary
parts, where the imaginary unit is the square
root of 1.
sqrt(-1)
To represent the imaginary part of complex
numbers, use either i or j.
c = [3+4i, 4+3j, -i, 10j]

Jorma Kekalainen

Digital Image Processing

227

Example
>> clear
>> sqrt(-1)
ans =
0.0000 + 1.0000i
>> c = [3+4i, 4+3j, -i, 10j]
c=
3.0000 + 4.0000i 4.0000 + 3.0000i 0.0000 - 1.0000i 0.0000 +10.0000i

Jorma Kekalainen

Digital Image Processing

228

Example
>> whos
Name Size
ans
c

Jorma Kekalainen

1x1
1x4

Digital Image Processing

232

Saving and restoring data

Workspace variables do not persist after you exit MATLAB.
Save your data for later use with the save command,
save myfile.mat
Saving preserves the workspace in your current working
folder in a compressed file with a .mat extension, called a
MAT-file.
To clear all the variables from the workspace, use the clear
command.
Restore data from a MAT-file into the workspace using
load.
load myfile.mat

236

Array creation
To create an array with four elements in a
single row, separate the elements with either
a comma (,) or a space.
A = [1 2 3 4] or A = [1,2,3,4]
returns
A=
1234
This type of array is a row vector.
Jorma Kekalainen

Digital Image Processing

237

Creating vectors
1st way:
a = 3 5 7 -2 8 13
a = [3 5 7 -2 8 13]
2nd way: vector = first value : step : last value
a = 0:6
a =0 1 2 3 4 5
b = 1:3:14
b =1 4 7 10 13
c = 5:-1:0
c =5 4 3 2 1 0
3rd way:
vector = linspace(minimum value, maximum value,
number of elements)
a = linspace(1, 3, 6) a =1.0000 1.4000 1.8000 2.2000 2.6000
b=linspace(1, 3, 5)

3.0000

b =1.0000 1.5000 2.0000 2.5000 3.0000

Jorma Kekalainen

Digital Image Processing

238

Array creation
To create a matrix that has multiple rows,
separate the rows with semicolons.
>>A= [1 2 3; 4 5 6; 7 8 10]
A=
1
2
3
4
5
6
7
8
10
Jorma Kekalainen

Digital Image Processing

239

Array creation
Another way to create a matrix is to use a function, such as
ones, zeros, or rand.
For example, create a 5-by-1 column vector of zeros, 1-by-5
row vector of ones and a 5-by-5 identity matrix with ones on
the main diagonal and elsewhere zeros
>>x = zeros(5,1), y=ones(1,5), z=eye(5)
x=
z=
0
1 0 0 0 0
0
0 1 0 0 0
0
0 0 1 0 0
0
0 0 0 1 0
0
0 0 0 0 1
y=
1
Jorma Kekalainen

1
Digital Image Processing

240

Array indexing
Every variable in Matlab is an array that can hold many
numbers.
When you want to access selected elements of an
array, use indexing.
For example, consider the 4-by-4 magic square M:
>>M= magic(4)
M=
16
2
3
13
5
11
10
8
9
7
6
12
4
14
15
1
Jorma Kekalainen

Digital Image Processing

241

Array indexing
There are two ways to refer to a particular
element in an array.
The most common way is to specify row and
column subscripts, such as
>>M(4,2)
ans =
M=
14
16 2 3

13
5 11 10 8
9 7 6 12
4 14 15
1

Jorma Kekalainen

Digital Image Processing

242

Linear indexing
Less common, but sometimes useful, is to use
a single subscript that traverses down each
column in order:
M=
>>M(8)
16 2 3 13
5 11 10 8
ans =
9 7 6 12
4 14 15 1
14
Using a single subscript to refer to a particular
element in an array is called linear indexing.
Jorma Kekalainen

Digital Image Processing

243

Note
If you try to refer to elements outside an array on the right side of an
assignment statement, MATLAB throws an error.
>>test = M(4,5)
Attempted to access M(4,5); index out of bounds because size(M)=[4,4].
However, on the left side of an assignment statement, you can specify
elements outside the current dimensions.
The size of the array increases to accommodate the newcomers
>>M(4,5) = 17
M=
16
2
3
13
0
5
11
10
8
0
9
7
6
12
0
4
14
15
1
17
Jorma Kekalainen

Digital Image Processing

244

Referring to multiple elements

To refer to multiple elements of an array, use the
colon operator, which allows you to specify a
range of the form start:end.
For example, list the elements in the first three
rows in the second column of M:
>>M(1:3,2)
ans =
M=
2
11
16 2 3 13 0
5 11 10 8 0
7
9 7 6 12 0
4 14 15
Jorma Kekalainen

Digital Image Processing

1 17
245

Referring to multiple elements

The colon alone, without start or end values,
specifies all of the elements in that dimension.
For example, select all the columns in the third
row of M:
M=

>>M(3,:)

16 2 3 13 0
5 11 10 8 0
9 7 6 12 0
4 14 15 1 17

Digital Image Processing

249

Matrix elements
Matlab also allows matrix elements to be obtained using a
single number; this number being the position when the
matrix is written out as a single column.
Thus in a (4*4) matrix as above, the order of elements is

So the element a(2,3) can also be obtained as a(10):

>> a(10)
ans =
-3
Jorma Kekalainen

Digital Image Processing

250

Colon operator (:)

256

Example
>> A=[1:3;4:6]
A=
1 2 3
4 5 6
>> B=[1 2;-1 -1;-2 0]
B=
1 2
-1 -1
-2 0
>> C=A*B
C=
-7 0
-13 3

Jorma Kekalainen

Digital Image Processing

257

Example
As an example of matrix powers:
>> a^3*b^4
ans =

Inversion is performed using the inv function:

>> inv(a)
ans =

261

Matrix and array operations

Matlab allows to process all of the values in a matrix using a
single arithmetic operator or function.
>>A + 10
A= [1 2 3; 4 5 6; 7 8 10]
ans =
A=
11
12
13
1
2
14
15
16
4
5
7
8
17
18
20
>>sin(A)
ans =
0.8415
0.9093
0.1411
-0.7568
-0.9589
-0.2794
0.6570
0.9894
-0.5440
Jorma Kekalainen

Digital Image Processing

262

3
6
10

Matrix and array operations

To transpose a matrix, use a single quote ('):
>>A'
ans =
1
4
7
2
5
8
3
6
10
A=

Jorma Kekalainen

Digital Image Processing

1
4
7

2 3
5 6
263
8 10

Matrix and array operations

We can perform standard matrix multiplication,
which computes the inner products between
rows and columns, using the * operator.
For example, confirm that a matrix times its
inverse returns the identity matrix:
>>p = A*inv(A)
p=
1.0000
0
-0.0000
0
1.0000
0
0
0
1.0000
Jorma Kekalainen

Digital Image Processing

264

Note
Notice that p is not a matrix of integer values.
Matlab stores numbers as floating-point values, and arithmetic operations
are sensitive to small differences between the actual value and its
floating-point representation.
We can display more decimal digits using the format command:
>>format long
>>p = a*inv(a)
p=
1.000000000000000
0
-0.000000000000000
0
1.000000000000000
0
0
0
0.999999999999998
Reset the display to the shorter format using
format short
format affects only the display of numbers, not the way Matlab computes
or saves them.
Jorma Kekalainen

Digital Image Processing

265

Note
All these commands work equally well on
vectors.
In fact, Matlab makes no distinction between
matrices and vectors; a vector merely being a
matrix with number of rows or columns equal
to 1.

Jorma Kekalainen

Digital Image Processing

266

Dot operators
A very distinctive class of operators in Matlab are those
which use dots; these operate in an element-wise
fashion element-wise operations.
For example, the command
a*b
performs the usual matrix multiplication of a and b.
But the corresponding dot operator:
a.*b
produces the matrix whose elements are the products
of the corresponding elements of a and b.
Jorma Kekalainen

Digital Image Processing

267

Dot product
That is, if
a.*b
then the element of dot product is c(i,j)=
a(i,j)*b(i,j)
b=[2 4 -7 -4;5 6 3 -2;1 -8 -5 -3;0 -6 7 -1]
b=
>> a.*b
2 4 -7 -4
5 6 3 -2
ans =
1 -8 -5 -3
0 -6

Jorma Kekalainen
Digital Image Processing
Note:
We have also dot division,
and dot powers.

7 -1

268

Example
The command a.^2 produces a matrix each
element of which is a square of the
corresponding elements of a:
>> a.^2
ans =

Jorma Kekalainen

Digital Image Processing

269

Matrix of reciprocals
Similarly, we can produce a matrix of
reciprocals by writing 1./a:
>> 1./a
ans =

Note: The value Inf is Matlab's version of infinity; Here it is returned for
Kekalainen
Digital Image Processing
270
the Jorma
operation
1/0.

274

Example

Jorma Kekalainen

>> zeros(3)
ans =
0 0 0
0 0 0
0 0 0
>> zeros(3,4)
ans =
0 0 0 0
0 0 0 0
0 0 0 0
>> zeros(1,2)
ans =
0Digital Image
0 Processing

275

Digital Image Processing

279

Example
We can first construct two matrices: one containing all the
row indices, and one containing all the column indices:
>> rows=(1:5)'*ones(1,5)
rows =
>> cols=ones(5,1)*(1:5)
11111
cols =
22222
12345
33333
12345
12345
44444
12345
55555
12345

Jorma Kekalainen

Digital Image Processing

280

Example
Now we can construct our matrix using rows and cols:
>> [cols,rows]=meshgrid(1:5,1:5)
>> A=rows+cols-1
cols =
A=
1 2 3 4 5
1 2 3 4 5
12345
1 2 3 4 5
23456
1 2 3 4 5
1 2 3 4 5
34567
rows =
1 1 1 1 1
45678
2 2 2 2 2
56789
3 3 3 3 3
4
5

4
5

Note: The construction of rows and cols can be done automatically with the meshgrid
function:
[cols,rows]=meshgrid(1:5,1:5)
Jorma Kekalainen
Image Processing
281
will produce
the two index matrices Digital
above.

Size of the matrix

The size of our matrix a can be obtained by
using the size function:
size(a)
which returns the number of rows and
columns of a.
>>size(A)
ans =
5
5
Jorma Kekalainen

Digital Image Processing

282

Note
Many functions in Matlab, when applied to a
matrix, work by applying the function to each
element in turn.
Such functions are the trigonometric and
exponential functions, and logarithms.
The use of functions in this way means that in
Matlab many iterations and repetitions can be
done with vectorization rather than by using
loops.
Jorma Kekalainen

Digital Image Processing

283

Vectorization
Vectorization, in Matlab, refers to an operation
carried out over an entire matrix or vector.
In most programming languages, applying an
operation to elements of a list or array will
require the use of a loop, or a sequence of nested
loops.
Vectorization in Matlab allows us to do without
loops in almost all instances, and is a very
efficient replacement for them.
Jorma Kekalainen

Digital Image Processing

284

Example
Suppose we wish to calculate the cosine values of all the
integer radians one to ten million.
We can do this with a for-loop:
for i=1:10^7,cos(i);end
We can measure the time of the operation with Matlab's tic,
toc timer: tic starts a stop watch timer, and toc stops it and
prints out the elapsed time in seconds.
Thus, on this computer:
>> tic,for i=1:10^7,cos(i);end,toc
Elapsed time is 5.836959 seconds.

Jorma Kekalainen

Digital Image Processing

285

Example
We can perform the same calculation with:
>> i=1:10^7;cos(i);
and print out the elapsed time with:
>> tic,i=1:10^7;cos(i);toc
Elapsed time is 1.311342 seconds.
Note that the second command applies the
cosine function to all the elements of the vector
1:10^7, whereas with the for loop, cosine is only
applied to each element of the loop in turn.
Jorma Kekalainen

Digital Image Processing

286

Example
As another example, we can easily generate
the first 10 square numbers with:
>> [1:10].^2
ans =
1 4 9 16 25 36 49 64 81 100
What happens here is that [1:10] generates a
vector consisting of the numbers 1 to 10; and
the dot operator .^2 squares each element in
turn.
Jorma Kekalainen

Digital Image Processing

287

Example
Vectorization can also be used with logical operators;
we can obtain all positive elements of the matrix a
above with:
>> a>0
ans =
1 0 0 1
1 1 0 1
1 0 0 0
0 1 0 1
The result consists of 1's only in the places where the
elements are positive.
Note: Matlab is designed to perform vectorized commands very quickly, and whenever
Jorma Kekalainen
Processing
288
possible
such a command should be Digital
usedImage
instead
of a for loop.

Digital Image Processing

2-D and 3-D Plots in
Matlab

Basic commands

Jorma Kekalainen

Digital Image Processing

290

Title and labels

Figure header can be added by using
command title
Axis explanations can be added by typing
xlabel, ylabel and zlabel
Line comments are added with command
legend
Construction lines are shown by typing grid on
Scale of the axes can be altered using
command axis
Jorma Kekalainen

Digital Image Processing

291

Line plots
The idea is straightforward: we create two
vectors x and y of the same size, then the
command
plot(x,y)
will draw two-dimensional line plot y against
x.
If y has been created from x by using a
vectorized function f(x) the plot will show the
graph of y=f(x)
Jorma Kekalainen

Digital Image Processing

292

Example
x=[0:0.1:2*pi];
plot(x,sin(x))
and the result is shown
beside.

Jorma Kekalainen

A simple plot

299

Example
To add plots to an existing figure, use hold.
x = 0:pi/100:2*pi;
y = sin(x);
plot(x,y)
hold on
y2 = cos(x);
plot(x,y2,'r:')
legend('sin','cos')
Until you use hold off or close the window, all plots appear
in the current figure window.
Jorma Kekalainen

Digital Image Processing

300

Note
It is possible to combine Greek alphabet to figures as well as sub-and
superscripts.

t = 0:0.1:500; y = sin(100.*t*10^-3).*exp(-8.*t*10^-3);
figure,plot(t, y);
title('{\itAe}^{-\alpha\itt}sin\beta{\itt} \alpha<<\beta');
xlabel('Time \musec.');
ylabel('Amplitude');

Jorma Kekalainen

Digital Image Processing

301

Note
Explanation for the title:

Subscripts are added by using underline character

( _ ): For example, A0 is given in form A_0
Jorma Kekalainen

Digital Image Processing

302

Example
Basic command: plot

First parameter: values in the x-axis

Second parameter: values in the y-axis
Additional parameters to customize the line type
and colors.
t = (0:0.1:10)';
y = sin(2*pi*0.2*t);
figure
plot(t,y,'r+')

Jorma Kekalainen

Digital Image Processing

303

Example
t = (0:0.1:10)';
y = sin(2*pi*0.2*t);
figure
plot(t,y,'r+')
hold on
line([0 10],[0 0])
xlabel('Time (s)')
ylabel('\it{sin(2\pift)}')
title('Graph')
grid on
Jorma Kekalainen

Digital Image Processing

304

Subplots
We can display multiple plots in different subregions of the
same window using the subplot function.
For example, create three plots in a 3-by-1 grid within a figure
window. The first two inputs to the subplot function indicate the number
of plots in each row and column. The third input specifies which
plot is active.
t=-3*pi:0.01:3*pi;
yb=cos(t)-(1/3)*cos(3*t)+(1/5)*cos(5*t);
yc=cos(t)-0.5*(1/3)*cos(3*t);
yd=cos(t)-2*(1/3)*cos(3*t);
subplot(311),plot(t,yb),grid
title('yb=cos(t)-(1/3)*cos(3*t)+(1/5)*cos(5*t)')
subplot(312),plot(t,yc),grid
title('yc=cos(t)-0.5*(1/3)*cos(3*t)')
subplot(313),plot(t,yd),grid
Jorma Kekalainen
Digital Image Processing
title('yd=cos(t)-2*(1/3)*cos(3*t)')

305

3-D plots
Three-dimensional plots typically display a
surface defined by a function in two variables,
z = f (x,y).
To evaluate z, first create a set of (x,y) points
over the domain of the function using
meshgrid.
[X,Y] = meshgrid(-2:.2:2);
Z = X .* exp(-X.^2 - Y.^2);
Jorma Kekalainen

Digital Image Processing

306

3-D plots
Then, create a surface plot.
surf(X,Y,Z)

Jorma Kekalainen

Both the surf function and its

companion mesh display surfaces in
three dimensions.
surf displays both the connecting
lines and the faces of the surface
in color.
mesh produces wireframe surfaces
that color only the lines connecting
the defining points.

Digital Image Processing

307

Example
Generate a cylinder defined
by the profile function
2+sin(t).
t = 0:pi/10:2*pi;
[X,Y,Z] = cylinder(2+cos(t));
surf(X,Y,Z)
axis square

Jorma Kekalainen

Digital Image Processing

308

Digital Image Processing

312

Class char
myText and otherText are arrays, like all Matlab variables.
Their class or data type is char, which is short for character.
>> whos myText
Name
Size
Bytes Class Attributes
myText

1x12

24 char

>> whos otherText

Name
Size
Bytes Class Attributes
otherText
Jorma Kekalainen

1x13

26 char
Digital Image Processing

313

Concatenating strings
You can concatenate strings with square brackets, just as you
concatenate numeric arrays.
>> longText = [myText,' - ',otherText]
longText =
Hi, neighbor - You're right!
>> whos
Name
Size
longText
myText
otherText
Jorma Kekalainen

1x28
1x12
1x13

Bytes Class Attributes

56 char
24 char
26 char
Digital Image Processing

314

Converting numeric values to

strings
To convert numeric values to strings, use functions, such as
num2str or int2str.
>> f = 69;
>>c = (f-32)/1.8;
>>tempText = ['Temperature is ',num2str(c), ' degrees
Celsius']
tempText =
Temperature is 20.5556 degrees Celsius

Jorma Kekalainen

Digital Image Processing

315

whos
>> whos
Name
c
f
longText
myText
otherText
tempText
Jorma Kekalainen

Size
1x1
1x1
1x28
1x12
1x13
1x38

Bytes Class
8
8
56
24
26
76

Attributes

double
double
char
char
char
char

Digital Image Processing

316

Interactivity: input
User can be asked for input,
The question to be asked from the user is given as a
parameter to the input-function
The input function returns the answer of the user.
age = input('How old are you?\n');
note the line feed (\n) in the question to obtain question and
answer on different rows
>> age = input('How old are you?\n');
How old are you?
20

Jorma Kekalainen

>> age = input('How old are you?');

How old are you?20

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

326

330

Example

These operators are different:

== is equal to
= is signed a value
Character strings cannot be compared using operator
==.
Use strcmp-function instead
>> 'cats' == 'dogs'
ans =
0 0 0 1
Jorma Kekalainen

>> strcmp('cats', 'dogs')

ans =
0
>> strcmp('cats', 'cats')
ans =
Digital Image Processing
1

331

Logical operators in Matlab

>> A = 1; B = 0;

TRUE = 1; FALSE = 0;

Jorma Kekalainen

>> A && A
ans =
1
>> A && B
ans =
0
>> TRUE & TRUE
ans =
1
>> TRUE & FALSE
ans =
Digital
0 Image Processing

>> A || B
ans =
1
>> B || B
ans =
0
>> TRUE | FALSE
ans =
1
>> TRUE|TRUE
ans =
1

332

for-loop
for variable = initialValue:step:finalValue
% do something
end
for index = -5:2:0
index
end

Jorma Kekalainen

index =
-5
index =
-3
index =
-1

Digital Image Processing

333

Examples
FOR I = 1:N,
FOR J = 1:N,
A(I,J) = 1/(I+J-1);
END
END
FOR S = 1.0: -0.1: 0.0, END
steps S with increments of -0.1
Note: The BREAK statement can be used to terminate the loop prematurely.
Jorma Kekalainen
Image Processing
See also
IF, WHILE, SWITCH, BREAK,Digital
END.

334

Example
Its possible to go through every matrix
element by using two internal for-loops
[M, N] = size(X);
total = 0;
for column = 1:N
for row = 1:M
total = total + X(row, column);
end
end

Jorma Kekalainen

Digital Image Processing

335

while-loop
while condition
% do something
end
element = 1;
while (element <= 10)
x(element) = element;
element = element + 1;
end
>> x
x=
1
Jorma Kekalainen

Digital Image Processing

9 10
336

for-loop vs. while-loop

for-loop is used when it is exactly known a priori how
many repetitions are to be made
e.g., going through every row and column in a matrix
number of repetitions does not necessarily have to be a
constant for instance, the size of the matrix to go through
can change from one execution to another

while-loop is used when the number of repetitions is

unknown in the beginning of the loop
e.g., the user feeds an unknown amount of numbers

Jorma Kekalainen

Digital Image Processing

337

Example
How much is x?
>> a = 0;
while (2)
a = a + 1;
end
>> a
a=
304383259

Process was stopped violently by control C

Jorma Kekalainen

Digital Image Processing

338

break & continue

break-command exits the loop and program
execution continues from the line after the
loop.
continue-command forces the loop to start
immediately from the beginning.

Jorma Kekalainen

Digital Image Processing

339

Example
index = 1;
while (index < 6)
if (index == 4)
break
end
index
index = index + 1;
end
index =
1
index =
2
index =
Jorma Kekalainen
3

index = 1;
while (index < 6)
if (index == 4)
index = index + 1;
continue
end
index
index = index + 1;
end

Digital Image Processing

index =
1
index =
2
index =
3
index =
5

340

Good practise
Program code should be easily read and
understood
easier to debug the errors
easier to remember what the code was intended for
after weeks or months
other people can more easily utilize the code in their
programs

Code is easily read when:

variables have descriptive names
indentation and empty space is used properly
it is well commented
Jorma Kekalainen

Digital Image Processing

341

0.5250 0.5641 0.4332 0.4950

Iteration #5

sampleMean =

0.5250 0.5641 0.4332 0.4950 0.4250

overallMean =
0.4884

Jorma Kekalainen

Digital Image Processing

346

Example: Running the script

When you run the previous calcmean1 script, it displays the
intermediate results, and then calculates the overall mean.
calcmean
Iteration #1
sampleMean =
0.3988
Iteration #2
sampleMean =
0.3988
0.4950
Iteration #3
sampleMean =
0.3988
0.4950
0.5365
Jorma Kekalainen

Digital Image Processing

347

Example: Running the script

Iteration #4
sampleMean =
0.3988 0.4950 0.5365 0.4870
Iteration #5
sampleMean =
0.3988 0.4950 0.5365 0.4870 0.5501
overallMean =
0.4935

Jorma Kekalainen

Digital Image Processing

348

Editor
Type edit or edit file_name to open Matlab
editor.
It is possible to run command string files from
the command line by typing the files name.
Command string files can also be run by
pressing the run button while the editor
window is open.

Jorma Kekalainen

Digital Image Processing

349

Iteration #4
sampleMean =
0.4886 0.4174 0.4447 0.4977 0.4250
Iteration #5
sampleMean =
0.4886 0.4174 0.4447 0.4977 0.5112
overallMean =
0.4719
Mean is less than expected
Jorma Kekalainen

Digital Image Processing

353

Example
sqrt(2)

exp

exponential

exp(1)

Log
Log2

(natural) logarithm (base e) log( exp(2) )

log2(16)
logarithm base 2

log10

logarithm base10

Trigonometric
MATLAB
function
sin

functions
Description

log10(1e4)

sine

Example
sin(pi/6)

cos

cosine

cos(pi/3)

tan

tangent

tan(pi/4)

asin

inverse sine (arcsine)

asin(1/2)

acos

inverse cosine (arccos)

acos(1/2)

atan

inverse
tangent (arctan)
Digital Image Processing

atan(1)

Jorma Kekalainen

357

Note
The arguments of sin, cos, tan are always in
radians (not degrees).
The results of the inverse trigonometric
functions will also be in radians.

Jorma Kekalainen

Digital Image Processing

358

To call a function
Suppose that our workspace includes
variables A and B, such as
A = [1 3 5];
B = [10 6 4];
To call a function, enclose its input arguments
in parentheses:
max(A);

Jorma Kekalainen

Digital Image Processing

359

To call a function
If there are multiple input arguments, separate them with commas:
max(A,B);
Return output from a function by assigning it to a variable:
maxA = max(A);
When there are multiple output arguments, enclose them in square
brackets:
[maxA,location] = max(A);
Enclose any character string inputs in single quotes:
disp('hello world');
To call a function that does not require any inputs and does not
return any outputs, type only the function name:
clc
The clc function clears the Command Window.
Jorma Kekalainen

Digital Image Processing

360

Example
>> A=magic(5)
A=
17 24 1 8 15
23 5 7 14 16
4 6 13 20 22
10 12 19 21 3
11 18 25 2 9
>> [maxA,location] = max(A)
maxA =
23 24 25 21 22
location =
2 1 5 4 3
Jorma Kekalainen

Digital Image Processing

361

User defined functions

Functions are used to make the code more easily readable
and to implement new features in Matlab
Differences between functions and command string files:
function can be given parameters which affect the execution of the
function
function can have one or more return values or the result of the
function.
every function has its own workspace created when the function is
called and destroyed when the execution of the function ends
functions cannot directly utilize variables declared in Matlabs
workspace or internal variables in other functions
internal variables of the function are lost if they are not explicitly
returned or saved
Jorma Kekalainen

Digital Image Processing

362

Syntax and description of

square bracket are not required.
The Matlab keyword function must be present.
Jorma Kekalainen

Digital Image Processing

366

Input and output arguments

When calling a function, the order of the input and output
arguments determines which variables correspond.
For example if a function is defined by
function [Ret, Vol, Div] = process(Prices, Dates)

and called by
[x, y, z] = process(A, B)

then the variable A in the calling program gives the value of

the variable Prices in the function, and the variable B will
give the values for the variable Dates within the function.
The variables may be scalars, vectors, matrices, strings or
more complicated structures.
Similarly, the value of Vol calculated within the function will
give the value of the variable y in the calling program.
Jorma Kekalainen

Digital Image Processing

367

Write a function
Open a file in a text editor.
Within the file, declare the function and add program
statements:
function f = fact(n)
f = prod(1:n);

Function fact accepts a single input argument n, and

returns the factorial of n in output argument f.
The definition statement is the first executable line of
any function.
Each function definition includes these elements.
Jorma
Kekalainen definitions are not valid
Digitalat
Image
Processing
Note:
Function
the
command line or within a script.

368

Write a function
If our function returns more than one output, enclose the output
names in square brackets, such as
function [one,two,three] = myfunction(x)

The body of a function can include valid Matlab expressions, control

flow statements, comments, blank lines, and nested functions.
Any variables that we create within a function are stored within a
workspace specific to that function, which is separate from the base
workspace.
Functions end with either an end statement, the end of the file, or
the definition line for another function, whichever comes first.
The end statement is required only when a function in the file
contains a nested function (a function completely contained within
its parent).

371

Parameters
Function is usually given one or more parameters
which it needs to execute its task
The following function avg_example(x) counts
the average value of the numbers given in
parameter vector
function [xmean]=avg_example(x)
% Getting the size of the vector
N = length(x);
% Counting the mean
xmean = 1 / N * sum(x);
Jorma
Kekalainen name = name of the
Digitalm-file.
Image Processing
Note:
Function

372

Example

Jorma Kekalainen

Digital Image Processing

373

Function calling from command

line
When the function is saved as avg_example.m, it can
be called from the command line in the usual manner.
>> numbers=1:10;
>> avg_example(numbers)
ans =
5.5000

>> numbers=(1:100);
>> avg_example(numbers)
ans =
50.5000

Functions can be given multiple parameters by listing

them all in parentheses after the function name
function weightedmean= weightedavg(x, w)

Jorma Kekalainen

Digital Image Processing

374

doc command or press the question mark in the upper part of

the Matlabs main window or press F1
Jorma Kekalainen

Digital Image Processing

379

Help

Jorma Kekalainen

Digital Image Processing

380

Now we know that the function is implemented using exp.

>> lookfor exp

Note: Matlab convention is to use uppercase names for functions in help texts, even
Jorma Kekalainen
Image Processing
384
though
the function itself is called inDigital
lowercase.

Helping documentation
All MATLAB functions have supporting
documentation that includes examples and
describes the function inputs, outputs, and
calling syntax.
There are several ways to access this
information from the command line:
Open the function documentation in a separate
window using the doc command.

doc mean
Jorma Kekalainen

Digital Image Processing

385

Helping documentation
Display function hints (the syntax portion of the
function documentation) in the Command Window by
pausing after you type the open parentheses for the
function input arguments.

mean(
View an abbreviated text version of the function
documentation in the Command Window using the
help command.

help mean
Access the complete product documentation by
clicking the help icon .
Jorma Kekalainen

Digital Image Processing

386

Exercise
Test and study all the previously presented
Matlab commands.

Jorma Kekalainen

Digital Image Processing

387

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

Images in Matlab

Matlab
Matlab is a data analysis and visualization tool which
has been designed with powerful support for matrices
and matrix operations.
Matlab has excellent graphics capabilities, and its own
powerful programming language.
One of the reasons that Matlab has become such an
important tool is through the use of sets of Matlab
programs designed to support a particular task.
These sets of programs are called toolboxes, and the
particular toolbox of interest to us is the image
processing toolbox.
Jorma Kekalainen

Digital Image Processing

390

Images in Matlab
The basic data structure in Matlab is the array, an
ordered set of real or complex elements.
This object is naturally suited to the
representation of images, real-valued ordered
sets of color or intensity data.
Matlab stores most images as two-dimensional
arrays (i.e., matrices), in which each element of
the matrix corresponds to a single pixel in the
displayed image.
Pixel or picture element usually denotes a single
dot on a computer display.
Jorma Kekalainen

Digital Image Processing

391

Example
For example, an image composed of 200 rows and 300
columns of different colored dots would be stored in
Matlab as a 200-by-300 matrix.
Some images, such as true color images, require a threedimensional array, where the first plane in the third
dimension represents the red pixel intensities, the second
plane represents the green pixel intensities, and the third
plane represents the blue pixel intensities.
This convention makes working with images in Matlab
similar to working with any other type of matrix data, and
makes the full power of Matlab available for image
processing applications.

396

Example
Create the image of a ramp (256*256):

Jorma Kekalainen

Digital Image Processing

397

Example
>> A=ones(256,1)*(0:255);
D=uint8(A);
figure,imshow(A)
figure,imshow(D)
whos
Name
Size
Bytes Class

A
D

Jorma Kekalainen

256x256
256x256

Attributes

524288 double
65536 uint8

Digital Image Processing

398

Matlab function and command

A Matlab function is a keyword which accepts
various parameters, and produces some sort of
output, e.g., a matrix, a string, a graph or figure.
A command is a particular use of a function.
Examples of commands might be
>> sin(pi/3)
>> [c,emap]=imread('kids.tif');
>> figure,imshow(c)
>> figure,imshow(c,emap)
Jorma Kekalainen

Digital Image Processing

399

Matrix
Matlab's standard data type is the matrix.
All data are considered to be matrices of some
sort.
Images, of course, are matrices whose elements
are the gray values (or possibly the RGB values) of
its pixels.
Single values are considered by Matlab to be
(1*1) matrices, while a string is merely a (1*n)
matrix of characters; n being the string's length.
Jorma Kekalainen

Digital Image Processing

400

Grayscale images
Matrices can be handled very efficiently in
Matlab.
Images may be considered as matrices whose
elements are the pixel values of the image.
Suppose you will have a Matlab command
window open, and in it the Matlab prompt >>
ready to receive commands.
Type in the command
>> c=imread('coins.png');
Jorma Kekalainen

Digital Image Processing

Note: PNG Portable Network Graphics is a graphics file format.

401

c=imread('coins.png');
This takes the gray values of all the pixels in
the grayscale image coins.png and puts them
all into a matrix c.
This matrix c is now a Matlab variable, and so
we can perform various matrix operations on
it.
In general the imread function reads the pixel
values from an image file, and returns a matrix
of all the pixel values.
Jorma Kekalainen

Digital Image Processing

402

c=imread('coins.png');
A few notes about this command:
It ends in a semicolon; this has the effect of not
displaying the results of the command to the screen.
As the result of this particular command is a matrix of
size 246*300 = 73800 elements, so we don't want all
its values displayed.
The name coins.png is given in single quote marks.
Without them, Matlab would assume that coins.png
was the name of a variable, rather than the name of a
file.
Jorma Kekalainen

Digital Image Processing

403

figure,imshow(c)
Now we can display this matrix as a grayscale image typing in
the command
>> figure, imshow(c)
This is really two commands on the one line.
Matlab allows many commands to be entered on the same
line; using commas to separate the different commands.
The two commands we are using here are:
figure, which creates a figure on the screen.
A figure is a window in which a graphics object can be placed.
Objects may include images, or various types of graphs.
imshow(c), which displays the matrix c as an image.
This is a display of the gray values of the pixels in the image.
Jorma Kekalainen

Digital Image Processing

404

figure,imshow ('coins.png')
If there are no figures open,
then an imshow command,
or any other command
which generates a graphics
object, will open a new
figure for displaying the
object.
However, it is good practice
to use the figure command
whenever you wish to
create a new figure.
We could display this image
directly, without saving its
gray values to a matrix, with
the command
>>imshow('coins.png')
Jorma Kekalainen

Note: Since coins.png is an 8-bit grayscale

image, the pixel values appear as integers in the
Digital Image Processing
405
range 0-255.

RGB (truecolor) images

The next figure shows an RGB

image of class double

0.5176 0.1608 0.0627

Jorma Kekalainen

Digital Image Processing

411

True color or RGB images

Matlab handles 24-bit RGB images in much the same way as
grayscale.
We can save the color values to a matrix and view the result:
>> a=imread('autumn.tif');
>> figure,imshow(a)

>> a(100,200,:)

ans =
25

ans(:,:,1) =
75

>> a(100,200,1:3)
ans(:,:,1) =
75

ans(:,:,2) =
25

ans(:,:,2) =
25
ans(:,:,3) =
ans(:,:,3) =
30

grayscale image
F=imread('flowers.tif');
figure, imshow(F)
title('F=imread(''flowers.tif'')')
R=F(:,:,1); G=F(:,:,2); B=F(:,:,3); % Component images
GrayF=.2989*R+0.5870*G+0.1140*B; % or GrayF=rgb2gray(F);
figure, imshow(GrayF)
title('GrayF=.2989*R+0.5870*G+0.1140*B')

Jorma Kekalainen

Digital Image Processing

419

Component images with grayscale

colormaps and grayscale image
R

Gray
B

Jorma Kekalainen

Digital Image Processing

420

Colormaps
Sometimes, the pixels are quantified to the interval [0,255].
These values are transformed through a colormap in the
computer to
Grayscale values, i.e. 0 black and 255 white or
Arbitrary colors (pseudocolors)

Sometimes, the pixels are floating point values.

These values are transformed to the interval [0,255] and
further through a colormap in the computer.
A true color image has 3 values per pixel.
These values are transformed individually through 3 color
maps in the computer and further to the red, green and
blue channel of the screen.
Jorma Kekalainen

Digital Image Processing

421

True color map

To D/A-converter and
further to the red
Jorma Kekalainen
channel of the screen.

To D/A-converter and
further
to the green
Digital Image Processing
channel of the screen.

Over 16
million
colors

To D/A-converter and
further to the blue422
channel of the screen.

Jorma Kekalainen

Digital Image Processing

426

Crop the image

>> F=imread('flowers.tif');
>> class(F)
ans =
Uint8
>> size(F)
ans =
362 500 3
Here
Here it
is: it is:
>> figure
>> image(F)
>> title('flowers.tif')
>> xlabel('Introduction to Image Processing')
>> truesize
>> R=F(40:190,330:480,:);
>> figure
>> image(R)
>> truesize
>>
Jorma Kekalainen

Digital Image Processing

Now close the

other image

427

Example

Jorma Kekalainen

Digital Image Processing

428

Indexed images
An indexed image consists of a data matrix, X, and a colormap matrix,
map.
map is an m-by-3 array of class double containing floating-point values in
the range [0, 1].
Each row of map specifies the red, green, and blue components of a single
color.
An indexed image uses "direct mapping" of pixel values to colormap
values.
The color of each image pixel is determined by using the corresponding
value of X as an index into map.
Values of X therefore must be integers.
The value 1 points to the first row in map, the value 2 points to the second
row, and so on.
Display an indexed image with the statements
image(X); colormap(map)
Note: A colormap is often stored with an indexed image and is automatically loaded with
the image when you use the imread function. However, you are not limited to using the
Jorma Kekalainen
429
default
colormapuse any colormap Digital
thatImage
youProcessing
choose.

Structure of an indexed image

The pixels in the image are represented by integers, which are pointers
(indices) to color values stored in the colormap.
The relationship between the values in the image matrix and the
colormap depends on the class of the image matrix. If the image
matrix is of class double, the value 1 points to the first row in the
colormap, the value 2 points to the second row, and so on. If the
image matrix is of class uint8 or uint16, there is an offsetthe
value 0 points to the first row in the colormap, the value 1 points
to the second row, and so on.
The offset is also used in
graphics file formats to
maximize the number of
colors that can be supported.
In the following image, the
image matrix is of class
double. Because there is no
offset, the value 5 points to
Jorma Kekalainen
the fifth
row of the colormap.

Digital Image Processing

430

Intensity images
An intensity image is a data matrix, I, whose values
represent intensities within some range.
An intensity image is represented as a single matrix,
with each element of the matrix corresponding to one
image pixel.
The matrix can be of class double, uint8, or uint16.
While intensity images are rarely saved with a
colormap, a colormap is still used to display them.
In essence, intensity images are treated as indexed
images.
Jorma Kekalainen

Digital Image Processing

431

Intensity images
To display an intensity image, we can
use the imagesc ("image scale")
function, which enables us to set the
range of intensity values.
imagesc scales the image data to use
the full colormap. Use the two-input
form of imagesc to display an intensity
image.
E.g.:
imagesc(I,clims); colormap(gray);
The second input argument
to imagesc specifies the desired
intensity range. The imagesc function
displays I by mapping the first value in
the range to the first colormap entry,
and the second value to the last
colormap entry. Values in between are
linearly distributed throughout the
remaining colormap colors
Jorma Kekalainen

10 and 60 scaled to the full range of
the gray colormap using the
statements
>> subplot(1,2,2)
clims = [10 60];
imagesc(X,clims)
colormap(gray)
axis image
title('CLim = [10 60]')

Digital Image Processing

435

Expanding midrange of grayscale

resolution

>> load clown

figure
imagesc(X)
colormap(map)

Jorma Kekalainen

Digital Image Processing

439

Example

Jorma Kekalainen

Digital Image Processing

440

imfinfo('clown.bmp')
imfinfo('clown.bmp')
ans =

Filename: 'C:\Users\ktkekjo\Documents\MATLAB\clown.bmp'

FileModDate: '17-Aug-2015 20:31:42'

FileSize: 65078

Format: 'bmp'

FormatVersion: 'Version 3 (Microsoft Windows 3.x)'

Width: 320

Height: 200

BitDepth: 8

ColorType: 'indexed'

FormatSignature: 'BM'

NumColormapEntries: 256

Colormap: [256x3 double]

Jorma Kekalainen

Digital Image Processing

441

Indexed color images

The command
>> imshow('ancienttrees.tif')
produces a nice color image of trees.
If we try saving it to a matrix first and then
displaying the result:
>> AT=imread('ancienttrees.tif');
>> figure,imshow(AT),
we obtain a dark, barely distinguishable image,
with single integer gray values, indicating that AT
is being interpreted as a single grayscale image.
Jorma Kekalainen

Digital Image Processing

442

Example
>> imshow('ancienttrees.tif')

Jorma Kekalainen

>> AT=imread('ancienttrees.tif');
>> figure,imshow(AT)

Digital Image Processing

443

446

Imfinfo

>> imwrite(X,map,'ancienttrees.tif')
>> imshow('ancienttrees.tif')
>> imfinfo('ancienttrees.tif')

ans =

Filename:
'C:\Users\ktkekjo\Documents\MATLAB\ancienttrees
.tif'
FileModDate: '23-kes-2015 16:40:36'
FileSize: 75764
Format: 'tif'
FormatVersion: []
Width: 350
Height: 258
BitDepth: 8
ColorType: 'indexed'
FormatSignature: [73 73 42 0]
ByteOrder: 'little-endian'
NewSubFileType: 0
BitsPerSample: 8
Compression: 'PackBits'

Jorma Kekalainen

PhotometricInterpretation: 'RGB Palette'

StripOffsets: [1x12 double]
SamplesPerPixel: 1
RowsPerStrip: 23
StripByteCounts: [1x12 double]
XResolution: 72
YResolution: 72
ResolutionUnit: 'Inch'
Colormap: [256x3 double]
PlanarConfiguration: 'Chunky'
TileWidth: []
TileLength: []
TileOffsets: []
TileByteCounts: []
Orientation: 1
FillOrder: 1
GrayResponseUnit: 0.0100
MaxSampleValue: 255
MinSampleValue: 0
Thresholding: 1
Offset: 73930

Digital Image Processing

447

Note
AT=imread('ancienttrees.tif');
imshow(AT)

Jorma Kekalainen

[AT,cmap]=imread('ancienttrees.tif');
figure,image(AT)
colormap(cmap)

Digital Image Processing

448

Colormap and index

In fact the image ancienttrees.tif is an example of
an indexed image, consisting of two matrices: a
color map, and an index to the color map.
Assigning the image to a single matrix picks up
only the index; we need to obtain the color map
as well:
>> [AT,map]=imread('ancienttrees.tif');
>> figure,imshow(AT,map)
Matlab stores the RGB values of an indexed
image as values of type double, with values
between 0 and 1.
Jorma Kekalainen

Digital Image Processing

256

ind(100,100,:)+1=201

emap(198:204,:)

Colormap

Number+1 at
pixel location is
an index into a
colormap.

green

blue

Digital Image Processing

Intensity values
are between 0
and 1.

*255 = [ 239 95 149 ]

452

Example

Jorma Kekalainen

24-bit truecolor

Digital Image Processing

8-bit colormapped

453

Colormapped image, indices, and colormap

>> [AT,map] =imread('ancienttrees.tif');
Indices contained in
AT(98:102,98:102)
95 105 95 93 105
112 105 105 105 105
117 117 112 105 105
112 114 112 112 105
97 111 117 112 112

actual values in
map(112:116,:)

Index+1

0.8392
0.6471
0.8078
0.7098
0.9686

0.7098
0.8078
0.8078
0.8392
0.7098

0.8706
0.9373
0.6118
0.8706
0.8706

255*map(112:116,:)
214
165
206
181
247
Jorma Kekalainen

Digital Image Processing

181
206
206
214
181

222
239
156
222
222
454

Example: Converting a colormapped image

to true color
AT is a mxnx1, 8-bit
image.
Each pixel has a value
between 0 and 255.

map is the colormap that is stored in

ancienttrees.tif along with image. map is a 256x3
type-double matrix, each row of which lists a color
in terms of its R, G, and B intensities, which are
given as fractions between 0 and 1.

>> [AT,map] = imread('ancienttrees.tif');

>> ATtrue = uint8(reshape(map(AT+1,:),[size(AT) 3])*255);

The m x n x 3 matrix of
intensity values is
reshaped into a m x n x
3 image of type double.
The values are scaled
to lie between 0 and
255 then converted to
Jorma Kekalainen
type uint8.

By concatenating ATs columns, Matlab rearranges

AT into a m x n x 1 list. Each number in the list
(if it has 1 added to it) refers to a row of the
colormap. Then, map(AT+1,:) produces a m x n x 3
matrix of intensity values of type double between
0 and 1.
Digital Image Processing

455

Example: Converting a colormapped image

to true color
>> [AT,map] = imread('ancienttrees.tif');
ATtrue = uint8(reshape(map(AT+1,:),[size(AT) 3])*255);
class(ATtrue)
size(ATtrue)
figure,imshow(ATtrue)
title('Converted to the true color image')
ans =
uint8
ans =
258 350 3
255*map(112:116,:)
214
165
206
181
247

181
206
206
214
181

222
239
156
222
222

Jorma Kekalainen

Digital Image Processing

456

How to make colormaps

Jorma Kekalainen

Digital Image Processing

457

Color bands of a truecolor image

displayed with grayscale colormaps

Digital Image Processing

459

Color bands of a truecolor image

displayed with tinted colormaps

Jorma Kekalainen

Digital Image Processing

460

Color bands of a truecolor image

displayed with tinted colormaps
>> I = imread('flowers.tif');

>> Red = I(:,:,1);

Scrambled/unscrambled

D = imread('Dee.jpg');
rng(2,'twister');
s=rng;
r = randperm(333);
c = randperm(500);
Scr = D(r,c,:);
figure
image(Scr)
truesize
title('Scrambled Image')
xlabel('What is it?')
imwrite(Scr,'Scrambled_Image.bmp',
'bmp');

Scrambled=imread('Scrambled_I
mage.bmp');
rng(2,'twister');
s=rng;
figure, imshow(Scrambled)
title('Scrambled image')
pause
r = randperm(333);
c = randperm(500);
Unscrambled(r,c,:)=Scrambled;
figure
image(Unscrambled)
truesize
title('Unscrambled Image')

Note: rng controls random number generation. Generator = 'twister' referred to the Mersenne
Jorma Kekalainen
Digital Image Processing
467
Twister
generator which is now Matlab startup
generator.

Example

Jorma Kekalainen

Digital Image Processing

468

Note: rng
Save the current generator settings in s:
s = rng;
Call rand to generate a vector of random values:
First, initialize the random number generator to make the results in this
example repeatable.
rng(1,'twister');
Save the generator settings in a structure, s.
s = rng;
x = rand(1,10)
Restore the original generator settings by calling rng.
Generate a new set of random values and verify that x and y are equal:
rng(s);
y = rand(1,10)
Jorma Kekalainen

Digital Image Processing

469

Obtaining information about

graphics files
The imfinfo function enables you to obtain information about
graphics files in any of the standard formats.
The information we obtain depends on the type of file, but it always
includes at least the following:
Name of the file, including the folder path if the file is not in the
current folder
File format
Version number of the file format
File modification date
File size in bytes
Image width in pixels
Image height in pixels
Number of bits per pixel
Image type: RGB (truecolor), intensity (grayscale), or indexed

Jorma Kekalainen

Digital Image Processing

470

imfinfo('ancienttrees.tif')
>> imfinfo('ancienttrees.tif')
ans =
Filename: [1x50 char]
FileModDate: '23-kes-2015 16:40:36'
FileSize: 75764
Format: 'tif'
FormatVersion: []
Width: 350
Height: 258
BitDepth: 8
ColorType: 'indexed'
FormatSignature: [73 73 42 0]
ByteOrder: 'little-endian'
NewSubFileType: 0
BitsPerSample: 8
Compression: 'PackBits'
Jorma Kekalainen

PhotometricInterpretation: 'RGB Palette'

Digital Image Processing

471

Information about our image

Much of those information is not useful to us; but we
can see the size of the image in pixels, the size of the
file (in bytes), the number of bits per pixel (this is
given by BitDepth), and the color type (in those case
indexed).
For comparison, let's look at the output of a true
color file (showing only the first few lines of the
output):
>>imfinfo('flowers.tif')
and binary file
>> imfinfo('text.png')
Jorma Kekalainen

65536 logical

Digital Image Processing

475

Data types and conversions

Elements in Matlab matrices may have a
number of different numeric data types; a few
of them are listed in table below.
These data types are also functions, we can
convert from one type to another.

Jorma
Kekalainen
Digital
Processing
Note:
There
are other data types; see
theImage
help
for datatypes.

476

Example
>> a=23;
>> b=uint8(a);
>> b
b=
23
>> whos a b
Name Size Bytes
Class
a
1x1 8
double array
b
1x1 1
uint8 array
Note: Even though the variables a and b have the same numeric value, they are of
different data types. An important consideration is that arithmetic operations are not
Jorma Kekalainen
Digital Image Processing
477
permitted
with the data types int8, int16,
uint8 and uint16.

Example
>> whos a b
Name Size
a
b

1x1
1x1

Bytes Class

Attributes

8 double
1 uint8

>> a/2
ans =
11.5000
>> b/2
ans =
12
Jorma Kekalainen

Digital Image Processing

478

Note
A grayscale image may consist of pixels whose values
are of data type uint8.
These images are thus reasonably efficient in terms
of storage space, since each pixel requires only one
byte.
Arithmetic operations are not permitted on this data
type; a uint8 image should be converted to double
before arithmetic is attempted.
However, most Image Processing Toolbox functions
accept uint8 and uint16 input.
Jorma Kekalainen

Digital Image Processing

479

uint8 function - converts to 8-bit

unsigned integer
intArray = uint8(array)
converts the elements of an array into unsigned 8-bit (1-byte) integers of
class uint8.
Input argument array is array of any numeric class, such as single or
double.
If array is already of class uint8, the uint8 function has no effect.
Output argument intArray is array of class uint8.
Values range from 0 to 28 1.
The uint8 function maps any values in array that are outside the limit to
the nearest endpoint.
For example,
>> uint8(2^8)
ans =
255
Jorma Kekalainen

Digital Image Processing

480

Converting between different

image types
We can convert images from one image type to
another. Table lists Matlab's functions for converting
between different image types.
Function

Use

Format

rgb2gray

RGB to grayscale

y=rgb2gray(x);

rgb2ind

RGB to indexed

[ind,cmap]=rgb2ind(x);

ind2rgb

Indexed to RGB

y=ind2rgb(ind,cmap);

gray2ind

Grayscale to indexed

[ind,cmap]=gray2ind(x);

ind2gray

Indexed to grayscale

y=ind2gray(ind,cmap);

Jorma Kekalainen

Digital Image Processing

481

rgb2gray
% RGB to grayscale y=rgb2gray(x)
I=imread('flowers.tif'); figure, imshow(I)
RGBsize=size(I), title('flowers.tif'), xlabel('Original image')
J=rgb2gray(I);
RGBsize =
figure, imshow(J), Graysize=size(J), title('rgb2gray')
362 500
xlabel('RGB to gray')
Graysize =

362 500

Jorma Kekalainen

Digital Image Processing

482

Note: 4 color

486

ind2rgb
% Indexed to RGB y=ind2rgb(x,map)
y=ind2rgb(ind,emap);
figure,imshow(y)
title('ind2rgb')
xlabel('Indexed to RGB using 6 colors')
emap, indsize=size(ind), RGBsize=size(y)
indsize =
362 500
RGBsize =
362 500

Jorma Kekalainen

Digital Image Processing

487

gray2ind
% Grayscale to indexed [y,map]=gray2ind(x,map)
[ind,map]=gray2ind(J);
figure, imshow(ind,map)
indsize=size(ind), mapsize=size(map)
title('gray2ind'), xlabel('Gray to indexed')
indsize =
362 500
mapsize =
64 3

Jorma Kekalainen

Digital Image Processing

488

ind2gray
% Indexed to grayscale y=ind2gray(x,map)
y=ind2gray(ind,map);
figure, imshow(y), title('ind2gray')
indsize=size(ind), mapsize=size(map),Graysize=size(y),
xlabel('Indexed to gray')
indsize =
362 500
mapsize =
64 3
Graysize =
362 500
Jorma Kekalainen

Digital Image Processing

489

494

Exercise 1
Test and study all the previously presented
Matlab command sequences.

Jorma Kekalainen

Digital Image Processing

495

Exercise 2
Type in the command
>> help imdemos
This will give you a list of, amongst other
things, all the sample PNG images which come
with the Image Processing Toolbox.
(a) Make a list of these sample images, and
(b) for each image determine its type (binary,
grayscale, true color or indexed color), and its
size (in pixels)
Jorma Kekalainen

Digital Image Processing

496

Exercise 3
Pick a grayscale image, e.g., cameraman.tif.
Using the imwrite function,write it to files of
type JPEG, PNG and BMP.
What are the sizes of those files?

Jorma Kekalainen

Digital Image Processing

497

Exercise 4

Repeat the previous question with some

(a) a binary image,
(b) an indexed color image,
(c) a true color image.

Jorma Kekalainen

Digital Image Processing

498

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

Displaying Images

(a) Original image

Jorma Kekalainen

(b) After conversion to type double

504

507

Results of scaling
Dividing by 512 darkens the image, as all
matrix values are now between 0 and 0.5, so
that the brightest pixel in the image is a midgray.
Dividing by 128 means that the range is 0-2,
and all pixels in the range 1-2 will be displayed
as white.
Thus the image has an over-exposed, washedout appearance.
Jorma Kekalainen

Digital Image Processing

508

Comment
The display of the result of a command whose
output is a matrix of type double can be
greatly affected by a judicious choice of a
scaling factor.

Jorma Kekalainen

Digital Image Processing

509

im2double
We can convert the original image to double more
properly using the function im2double.
This applies correct scaling so that the output values
are between 0 and 1.
So the commands
cd=im2double(c);
imshow(cd)
will produce a correct image.

Jorma Kekalainen

Digital Image Processing

510

double vs. im2double

It is important to make the distinction between the two
functions double and im2double: double changes the
data type but does not change the numeric values;
im2double changes both the numeric data type and
the values.
The exception, of course, is if the original image is of
type double, in which case im2double does nothing.
Although the command double is not of much use for
direct image display, it can be very useful for image
arithmetic as we have seen with scaling examples.

Jorma Kekalainen

Digital Image Processing

511

uint8 vs. im2uint8

Corresponding to the functions double and
im2double are the functions uint8 and im2uint8.
If we take our image cd of type double, properly
scaled so that all elements are between 0 and 1,
we can convert it back to an image of type uint8
in two ways:
c2=uint8(255*cd);
c3=im2uint8(cd);
Use of im2uint8 is to be preferred; it takes other
data types as input, and always returns a correct
result.
Jorma Kekalainen

Digital Image Processing

512

Image types
>> whos
Name
c
c2
c3
cd
cd1
cd512
Jorma Kekalainen

Size
256x256
256x256
256x256
256x256
256x256
256x256

Bytes
65536
65536
65536
524288
524288
524288
Digital Image Processing

Class

Attributes

uint8
uint8
uint8
double
double
double
513

521

Displaying each image in a

separate figure
The simplest way to display multiple images is to display them in
separate figure windows.
Matlab does not place any restrictions on the number of images
you can display simultaneously.
imshow always displays an image in the current figure.
If we display two images in succession, the second image replaces
the first image.
To view multiple figures with imshow, use the figure command to
explicitly create a new empty figure before calling imshow for the
next image.
For example, to view the first three frames in an array of truecolor
images F,

F=imread('peppers.png');
imshow(F(:,:,1))
figure, imshow(F(:,:,2))
figure, imshow(F(:,:,3))

Jorma Kekalainen

Digital Image Processing

522

Each image displayed separately

Jorma Kekalainen

Digital Image Processing

523

Displaying multiple images in the

same figure
We can use the imshow function with the Matlab
subplot function or the Matlab subimage function to
display multiple images in a single figure window.
subplot divides a figure into multiple display regions.
The syntax of subplot is subplot(m,n,p).
This syntax divides the figure into an m-by-n matrix of
display regions and makes the pth display region active.

Note: When we use subplot to display multiple color images in one figure window,
the images must share the colormap of the last image displayed. In some cases the
display results can be unacceptable. As an alternative, we can use the subimage
Jorma Kekalainen
Digital Image Processing
524
function,
or we can map all images to
the same colormap as we load them.

Displaying multiple images in the

same figure
E.g., we can use this syntax to display two images side by side

[X1,map1]=imread('forest.tif');
[X2,map2]=imread('trees.tif');
subplot(1,2,1), imshow(X1,map1)
subplot(1,2,2), imshow(X2,map2)

subimage converts images to truecolor before displaying them and

therefore circumvents the colormap sharing problem.
This example uses subimage to display the forest and the trees
images with better results

[X1,map1]=imread('forest.tif');
[X2,map2]=imread('trees.tif');
subplot(1,2,1), subimage(X1,map1)
subplot(1,2,2), subimage(X2,map2)

Bit planes
Grayscale images can be transformed into a sequence of
binary images by breaking them up into their bit-planes.
If we consider the gray value of each pixel of an 8-bit image
as an 8-bit binary word, then the 0th bit plane consists of
the last bit of each gray value.
Since this bit has the least effect in terms of the magnitude
of the value, it is called the least significant bit, and the
plane consisting of those bits the least significant bit plane.
Similarly the 7th bit plane consists of the first bit in each
value.
This bit has the greatest effect in terms of the magnitude of
the value, so it is called the most significant bit, and the
plane consisting of those bits the most significant bit plane.
Jorma Kekalainen

Digital Image Processing

531

Example
If we take a grayscale image, we start by making it a matrix of type double;
this means we can perform arithmetic on the values. E.g.,
>> c=imread('cameraman.tif');
>> cd=double(c);
We now isolate the bit planes by simply dividing the matrix cd by
successive powers of 2, taking the remainder, and seeing if the final bit is 0
or 1.
We can do this with the mod function.
c0=mod(cd,2);
c1=mod(floor(cd/2),2);
c2=mod(floor(cd/4),2);
c3=mod(floor(cd/8),2);
c4=mod(floor(cd/16),2);
c5=mod(floor(cd/32),2);
c6=mod(floor(cd/64),2);
c7=mod(floor(cd/128),2);
A corresponding example is seen in the following figures.
Jorma Kekalainen

Digital Image Processing

532

Example: The least significant bit plane

(= 0th bit plane)

c=imread('Hedyg-scale.tif');
cd=double(c);
c0=mod(cd,2);
imshow(c0)

544

All commands used in the

previous example

c=imread('Hedyg-scale.tif');
cd=double(c);
>> c0=mod(cd,2);
c1=mod(floor(cd/2),2);
c2=mod(floor(cd/4),2);
c3=mod(floor(cd/8),2);
c4=mod(floor(cd/16),2);
c5=mod(floor(cd/32),2);
c6=mod(floor(cd/64),2);
c7=mod(floor(cd/128),2);

Jorma Kekalainen

imshow(c0)
figure, imshow(c1)
figure, imshow(c2)
figure, imshow(c3)
figure, imshow(c4)
figure, imshow(c5)
figure, imshow(c6)
figure, imshow(c7)
cc=2*(2*(2*(2*(2*(2*(2*c7
+c6)+c5)+c4)+c3)+c2)+c1)+c
0;
figure,imshow(uint8(cc))

Digital Image Processing

545

Exercise
Transform the image cameraman.tif into a
sequence of binary images by breaking them
up into their bit-planes.

Jorma Kekalainen

Digital Image Processing

546

Solution

c=imread('cameraman.tif');
cd=double(c);
%We now isolate the bit planes by simply dividing the matrix cd by successive powers of 2, taking
%the remainder, and seeing if the final bit is 0 or 1. We can do this with the mod function.
c0=mod(cd,2);
c1=mod(floor(cd/2),2);
c2=mod(floor(cd/4),2);
c3=mod(floor(cd/8),2);
c4=mod(floor(cd/16),2);
c5=mod(floor(cd/32),2);
c6=mod(floor(cd/64),2);
c7=mod(floor(cd/128),2);
>> imshow(c0)
>> figure, imshow(c1)
>> figure, imshow(c2)
>> figure, imshow(c3)
>> figure, imshow(c4)
>> figure, imshow(c5)
>> figure, imshow(c6)
>> figure, imshow(c7)
>>

552

Example

Jorma Kekalainen

>>
AA=2*(2*(2*(2*(2*(2*(2*A7+A6)+A5)+A4)+A3)+A2)+A1)+A0
AA =
207 24 40 36 167
230 71 247 107 9
32 139 244 233 216
232 244 123 202 238
161 246 204 244 173
>> Ath=A>127
Ath =
1 0 0 0 1
1 0 1 0 0
0 1 1 1 1
1 1 0 1 1
1 1 1 1 1
>> A=floor(255*rand(5))
A=
207 24 40 36 167
230 71 247 107 9
32 139 244 233 216
232 244 123 202 238
161 246 204 244 173

Digital Image Processing

556

Digital Image Processing

Spatial Resolution

Spatial resolution
Spatial resolution is the density of pixels over
the image: the greater the spatial resolution,
the more pixels are used to display the image.
We can experiment with spatial resolution
with Matlab's imresize function.

Jorma Kekalainen

Digital Image Processing

558

imresize function
Suppose we have an 256*256 8-bit grayscale image saved to
the matrix x.
Then the command
imresize(x,1/2, 'nearest');
will halve the size of the image by taking out every other row
and every other column, thus leaving only those matrix
elements whose row and column indices are even:

imresize(x,1/2, 'nearest')

Jorma Kekalainen

Digital Image Processing

559

imresize function
If we apply imresize to The effective resolution of this
the result with the
new image is only 128*128.
parameter 2 rather than We can do all this at the same
1/2, all the pixels are
time:
repeated to produce an imresize(imresize(x,1/2,
'nearest'),2,'nearest');
image with about the
same size as the
original, but with half
the resolution in each
direction:
Jorma Kekalainen

Digital Image Processing

560

Example: Spatial resolution

>> A=floor(255*rand(6))
A=

28
198
99
61
102
24

Jorma Kekalainen

33 90 186 48 20
240 209 165 175 236
243 3 114 46 197
146 10 139 93 124
15 43 75 159 111
59 165 189 198 113
Digital Image Processing

561

Example: Spatial resolution

>> imresize(imresize(A,1/2, 'nearest'),2,'nearest')
ans =

240
240
146
146
59
59

Jorma Kekalainen

240 165 165 236 236

240 165 165 236 236
146 139 139 124 124
146 139 139 124 124
59 189 189 113 113
59 189 189 113 113
Digital Image Processing

562

Example: Spatial resolution

>> imresize(imresize(A,1/4, 'nearest'),4,'nearest')
ans =

3
3
3
3
165
165
165
165

Jorma Kekalainen

3 3 3
3 3 3
3 3 3
3 3 3
165 165
165 165
165 165
165 165

197
197
197
197
165
165
165
165

197
197
197
197
113
113
113
113

Digital Image Processing

113
113
113
113

563

Example: Spatial resolution

>> imresize(imresize(A,1/8, 'nearest'),8,'nearest')
ans =

159
159
159
159
159
159
159
159

Jorma Kekalainen

159
159
159
159
159
159
159
159

Digital Image Processing

159
159
159
159
159
159
159
159

564

Example: Spatial resolution

>>
Aold=imresize_old(imresize_old(
A,1/2),2)

>>
Aold1=imresize_old(imresize_old(
A,1/4),4)

Aold =

Aold1 =

240
240
146
146
59
59

240 165 165 236 236

240 165 165 236 236
146 139 139 124 124
146 139 139 124 124
59 189 189 113 113
59 189 189 113 113

Jorma Kekalainen

3
3
3
3

Digital Image Processing

3
3
3
3

565

Note
About the same effect we can achieve
applying the command
imresize_old( imresize_old(x,1/2),2);

Jorma Kekalainen

Digital Image Processing

566

Effective resolution
By changing the parameters of imresize, we
can change the effective resolution of the
image to smaller amounts:

Jorma Kekalainen

Digital Image Processing

567

Example
To see the effects of these commands,
suppose we apply them to the image
Deepiga.png:
x=imread('Deepiga.png');
The effects of decreasing blockiness or
pixelization become quite pronounced as the
resolution increases.

Jorma Kekalainen

Digital Image Processing

568

Example

figure,imshow(x)

Jorma Kekalainen

Digital Image Processing

Digital Image Processing

582

Arithmetic operations
These operations act by applying a simple
function y=f(x) to each gray value in the image.
Thus f(x) is a function which maps the range 0
255 onto itself.
Simple functions include adding or subtract a
constant value to each pixel: y=xC or multiplying
each pixel by a constant: y=Cx.
In each case we may have to fix the output
slightly in order to ensure that the results are
integers in the range 0 255.
Jorma Kekalainen

Digital Image Processing

583

Arithmetic operations
We can do this by first rounding the result (if
necessary) to obtain an integer, and then clipping
the values by setting:
255, if y > 255
y=
0, if y < 0

We can see how these operations affect an image

Image complementation
The complement of a grayscale image is its
photographic negative.
If an image matrix m is of type double and so its
gray values are in the range 0.0 1.0, we can
obtain its negative with the command
>> 1-m
If the image is binary, we can use
>> ~m
If the image is of type uint8, the best approach is
the imcomplement function.
Jorma Kekalainen

Digital Image Processing

594

Binary conversion
BI=im2bw(Deepiga,0.5);% Binary conversion with threshold
>> imshow(BI)
>> figure, imshow(~BI)

Jorma Kekalainen

Digital Image Processing

595

Note: im2bw

Convert image to binary image, based on

threshold
Syntax

BW = im2bw(I, level)
BW = im2bw(X, map, level)

BW = im2bw(RGB, level)
Description
BW = im2bw(I, level) converts the grayscale
image I to a binary image.
The output image BW replaces all pixels in the
input image with luminance greater than level

with the value 1 (white) and replaces all other

pixels with the value 0 (black).
Specify level in the range [0,1].
This range is relative to the signal levels
possible for the image's class.
Therefore, a level value of 0.5 is midway
between black and white, regardless of class.

Jorma Kekalainen

To compute the level argument, you

can use the function graythresh.
If you do not specify level, im2bw
uses the value 0.5.
BW = im2bw(X, map, level) converts
the indexed image X with colormap
map to a binary image.
BW = im2bw(RGB, level) converts the
truecolor image RGB to a binary
image.
If the input image is not a grayscale
image, im2bw converts the input
image to grayscale, and then converts
this grayscale image to binary by
thresholding.

Digital Image Processing

596

Image complementation
Next figure shows the complement function y=255-x, and the
result of the commands
>> cc=imcomplement(c);
>> imshow(cc)

Jorma Kekalainen

Digital Image Processing

597

Part complementation
Interesting special effects can be obtained by
complementing only part of the image.
E.g., by taking the complement of pixels of gray
value 128 or less, and leaving other pixels
untouched.
Or we could take the complement of pixels which
are 128 or greater, and leave other pixels
untouched.
Figure shows these functions.
The effect of these functions is called solarization.
Jorma Kekalainen

Digital Image Processing

598

Solarization

Complementing only dark pixels

Jorma Kekalainen

Complementing only light pixels

Digital Image Processing

599

Histograms
Given a grayscale image, its histogram consists of
the histogram of its gray levels; that is, a graph
indicating the number of times each gray level
occurs in the image.
We can deduce a great deal about the
appearance of an image from its histogram:
In a dark image, the gray levels (and hence the
histogram) would be clustered at the lower end.
In a uniformly bright image, the gray levels would be
clustered at the upper end.
In a well contrasted image, the gray levels would be
well spread out over much of the range.
Jorma Kekalainen

Digital Image Processing

600

Different contrasted images

Very low contrast, low contrast and medium
contrast images

Jorma Kekalainen

Digital Image Processing

601

imhist
We can view the histogram of an image in
Matlab by using the imhist function:
>> p=imread('pout.tif');
>> imshow(p),figure,imhist(p),axis tight
(the axis tight command ensures the axes of
the histogram are automatically scaled to fit
all the values in).

Jorma Kekalainen

Digital Image Processing

602

Image and its histogram

pout.tif

605

Histogram stretching
We can stretch the gray
levels in the centre of the
range out by applying the
piecewise linear function
shown at the right.
This function has the effect
of stretching the gray levels
5-9 to gray levels 2-14
according to the equation:

The equation yields:

where i is the original gray

level and j its result after the
transformation.
Jorma Kekalainen
Digital Image Processing

606

Stretching function
Gray levels outside this
range are either left alone
as in this case (or
transformed according to
the linear functions at the
ends of the graph).

Jorma Kekalainen

Digital Image Processing

607

Corresponding stretched
histogram
A stretched histogram indicates an image
with greater contrast than the original.

Jorma Kekalainen

Digital Image Processing

A histogram of a poorly
contrasted image

608

Use of imadjust in stretching

To perform histogram
stretching in Matlab the
imadjust function may
be used.
In its simplest form, the
command
imadjust(im,[a,b],[c,d])
stretches the image
according to the
function shown in
figure.

The stretching function

given by imadjust

Note: Pixel values less than a are all converted to c and pixel values greater than b are
Jorma Kekalainen
Digital Image Processing
609
all converted
to d.

Use of imadjust in stretching

Since imadjust is designed to work equally well on images of
type double, uint8 or uint16 the values of a, b, and d must be
between 0 and 1; the function automatically converts the
image (if needed) to be of type double.
If either of [a,b] or [c,d] are chosen to be [0,1], the
abbreviation [] may be used.
Thus, for example, the command
>> imadjust(im,[],[])
does nothing, and the command
>> imadjust(im,[],[1,0])
inverts the gray values of the image,
to produce a result similar
to a photographic negative.
Jorma Kekalainen
Digital Image Processing
610

Example

P=imread('pout.tif');
figure, imshow(P)
title('Image before histogram streching')
figure, imhist(P)
title('Image histogram before histogram streching')
Padj=imadjust(P,[75/255,150/255],[0,1]);
figure, imshow(Padj)
title('Image after histogram streching')
figure, imhist(Padj)
title('Image after histogram streching')
figure,plot(P,Padj,'.'),axis tight
title('Stretching function imadjust(P,[75/255,150/255],[0,1])')

Jorma Kekalainen

Digital Image Processing

611

617

Stretching function
We may view the imadjust stretching function
with the plot function.
For example,
>> plot(t,th,'.'),axis tight
produces the plot shown in the following figure.
Since t and th are matrices which contain the
original values and the values after the imadjust
function, the plot function simply plots them,
using dots () to do it.
Jorma Kekalainen

Digital Image Processing

618

Stretching function
Stretching function used in the previous tire image.
=0.5 (<1 concave downward)

Jorma Kekalainen

Digital Image Processing

619

Piecewise linear stretching function

We can write our own function to perform piecewise linear
stretching as shown in figure below.
To do this, we will make use of the find function, to find the
pixel values in the image between ai and ai +1.
Since the line between the coordinates (ai, bj)
and (ai+1, bj+1) has the equation

the heart of our function will be the lines

pix=find(im >= a(i) & im < a(i+1));
out(pix)=(im(pix)-a(i))*(b(i+1)-b(i))/(a(i+1)-a(i))+b(i);
where im is the input image and out is the output image.
Jorma Kekalainen

Digital Image Processing

620

Matlab function for piecewise

linear stretching function
A simple procedure which takes as inputs
images of type uint8 or double.
function out = histpwl(im,a,b)
% HISTPWL(IM,A,B) applies a piecewise linear
%transformation to the pixel values
% of image IM, where A and B are vectors
% containing the x and y coordinates of the
% ends of the line segments.
%IM can be of type UINT8 or DOUBLE, and
% the values in A and B must be between
% 0 and 1.
% For example:
% histpwl(x,[0,1],[1,0])
% simply inverts the pixel values.
%
Jorma Kekalainen

classChanged = 0;
if ~isa(im, 'double'),
classChanged = 1;
im = im2double(im);
end
if length(a) ~= length (b)
error('Vectors A and B must be of equal size');
end
N=length(a);
out=zeros(size(im));
for i=1:N-1
pix=find(im>=a(i) & im<a(i+1));
out(pix)=(im(pix)-a(i))*(b(i+1)-b(i))/(a(i+1)-a(i))+b(i);
end
pix=find(im==a(N));
out(pix)=b(N);
if classChanged==1
out = uint8(255*out);
end

Digital Image Processing

621

Example: histpwl
As an example of the use of this function:
>> th=histpwl(t,[0 .25 .5 .75 1],[0 .75 .25 .5 1]);
>> imshow(th)
>> figure,plot(t,th,'.'),axis tight
produces the figures shown below.

Jorma Kekalainen

Digital Image Processing

622

Digital Image Processing

626

Example

To equalize this
histogram, we
form running
totals of the ni ,
and multiply each
by 15/360=1/24.

Jorma Kekalainen

Digital Image Processing

627

Example
We now have the following transformation of gray
values, obtained by reading off the first and last
columns in the previous table:

The histogram of the j values is shown in the

following figure.
This is far more spread out than the original
histogram, and so the resulting image should exhibit
greater contrast.
Jorma Kekalainen

Digital Image Processing

628

Jorma Kekalainen

Digital Image Processing

636

Example

[he,emap]=gray2ind(H);
figure, imshow(he)
title('Index values only')
figure,imhist(he(:,:,1)),axis tight
title('Index matrix consisting low values only')

Jorma Kekalainen

Digital Image Processing

637

After equalization
hee=histeq(he(:,:,1));
figure, imshow(hee),
title('After equalization')
figure,imhist(hee),axis tight
title('After equalization')

Jorma Kekalainen

Digital Image Processing

638

Example

Jorma Kekalainen

Digital Image Processing

639

Comment
Consider the histogram
beside.
To apply histogram
stretching, we would
need to stretch out the
values between gray
levels 9 and 13.
Thus, we would need to
apply a piecewise
function similar to that
shown beside.
Jorma Kekalainen

Digital Image Processing

640

Comment
Let's consider the
Cumulative histogram
cumulative histogram,
which is shown beside.
The dashed line is simply
joining the top of the
histogram bars.
However, it can be
interpreted as an
appropriate histogram
stretching function.
To do this, we need to
scale the y values so that
they are between 0 and
But this is precisely the method
15, rather than 0 and 360. described in the histogram
Jorma Kekalainen

Digital Image Processing

equalization section.

641

Digital Image Processing

Exercises

Exercise on image import and

export
(a) Read and display any image with very bad
contrast (Construct such an image if needed)
(b) Check the image appearance in the
workspace
(c) Improve image contrast
(d) Write the image to a disk file
(e) Check the contents of the newly written
file
Jorma Kekalainen

Digital Image Processing

643

(a) Read and display an image

First, we clear the Matlab workspace of any
variables and close open figure windows.
clear, close all
To read an image, we use the imread command.
Here we read an image file ex4-fig1.png in an
array named I.
I = imread('ex4-fig1.png');
An acronym png means Portable Network Graphics
(PNG) which is one of many graphics file formats.
Note: For the list of supported graphics file formats, see the imread
Jorma Kekalainen
Digital Image Processing
644
function
reference documentation.

(a) Read and display an image

>> imfinfo('ex4-fig1.png')
ans =

Filename: [1x46 char]

FileModDate: '07-Aug-2015 21:48:40'

FileSize: 16388

Format: 'png'

FormatVersion: []

Width: 357

Height: 221

BitDepth: 8

ColorType: 'grayscale'

FormatSignature: [137 80 78 71 13 10 26 10]

Colormap: []

Comparison of images

>> subplot(121),imshow(I)
>> title('Grayscale image ex4-fig1.png')
>> subplot(122),imshow(I2)
>> title('Equalized image')
>> truesize

Digital Image Processing

Thresholding

Single thresholding
A grayscale image is turned into a binary (black and
white) image by first choosing a gray level T in the
original image, and then turning every pixel black or
white according to whether its gray value is greater than
or less than T
white if its gray level is >T
A pixel becomes
black if its gray level is T
Thresholding is a vital part of image segmentation,
where we wish to isolate objects from the background.
It is also an important component of robot vision.
Jorma Kekalainen

Jorma Kekalainen

148
110
113
106
113
183
119
114

145
185
193
158
108
137
188
135

178
155
136
130
127
162
179
123

132
118
146
178
144
105
100
134

174
165
108
170
139
169
151
183

IM=uint8(data)
thIM1=IM>100
thIM2=IM>150
figure, imshow(IM)
title('The small image before thresholding')
figure, imshow(thIM1)
title('The small image after thresholding>100')
figure, imshow(thIM2)
title('The small image after thresholding>150')

Digital Image Processing

672

Example
IM =
3 148
2 176
0 100
0 155
9 196
6 188
9 122
8 176

117
174
124
170
138
143
156
137

148
110
113
106
113
183
119
114

Jorma Kekalainen

145
185
193
158
108
137
188
135

178
155
136
130
127
162
179
123

132
118
146
178
144
105
100
134

174
165
108
170
139
169
151
183

thIM1 =
0 1
0 1
0 0
0 1
0 1
0 1
0 1
0 1
thIM2 =
0 0
0 1
0 0
0 1
0 1
0 1
0 0
0 1

1
1
1
1
1
1
1
1

1
1
1
1
1
1
0
1

1
1
1
1
1
1
1
1

0
1
0
1
0
0
1
0

0
0
0
0
0
1
0
0

0
1
1
1
0
0
1
0

1
1
0
0
0
1
1
0

0
0
0
1
0
0
0
0

1
1
0
1
0
1
1
1

Digital Image Processing

673

Note
>> whos
Name

Size

8x8
8x8
8x8
8x8

IM
data
thIM1
thIM2

Jorma Kekalainen

Bytes Class
64
512
64
64
Digital Image Processing

Attributes

uint8
double
logical
logical
674

Example
The rice image shown previously has light grains on a dark
background; an image with dark objects over a light
background may be treated the same:
>> b=imread('bacteria.tif');
>> imshow(b),figure,imshow(b>100)
will produce the images shown below

Jorma Kekalainen

Digital Image Processing

675

im2bw function
Besides the above method, Matlab has the im2bw
function, which thresholds an image of any data
type, using the general syntax
im2bw(image,level)
where level is a value between 0 and 1 (inclusive),
indicating the fraction of gray values to be turned
white.
This command will work on grayscale,
colored and indexed images of data
type
Jormauint8,
Kekalainen uint16 or double.
Digital Image Processing

676

im2bw function
For example, the thresholded rice and bacteria images above
could be obtained using
>> im2bw(r,0.47);
>> im2bw(b,0.39);
The im2bw function automatically scales the value level to a
gray value appropriate to the image type, and then performs a
thresholding by our first method.

Jorma Kekalainen

Digital Image Processing

677

Exercise
Consider the following 8*8 image
3 148 117 148 145 178 132 174
2 176 174 110 185 155 118 165
0 100 124 113 193 136 146 108
0 155 170 106 158 130 178 170
9 196 138 113 108 127 144 139
6 188 143 183 137 162 105 169
9 122 156 119 188 179 100 151
8 176 137 114 135 123 134 183
Using im2bw function threshold it at
(a) level 100
(b) level 150
Jorma Kekalainen

Digital Image Processing

678

Solution
>> data=[3 148 117 148 145 178 132 174
2 176 174 110 185 155 118 165
0 100 124 113 193 136 146 108
0 155 170 106 158 130 178 170
9 196 138 113 108 127 144 139
6 188 143 183 137 162 105 169
9 122 156 119 188 179 100 151
8 176 137 114 135 123 134 183
]
data =
3 148 117 148 145 178 132 174
2 176 174 110 185 155 118 165
0 100 124 113 193 136 146 108
0 155 170 106 158 130 178 170
9 196 138 113 108 127 144 139
6 188 143 183 137 162 105 169
9 122 156 119 188 179 100 151
8 176 137 114 135 123 134 183
Jorma Kekalainen

IM=uint8(data)
thIM1=im2bw(IM,100/255)
thIM2= im2bw(IM,150/255)
figure, imshow(IM)
title('The small image before thresholding')
figure, imshow(thIM1)
title('The small image after thresholding>100')
figure, imshow(thIM2)
title('The small image after thresholding>150')

Digital Image Processing

679

Solution
data =
3 148
2 176
0 100
0 155
9 196
6 188
9 122
8 176

117
174
124
170
138
143
156
137

148
110
113
106
113
183
119
114

Jorma Kekalainen

145
185
193
158
108
137
188
135

178
155
136
130
127
162
179
123

132
118
146
178
144
105
100
134

174
165
108
170
139
169
151
183

thIM1 =
0 1
0 1
0 0
0 1
0 1
0 1
0 1
0 1

1
1
1
1
1
1
1
1

1
1
1
1
1
1
0
1

1
1
1
1
1
1
1
1

thIM2 =
0 0 0 0 0 1
0 1 1 0 1 1
0 0 0 0 1 0
0 1 1 0 1 0
0 1 0 0 0 0
0 1 0 1 0 1
0 0 1 0 1 1
0 Digital
1 Image
0 0Processing
0 0

0
0
0
1
0
0
0
0

1
1
0
1
0
1
1
1

680

Example
Besides isolating objects from the background,
thresholding provides a very simple way of
showing hidden aspects of an image.
For example, the image paper.tif appears all
white, as nearly all the gray values are very
high.
However, thresholding at a high level
produces an image of far greater interest.
Jorma Kekalainen

Digital Image Processing

681

Example
We can use the commands
>> p=imread('paper1.tif');
>> imshow(p),figure,imshow(p>241)
to provide the images shown below

Jorma Kekalainen

Digital Image Processing

682

Double thresholding
Here we choose two values T1 and T2 and apply a thresholding
operation as
white if its gray level is between T1 and T2
A pixel becomes
black if its gray level is otherwise
We can implement this by a simple variation on the above
method:
X>T1 & X<T2
Since the ampersand acts as a logical and, the result will only
produce a one where both inequalities are satisfied.
Jorma Kekalainen

Digital Image Processing

683

Double threshold it at
(a) level 100
(b) level 150

Apply thresholding operation so that

white if its gray level is between 100 and 150
A pixel becomes
Jorma Kekalainen
Processing
blackDigital
if itsImage
gray
level is otherwise

684

Example
>> data=[3 148 117 148 145 178 132 174
2 176 174 110 185 155 118 165
0 100 124 113 193 136 146 108
0 155 170 106 158 130 178 170
IM=uint8(data)
9 196 138 113 108 127 144 139
thIM=IM>100 & IM<150
6 188 143 183 137 162 105 169
figure, imshow(IM)
9 122 156 119 188 179 100 151
title('Image before thresholding')
8 176 137 114 135 123 134 183
figure, imshow(thIM)
]

title('Image after thresholding (IM>100 & IM<150)')

data =
3 148 117
2 176 174
0 100 124
0 155 170
9 196 138
6 188 143
9 122 156
Jorma Kekalainen
8 176 137

148
110
113
106
113
183
119
114

145
185
193
158
108
137
188
135

178
155
136
130
127
162
179
123

132
118
146
178
144
105
100
134

174
165
108
170
139
169
151
183 Digital Image Processing

685

Example
IM =
3 148
2 176
0 100
0 155
9 196
6 188
9 122
8 176

117
174
124
170
138
143
156
137

148
110
113
106
113
183
119
114

Jorma Kekalainen

145
185
193
158
108
137
188
135

178
155
136
130
127
162
179
123

132
118
146
178
144
105
100
134

174
165
108
170
139
169
151
183

thIM =
0 1
0 0
0 0
0 0
0 0
0 0
0 1
0 0

Digital Image Processing

1
0
1
0
1
1
0
1

1
1
1
1
1
0
1
1

1
0
0
0
1
1
0
1

0
0
1
1
1
0
0
1

1
1
1
0
1
1
0
1

0
0
1
0
1
0
0
0

686

Example
Consider the following sequence of commands, which start by
producing an 8-bit gray version of the indexed image spine.tif:
Double thresholding brings out subtle
[x,map]=imread('spine.tif');
features of the spine which single
s=uint8(ind2gray(x,map));
thresholding would be unable to do.
figure,imshow(s),figure,imshow(s>115 & s<125)

Jorma Kekalainen

Digital Image Processing

687

Note
We can obtain similar results using im2bw:
imshow(im2bw(x,map,0.45)&~im2bw(x,map,0.5))

Jorma Kekalainen

Digital Image Processing

688

Note: ~
~ means logical not.
~A performs a logical not of input array A, and
returns an array containing elements set to
either logical 1 (TRUE) or logical 0 (FALSE).
An element of the output array is set to 1 if A
contains a zero value element at that same
array location.
Otherwise, that element is set to 0.
Jorma Kekalainen

Digital Image Processing

689

Applications of thresholding
Thresholding can be useful in the following situations:
(a) When we want to remove unnecessary detail from an
image and to concentrate on essentials.
(b) To bring out hidden detail.
(c) When we want to remove a varying background from text
or a drawing.
Examples of (a) were given in the rice and bacteria images: by
removing all gray level information, the rice and bacteria were
reduced to binary blobs.
But this information may be all we need to investigate sizes,
shapes, or numbers of blobs.
Jorma Kekalainen

Digital Image Processing

690

Example

693

Example

R=rand(107,122)*128+127;
L=imread('logo.tif');
figure, imshow(L)
TL=uint8(R.*double(not(L)));
figure, imshow(TL)
figure, imshow(TL>100)

Jorma Kekalainen

Digital Image Processing

694

Exercise
Superimpose the image text.png onto the
image cameraman.tif:
t=imread('text.png');
c=imread('cameraman.tif');
m=uint8(double(c).*double(~t));
Can you threshold this new image m to isolate
the text?

Jorma Kekalainen

Digital Image Processing

695

Solution

Superimpose the image text.png onto the image cameraman.tif:

t=imread('text.png');
c=imread('cameraman.tif');
m=uint8(double(c).*double(~t));
figure, imshow(m),figure, imshow(m>100)
figure, imshow(m>10),figure, imshow(m>5)

Ans. Yes, we can threshold this new image m

to isolate the text!

The idea is to move a mask - a rectangle (usually with sides of
odd length) or other shape over the given image.
As we do this, we create a new image whose pixels have gray
values calculated from the gray values under the mask, as
shown below.

Jorma Kekalainen

Digital Image Processing

700

Filter
The combination of mask and function is called a
filter.
If the function by which the new gray value is
calculated is a linear function of all the gray
values in the mask, then the filter is called a
linear filter.
We can implement a linear filter by multiplying
all elements in the mask by corresponding
elements in the neighborhood, and adding up all
these products.
Jorma Kekalainen

Digital Image Processing

701

Spatial filter
Suppose that the 3*5 mask
values are given by:

and that corresponding

pixel values are

We now multiply and add:

Jorma Kekalainen

Digital Image Processing

702

Spatial filtering
Spatial filtering requires three steps:
1. position the mask over the current pixel,
2. form all products of filter elements with the
corresponding elements of the neighborhood,
3. add up all the products.
This process must be repeated for every pixel
in the image.

Jorma Kekalainen

Digital Image Processing

703

Spatial filtering schema

Jorma Kekalainen

Digital Image Processing

704

Example
An example: One important linear filter is to use a (3*3) mask
and take the average of all nine values within the mask.
This value becomes the gray value of the corresponding pixel
in the new image.
We may describe this operation as follows:

where e is gray value of the current pixel in the original image,

and the average is the gray value of the corresponding pixel in
the new image.
Jorma Kekalainen
Digital Image Processing
705

Example
To apply this to an image, consider the 5*5 image obtained by:
x=uint8(10*magic(5))
x=
170
240
10
80
150
230
50
70
140
160
40
60
130
200
220
100
120
190
210
30
110
180
250
20
90
We may regard this array as being made of nine overlapping 3*3
neighborhoods.
The output of our working will thus consist only of nine values.
Jorma Kekalainen
Digital Image Processing
Note:
We shall see later how to obtain
25 values in the output.

706

Example
Consider the top left 3*3 neighborhood of our image x:
170 240 10
80
150
230 50
70
140 160
40
60
130 200 220
100 120 190 210 30
110 180 250 20
90
Now we take the average of all these values:
mean2(x(1:3,1:3))
ans =
111.1111
which we can round to 111.
Jorma Kekalainen
Digital Image Processing
707
Note:
mean2(A) computes the
average or mean of matrix elements.

Example
Now we can move to the second neighborhood:
170 240 10
80
150
230 50
70
140 160
40
60
130 200 220
100 120 190 210 30
110 180 250 20
90
and take its average:
mean2(x(1:3,2:4))
ans =
108.8889
We can round this either down to 108, or to the nearest
integer 109.
Jorma Kekalainen
Digital Image Processing

708

Example
If we continue in this manner, we will build up
the following output:
111.1111
108.8889 128.8889
110.0000
130.0000 150.0000
131.1111
151.1111 148.8889
This array is the result of filtering x with the
3*3 averaging filter.

Jorma Kekalainen

Digital Image Processing

709

Matrix notation
It is convenient to describe a linear filter simply in terms of
the coefficients of all the gray values of pixels within the
mask.
This can be written as a matrix.
The averaging filter above, for example, could have its output
written as

and so we can describe this filter by the matrix

Jorma Kekalainen

Digital Image Processing

710

Example
The filter

would operate on gray values as

a-2b+c-2d+4e-2f+g-2h+i

Jorma Kekalainen

Digital Image Processing

711

Edges of the image

There is an obvious problem in applying a filter at the
edge of the image, where the mask partly falls
outside the image.
In such a case, as illustrated below there will be a
lack of gray values to use in the filter function.
A mask at the edge of an image

Note: There is no single best approach; the method must be dictated by

the problem at hand; by the filter being used, and by the result
Jorma Kekalainen
Digital Image Processing
712
required.

Dealing with edge problem

a=ones(3,3)/9
240 10 80 150
50 70 140 160
60 130 200 220
120 190 210 30
180 250 20 90

a=
0.1111 0.1111 0.1111
0.1111 0.1111 0.1111
0.1111 0.1111 0.1111

filter2(a,x,'same')
ans =
76.6667
87.7778
66.6667
67.7778
56.6667

85.5556
111.1111
110.0000
131.1111
105.5556

Jorma Kekalainen

uint8(filter2(a,x,'same'))
65.5556 67.7778 58.8889
108.8889 128.8889 105.5556
130.0000 150.0000 106.6667
151.1111 148.8889 85.5556
107.7778 87.7778 38.8889
Digital Image Processing

ans =
77 86
88 111
67 110
68 131
57 106

66 68 59
109 129 106
130 150 107
151 149 86
108 88 39
717

filter2(filter,image,'valid')
filter2(filter,image,'valid') applies the mask
only to inside pixels.
We can see that the result will always be
smaller than the original:
filter2(a,x,'valid')

uint8( filter2(a,x,'valid'))

ans =
111.1111 108.8889 128.8889
110.0000 130.0000 150.0000
131.1111 151.1111 148.8889

ans =
111 109 129
110 130 150
131 151 149

Jorma Kekalainen

Digital Image Processing

718

Note
We can obtain the previous result of 'same' by padding with
zeros and using 'valid':
x2=zeros(7,7); x2(2:6,2:6)=x
x2 =

filter2(a,x2,'valid')

Jorma Kekalainen

Digital Image Processing

719

filter2(filter,image,'full')
filter2(filter,image,'full') returns a
result larger than the original; it
does this by padding with zero, and
applying the filter at all places on
and around the image where the
mask intersects the image matrix.
filter2(a,x,'full')
ans =
18.8889
44.4444
48.8889
41.1111
27.7778
23.3333
12.2222

45.5556
76.6667
87.7778
66.6667
67.7778
56.6667
32.2222

46.6667
85.5556
111.1111
110.0000
131.1111
105.5556
60.0000

uint8(filter2(a,x,'full'))
ans =
19 46
44 77
49 88
41 67
28 68
23 57
12 32

47
86
111
110
131
106
60

37 27 26 17
66 68 59 34
109 129 106 59
130 150 107 46
151 149 86 38
108 88 39 13
50 40 12 10

36.6667 26.6667 25.5556 16.6667

65.5556 67.7778 58.8889 34.4444
108.8889 128.8889 105.5556 58.8889
130.0000 150.0000 106.6667 45.5556
151.1111 148.8889 85.5556 37.7778
107.7778 87.7778 38.8889 13.3333
50.0000 40.0000 12.2222 10.0000

filter2(a,x,'same')
filter2(a,x,'valid')

Note: The shape parameter, being optional, can be omitted; in which case the default
Kekalainen
Digital Image Processing
720
valueJorma
is 'same'.

Compare
filter2(a,x,'same')
ans =
76.6667
87.7778
66.6667
67.7778
56.6667

85.5556
111.1111
110.0000
131.1111
105.5556

65.5556 67.7778 58.8889

108.8889 128.8889 105.5556
130.0000 150.0000 106.6667
151.1111 148.8889 85.5556
107.7778 87.7778 38.8889

filter2(a,x,'valid')
ans =
111.1111 108.8889 128.8889
110.0000 130.0000 150.0000
131.1111 151.1111 148.8889

conv2(double(x),a,'same')
ans =
76.6667 85.5556
87.7778 111.1111
66.6667 110.0000
67.7778 131.1111
Jorma Kekalainen
56.6667 105.5556

65.5556 67.7778 58.8889

108.8889 128.8889 105.5556
130.0000 150.0000 106.6667
151.1111 148.8889 85.5556
Digital Image Processing
107.7778 87.7778 38.8889

conv2(double(x),a,'valid')
ans =
111.1111 108.8889 128.8889
110.0000 130.0000 150.0000
721
131.1111 151.1111 148.8889

Compare
conv2(double(x),a,'full')
ans =
18.8889
44.4444
48.8889
41.1111
27.7778
23.3333
12.2222

45.5556
76.6667
87.7778
66.6667
67.7778
56.6667
32.2222

46.6667
85.5556
111.1111
110.0000
131.1111
105.5556
60.0000

or conv2(double(x),a)
36.6667 26.6667 25.5556 16.6667
65.5556 67.7778 58.8889 34.4444
108.8889 128.8889 105.5556 58.8889
130.0000 150.0000 106.6667 45.5556
151.1111 148.8889 85.5556 37.7778
107.7778 87.7778 38.8889 13.3333
50.0000 40.0000 12.2222 10.0000

filter2(a,x,'full')
ans =
18.8889 45.5556
44.4444 76.6667
48.8889 87.7778
41.1111 66.6667
27.7778 67.7778
23.3333 56.6667
Jorma Kekalainen
12.2222 32.2222

46.6667
85.5556
111.1111
110.0000
131.1111
105.5556
60.0000

36.6667 26.6667 25.5556 16.6667

65.5556 67.7778 58.8889 34.4444
108.8889 128.8889 105.5556 58.8889
130.0000 150.0000 106.6667 45.5556
151.1111 148.8889 85.5556 37.7778
107.7778 87.7778 38.8889 13.3333
Digital Image Processing
50.0000
40.0000 12.2222 10.0000

filter2(a,x,'same')
filter2(a,x,'valid')

722

Note
C = conv2(A,B) computes the two-dimensional convolution of
matrices A and B.
If one of these matrices describes a two-dimensional finite impulse
response (FIR) filter, the other matrix is filtered in two dimensions.
The size of C is determined as follows: if [ma,na] = size(A), [mb,nb]
= size(B), and [mc,nc] = size(C), then mc = max([ma+mb-1,ma,mb])
and nc = max([na+nb-1,na,nb]).
C = conv2(...,shape) returns a subsection of the two-dimensional
convolution, as specified by the shape parameter:
'full' Returns the full two-dimensional convolution (default).
'same' Returns the central part of the convolution of the same size
as A.
'valid' Returns only those parts of the convolution that are
computed without the zero-padded edges. Using this option, size(C)
= max([ma-max(0,mb-1),na-max (0,nb-1)],0).
Jorma Kekalainen
Image Processing
Note:
All numeric inputs to conv2Digital
must
be of type double or single.

723

Two-dimensional convolution
If a and b are functions of two discrete variables, n1 and n2,
then the formula for the two-dimensional convolution of a
and b is

In practice, of course, conv2 computes the convolution for

finite intervals.
Note that matrix indices in Matlab software always start at
1 rather than 0.
Therefore, matrix elements A(1,1), B(1,1), and C(1,1)
correspond to mathematical quantities a(0,0), b(0,0), and
c(0,0).
Jorma Kekalainen

Digital Image Processing

724

fspecial function
We can create our filters by hand, or by using the fspecial
function; this has many options which makes for easy creation
of many different filters.
We shall use the average option, which produces averaging
filters of given size; thus fspecial('average',[5,7]) will return
an averaging filter of size 5*7 or more simply
fspecial('average',11) will return an averaging filter of size
11*11.
If we leave out the final number or vector, the 3*3 averaging
filter is returned.

Jorma Kekalainen

Digital Image Processing

725

Example
For example, suppose we apply the 3*3 averaging filter to an
image as follows:
c=imread('cameraman.tif');
f1=fspecial('average');
cf1=filter2(f1,c);
We now have a matrix of data type double.
To display this, we can do any of the following:
transform it to a matrix of type uint8, for use with imshow,
divide its values by 255 to obtain a matrix with values in
the 0.1 1.0 range, for use with imshow,
use mat2gray to scale the result for display (later more on
this function).
Jorma Kekalainen

Digital Image Processing

726

Example
Using the second method i.e. functions
figure,imshow(c),figure,imshow(cf1/255)
will produce the images shown below.
c=imread('cameraman.tif');
f1=fspecial('average');
or
cf1=filter2(f1,c);
figure,imshow(c),figure,imshow(cf1/255)

imshow(uint8(cf1))

729

Example
f1=fspecial('average',[21,21]);
cf1=filter2(f1,c);
figure,imshow(cf1/255)
f1=fspecial('average',[29,29]);
cf1=filter2(f1,c);
figure,imshow(cf1/255)
The averaging filter blurs the image; the
edges in particular are less distinct than in
the original.
The image can be further blurred by using
an averaging filter of larger size.
Jorma Kekalainen

Digital Image Processing

730

Comment on average filtering

Notice how the zero padding used at the edges has resulted in a dark
border appearing around the image.
This is especially noticeable when a large filter is being used.
If this is an unwanted artifact of the filtering; if for example it changes the
average brightness of the image, then it may be more appropriate to use
the 'valid' shape option.
The resulting image after these filters may appear to be much worse than
the original.
However, applying a blurring filter to reduce details in an image may the
perfect operation for autonomous machine recognition, or if we are only
concentrating on such aspects of the image as numbers of objects or
amount of dark and light areas.
In those cases, too much details may obscure the outcome.

Jorma Kekalainen

Digital Image Processing

731

Frequency concept
Fundamentally, the frequencies of an image are the
amount by which gray values change with distance.
High frequency components are characterized by large
changes in gray values over small distances;
Example of high frequency components are edges and
noise.
Low frequency components, on the other hand, are
parts of the image characterized by little change in the
gray values.
These may include backgrounds and skin textures.
Jorma Kekalainen

Digital Image Processing

732

Low and high pass filters

A high pass filter passes over the high frequency components,
and reduces or eliminates low frequency components.
A low pass filter passes over the low frequency components,
and reduces or eliminates high frequency components.
E.g. the 3*3 averaging filter is low pass filter, as it tends to blur
edges.

The filter
Jorma Kekalainen

is a high pass filter.

0.0564
0.7146
-4.9048
0.7146
0.0564

0.0468
0.3167
0.7146
0.3167
0.0468

0.0448
0.0468
0.0564
0.0468
0.0448

cf1=filter2(f1,c);
figure,imshow(cf1/100)

Jorma Kekalainen

Digital Image Processing

736

Example
Lets consider block of similar values pixels,

and apply the previous high pass filter

to the central four:

The resulting values are

close to zero, which is the
expected result of applying a
high pass filter to a low
frequency component.

Jorma
Kekalainen
Digital Image
Processing
737
Note:
How
to deal with negative outside
of the
displayable range we shall see next.

Exercise
Verify the calculation in the previous example:
apply the high pass filter

to the central four elements

Jorma Kekalainen

Digital Image Processing

738

Solution
filter2(filter,image,'valid') applies the mask
only to inside pixels.
>> A=[150 152 148 149;147 152 151 150;152 148 149 151;151 149 150 148]
A=
150 152 148 149
147 152 151 150
152 148 149 151
151 149 150 148
>> f2=[1 -2 1;-2 4 -2;1 -2 1]
f2 =
1 -2 1
-2 4 -2
1 -2 1
>> cf2=filter2(f2,A, 'valid')
cf2 =
11 6
Jorma Kekalainen
Digital Image Processing
-13 -5

739

Values outside the displayable

range
For image display we would like the gray
values of the pixels to lie between 0 and 255.
However, the result of applying a linear filter
may create values which lie outside this range.
There are ways of dealing with values outside
of this displayable range
Make negative values positive
Clip values
Scaling transformation.
Jorma Kekalainen

Digital Image Processing

740

Make negative values positive

This will certainly deal with negative values,
but not with values greater than 255.
Hence, this can only be used in specific
circumstances;
for example, when there are only a few negative
values, and when these values are themselves
close to zero.

Jorma Kekalainen

Digital Image Processing

741

Clip values

We apply the following thresholding type operation to the

gray values x produced by the filter to obtain a displayable
value y:

This will produce an image with all pixel values in the required
range, but is not suitable if there are many gray values outside
the 0 - 255 range; in particular, if the gray values are equally
spread over a larger range.
In such a case this operation will tend to destroy the results of
the filter.
Jorma Kekalainen

Digital Image Processing

742

Scaling transformation
Suppose the minimum gray value produced by the filter is gL
and the maximum value is gH.
Now we can transform all values in the range gL gH to the
range 0 - 255 by the linear transformation
Since the gradient of the line is
255/(gH g L),the equation of the
line is

Applying this transformation to all gray levels x produced by

the filter will result (after any necessary rounding) in an image
which can be displayed.
Jorma Kekalainen

Digital Image Processing

743

Example
Let's apply the high pass filter
to the cameraman image
c=imread('cameraman.tif');
f2=[1 -2 1;-2 4 -2;1 -2 1];
cf2=filter2(f2,c);
Now the maximum and minimum values of the matrix cf2 are
well over 255 (593) and well below 0 (-541).
So we use the mat2gray function and get the result
figure,imshow(mat2gray(cf2));
Jorma Kekalainen

Digital Image Processing

744

Example
f2=[1 -2 1;-2 4 -2;1 -2 1]
cf2=filter2(f2,c);
f2 =

1 -2 1

-2 4 -2

1 -2 1
cf2=filter2(f2,c);
figure,imshow(mat2gray(cf2));

Jorma Kekalainen

Digital Image Processing

745

Example
We can generally obtain a
better result by dividing
the result of the filtering
by a constant before
displaying it:
figure,imshow(cf2/60)

Note: High pass filters are often used for edge detection. This can be
Kekalainen
Image Processing
746
seenJorma
quite
clearly in the imageDigital
above.

749

Example
>> data=[20 20 20 10 10 10 10 10 10
>> f2=[-1 -1 0;-1 0 1;0 1 1]
20 20 20 20 20 20 20 20 10
f2 =
20 20 20 10 10 10 10 20 10
-1 -1 0
20 20 10 10 10 10 10 20 10
-1 0 1
20 10 10 10 10 10 10 20 10
0 1 1
10 10 10 10 20 10 10 20 10
10 10 10 10 10 10 10 10 10
20 10 20 20 10 10 10 20 20
dataf2v=filter2(f2,data,'valid')
20 10 10 20 10 10 20 10 20]
dataf2s=filter2(f2,data,'same')
data =
dataf2f=filter2(f2,data,'full')
20 20 20 10 10 10 10 10 10
20 20 20 20 20 20 20 20 10
20 20 20 10 10 10 10 20 10
20 20 10 10 10 10 10 20 10
20 10 10 10 10 10 10 20 10
10 10 10 10 20 10 10 20 10
10 10 10 10 10 10 10 10 10
20 10 20 20 10 10 10 20 20
Jorma Kekalainen
750
20 10 10 20 10 10 20 Digital
10 Image
20 Processing Note: filter2(filter,image,shape)

Example
data =
20
20
20
20
20
10
10
20
20

20
20
20
20
10
10
10
10
10

20
20
20
10
10
10
10
20
10

10
20
10
10
10
10
10
20
20

10
20
10
10
10
20
10
10
10

10
20
10
10
10
10
10
10
10

10
20
10
10
10
10
10
10
20

10
20
20
20
20
20
10
20
10

10
10
10
10
10
10
10
20
20

>> dataf2v=filter2(f2,data,'valid')
dataf2v =
0 -10 -10 0 0 10 0
-10 -30 -30 -20 -20 0 -10
-30 -30 -10 0 0 20 0
-30 -10 10 10 0 20 0
-10 0 10 0 -10 10 -10
10 20 10 -10 -10 10 10
0 20 0 -10 10 20 20

>> dataf2f=filter2(f2,data,'full')
dataf2f =
20 40 40 30 20 20 20 20 20 10 0
>> dataf2s=filter2(f2,data,'same')
40 60 40 30 30 40 40 40 30 0 -10
dataf2s =
40 40 0 -10 -10 0 0 10 0 -30 -20
60 40 30 30 40 40 40 30 0
40 40 -10 -30 -30 -20 -20 0 -10 -40 -20
40 0 -10 -10 0 0 10 0 -30
40 30 -30 -30 -10 0 0 20 0 -40 -20
40 -10 -30 -30 -20 -20 0 -10 -40
30 10 -30 -10 10 10 0 20 0 -40 -20
30 -30 -30 -10 0 0 20 0 -40
20 10 -10 0 10 0 -10 10 -10 -40 -20
10 -30 -10 10 10 0 20 0 -40
30 30 10 20 10 -10 -10 10 10 -20 -20
10 -10 0 10 0 -10 10 -10 -40
40 30 0 20 0 -10 10 20 20 -20 -30
30 10 20 10 -10 -10 10 10 -20
20 -10 -40 -20 -40 -40 -10 -20 -30 -50
-40
30
20 0 -10 10 20 20 -20 Digital Image Processing
Jorma0Kekalainen
751
0 -20 -30 -20 -30 -30 -20 -30 -30 -30 -20
-10 -40 -20 -40 -40 -10 -20 -30 -50

Example
>> whos
Name

Size

data
dataf2f
dataf2s
dataf2v
f2

9x9
11x11
9x9
7x7
3x3

Bytes Class
648
968
648
392
72

Attributes

double
double
double
double
double

The mat2gray function automatically

scales the matrix elements to
displayable values.
For any matrix it applies a linear
transformation to its elements, with the
lowest value mapping to 0.0, and
the highest value mapping to 1.0.
This means the output of mat2gray is
always of type double.
The function also requires that the input
type is double.
Digital Image Processing
752

figure,imshow(mat2gray(data))
title('imshow(mat2gray(data)')
figure,imshow(mat2gray(dataf2v))
title('imshow(mat2gray(dataf2v)')
figure,imshow(mat2gray(dataf2s))
title('imshow(mat2gray(dataf2s)')
figure,imshow(mat2gray(dataf2f))
title('imshow(mat2gray(dataf2f)')

Jorma Kekalainen

Example

Jorma Kekalainen

Digital Image Processing

753

Note
Matlab also has an imfilter function, which has the syntax
imfilter(x,f);
where x is an image matrix (of any type), and f is a filter.
It differs from filter2 in the different parameters it takes (read its help file),
and in that the output is always of the same class as the original image.
B = imfilter(A,h) filters the multidimensional array A with the
multidimensional filter h.
The array A can be logical or a nonsparse numeric array of any class and
dimension.
The result B has the same size and class as A.
imfilter computes each element of the output, B, using doubleprecision
floating point.
If A is an integer or logical array, imfilter truncates output elements that
exceed the range of the given type, and rounds fractional values.
Jorma Kekalainen

Digital Image Processing

754

Exercise
The array below represents a small grayscale image. Compute
the images that result when the image is convolved with each
of the masks (a) to (f) shown
(a)
0 -1 -1
1 0 -1
110
(b)
-1 -1 -1
222
-1 -1 -1
(c)
-1 2 -1
-1 2 -1
-1 2 -1
(d)
-1 -1 -1
-1 8 -1
Jorma Kekalainen
-1 -1 -1

(e)
111
111
111
(f)
-1 0 1
-1 0 1
-1 0 1
(g)
0 -1 0
-1 4 -1
0 -1 0

20 20 20 10 10 10 10 10 10
20 20 20 20 20 20 20 20 10
20 20 20 10 10 10 10 20 10
20 20 10 10 10 10 10 20 10
20 10 10 10 10 10 10 20 10
10 10 10 10 20 10 10 20 10
10 10 10 10 10 10 10 10 10
20 10 20 20 10 10 10 20 20
20 10 10 20 10 10 20 10 20

Digital Image Processing

755

Exercise
Obtain a grayscale image of a mandrill with
the following commands:
>> load('mandrill.mat');
>> m=im2uint8(ind2gray(X,map));
Apply all the filters (a) - (g) listed in the
previous exercise to this image.
Can you now see what each filter does?

Jorma Kekalainen

Digital Image Processing

756

Exercise
Apply larger and larger averaging filters to the
image of the mandrill.
What is the smallest sized filter for which the
whiskers cannot be seen?

produces to a flatter curve, and a small value leads to a
sharper curve.

Jorma Kekalainen

Digital Image Processing

760

Two dimensional Gaussian

function
A two dimensional Gaussian function is given by

The command fspecial('gaussian') produces a discrete version of this

779

Example

c=imread('cameraman.tif');
figure,imshow(c),title('Original image')
cmed=medfilt2(c);
figure,imshow(cmed)
title('Filtered by median filter')

Jorma Kekalainen

Median filtering effects here are

infinitesimal. However, median
filtering seems almost tailor-made for
removal of some type of noise.

Digital Image Processing

780

Rank-order filters
Median filtering is a special case of a more general
process called rank-order filtering.
Rather than take the median of a set, we order the set
and take the n-th value, for some predetermined value
of n.
Matlab implements rank-order filtering with the
ordfilt2 function.
In fact the procedure for medfilt2 is really just a
wrapper for a procedure which calls ordfilt2.
There is only one reason for using rank-order filtering
instead of median filtering, and that is that it allows us
to choose the median of non-rectangular masks.
Jorma Kekalainen

Digital Image Processing

781

Rank-order filters
For example, if we decided to use as a mask a (3*3)
cross shape:

then the median would be the third of these five

values when sorted.
The command to do this is
ordfilt2(c,3,[0 1 0;1 1 1;0 1 0]);
Jorma Kekalainen

Digital Image Processing

782

Example

c=imread('cameraman.tif');
figure,imshow(c),title('Original image')
ccross=ordfilt2(c,3,[0 1 0;1 1 1;0 1 0]);
figure,imshow(ccross)
title('Filtered by cross-shaped median filter')

Jorma Kekalainen

Digital Image Processing

The same observations as in

the previous case of the
rectangular 3-by-3 mask.

783

Rank-order filters
In general, the second argument of ordfilt2
gives the value of the ordered set to take, and
the third element gives the domain; the nonzero values of which specify the mask, e.g.,
ccross=ordfilt2(c,3,[0 1 0;1 1 1;0 1 0]);
If we wish to use a cross with size and width 5
(so containing nine elements), we can use:
>> ordfilt2(c,5,[0 0 1 0 0;0 0 1 0 0;1 1 1 1 1;0 0 1 0 0;0 0 1 0 0]);

Jorma Kekalainen

Digital Image Processing

784

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

Noises and Noise Cleaning
Methods

Definition of noise
We may define noise to be any degradation in the image
signal, caused by external disturbance.
If an image is being sent electronically from one place to
another, via satellite or wireless transmission, or through
cable, we may expect errors to occur in the image signal.
These errors will appear on the image output in different ways
depending on the type of disturbance in the signal.
Usually we know what type of errors to expect, and hence the
type of noise on the image; hence we can choose the most
appropriate method for reducing the effects.
Cleaning an image corrupted by noise is thus an important
area of image restoration.
Jorma Kekalainen

Digital Image Processing

787

Types of noise
Now we will investigate some of the standard
noise forms, and the different methods of
eliminating or reducing their effects on the
image.
Impulse noise (also called shot noise, binary noise
or salt and pepper noise)
Gaussian noise
Speckle noise
Periodic noise
Note: Other theoretical noise models are Poisson, Rayleigh, Gamma and Exponential
Jorma Kekalainen
Digital Image Processing
788
noises.
They can be used for approximating
skewed histograms.

Impulse noise
This degradation can be caused by sharp, sudden
disturbances in the image signal; its appearance is
randomly scattered white or black (or both) pixels
over the image.
To demonstrate its appearance, we will use the
familiar gray-scale image:
c=imread('cameraman.tif');
figure, imshow('cameraman.tif');
title('Clean image')

Jorma Kekalainen

Digital Image Processing

789

Imnoise function
To add noise, we use the Matlab function imnoise, which
takes a number of different parameters.
To add salt and pepper noise:
>> c_sp=imnoise(c,'salt & pepper');
The amount of noise added is 10% (= default value) to add
more or less noise we include an optional parameter, being a
value between 0 and 1 indicating the fraction of pixels to be
corrupted.
E.g.,
>> imnoise(c,'salt & pepper',0.2);
would produce an image with 20% of its pixels corrupted by
salt and pepper noise.
Jorma Kekalainen

Digital Image Processing

790

Impulse noise
c=imread('cameraman.tif');
c_sp=imnoise(c,'salt & pepper');
figure, imshow(c_sp)
title('Image with 10% of pixels corrupted by binary noise')
c_sp=imnoise(c,'salt & pepper',0.2);
figure, imshow(c_sp)
title('Image with 20% of pixels corrupted by binary noise')

Jorma Kekalainen

Digital Image Processing

Image corrupted by Gaussian and

speckle noise
Gaussian noise

Speckle noise

Note: Although Gaussian noise and speckle noise appear superficially similar, they are
formed by two totally different methods, and, thus require different approaches for
Kekalainen
Digital Image Processing
798
theirJorma
removal.

Periodic noise
If the image signal is subject to a periodic, rather than a
random disturbance, we might obtain an image
corrupted by periodic noise.
Periodic noise may occur if the imaging equipment (the
acquisition or networking hardware) is subject to
electronic disturbance of a repeating nature, such
noise may be caused by an electric motor.
The effect of this kind noise is bars over the image.
The function imnoise does not have a periodic option,
but we can create our own periodic noise by adding a
periodic matrix (using a trigonometric function) to our
image.
Jorma Kekalainen

Digital Image Processing

799

Periodic noise

Jorma Kekalainen

Digital Image Processing

800

Periodic noise
We can create periodic noise by overlaying an image with a
trigonometric function:
c=imread('cameraman.tif');
s=size(c);
[x,y]=meshgrid(1:s(1),1:s(2));
pn=sin(x+y)+1;
c_pn=(im2double(c)+pn/2)/2;
figure,imshow(c_pn)
title('Image with periodic noise pn=sin(x+y)+1')
The first lines simply create a suitable sine matrix function, and
adjusts its output to be in the range 0 - 2.
The c_pn line first adjusts the cameraman image to be double class;
adds the scaled sine function to it, and divides by 2 to produce a
matrix of type double with all elements in the range 0.0 1.0.
This can be viewed directly with imshow.
Jorma Kekalainen

Digital Image Processing

801

Example
c=imread('cameraman.tif');
s=size(c);
[x,y]=meshgrid(1:s(1),1:s(2));
pn=sin(x+y)+1;
c_pn=(im2double(c)+pn/2)/2;
figure,imshow(c_pn)
title('Image with periodic noise pn=sin(x+y)+1')

Jorma Kekalainen

Digital Image Processing

802

c=imread('cameraman.tif');
c_sp=imnoise(c,'salt & pepper');
figure, imshow(c_sp)
title('Image with 10% of pixels corrupted by binary noise')
a3=fspecial('average');
c_sp_a3=filter2(a3,c_sp);
figure, imshow(uint8(c_sp_a3))
title('Image with 10% of pixels corrupted by binary noise filtered
by 3*3 averaging filter')
a5=fspecial('average',[5,5]);
c_sp_a5=filter2(a5,c_sp);
figure, imshow(uint8(c_sp_a5))
title('Image with 10% of pixels corrupted by binary noise filtered
by 5*5 averaging filter')

Jorma Kekalainen

Digital Image Processing

807

Effect of low pass filtering

Given that pixels corrupted by salt
and pepper noise are high
frequency components of an image,
we should expect a low-pass filter
should reduce them.
So we might try filtering with an
average filter:
a3=fspecial('average');
c_sp_a3=filter2(a3,c_sp);
The result is not noticeably better
than the noisy image.
The noise is not so much removed
as smeared over the image.
Note: Impulse noise = binary noise = salt and pepper
Kekalainen
Digital Image Processing
noiseJorma
= shot
noise.

808

Effect of low pass filtering

The effect is even more pronounced if
we use a larger averaging filter:
>> a7=fspecial('average',[7,7]);
>> c_sp_a7=filter2(a7,c_sp);
>> a11=fspecial('average',[11,11]);
>> c_sp_a11=filter2(a11,c_sp);
(5*5) averaging

Jorma Kekalainen

Digital Image Processing

809

Example

a7=fspecial('average',[7,7]);
c_sp_a7=filter2(a7,c_sp);
figure, imshow(uint8(c_sp_a7))
title('Image with 10% of pixels corrupted by binary noise
filtered by 7*7 averaging filter')
a11=fspecial('average',[11,11]);
c_sp_a11=filter2(a11,c_sp);
figure, imshow(uint8(c_sp_a11))
title('Image with 10% of pixels corrupted by binary noise
filtered by 11*11 averaging filter')

Jorma Kekalainen

Digital Image Processing

810

Median filtering
Median filtering seems almost tailor-made for
removal of salt and pepper noise.
Recall that the median of a set is the middle
value when they are sorted.
If there are an even number of values, the
median is the mean of the middle two.
A median filter is an example of a non-linear
spatial filter.
Jorma Kekalainen

Digital Image Processing

811

Example
Using a 3*3 mask, the output value is the median of
the values in the mask.

We see that very large or very small values - noisy

values - will end up at the top or bottom of the
sorted list.
Thus the median will in general replace a noisy value
with one closer to its surroundings.
Jorma Kekalainen

Digital Image Processing

812

Median filtering
In Matlab, median filtering is
implemented by the medfilt2 function:
>> c_sp_m3=medfilt2(c_sp);
The result is a vast improvement on using
averaging filters.

Jorma Kekalainen

Digital Image Processing

Cleaning salt and pepper

noise with a median filter

813

Example

% Cleaning binary (salt & pepper noise) noise with median filter
C=imread('cameraman.tif');
figure,imshow(C)
title('Clean image')
Csp=imnoise(C,'salt & pepper');%The amount of noise added to 10%
figure, imshow(Csp)
title('Image with 10% of its pixels corrupted by binary noise')
Cspm3=medfilt2(Csp);% class: uint8
figure, imshow(Cspm3)
title('Image filtered by 3*3 median filter')
xlabel('Image with 10% of its pixels corrupted by binary noise')

Jorma Kekalainen

Digital Image Processing

814

Using median filter on more noise

This would produce an

If we corrupt more pixels with noise:

image with 20% of its
>> c_sp2=imnoise(c,'salt & pepper',0.2); pixels corrupted by noise.
then medfilt2 still does a remarkably good job.
20% salt & pepper noise

Jorma Kekalainen

Digital Image Processing

After (3*3) median filtering

815

M av

1 n
1 n
1 n
= (M + N i ) = M + N i
n i =1
n i =1
n i =1
1 n
= M + Ni
n i =1

Image Processing
M av n
MDigital
mean = M

Jorma Kekalainen

821

Image averaging
Since Ni is normally distributed with mean 0, it
can be readily shown that the average of all
the Ni s will be close to zero; actually the
greater the number of Ni s the closer to zero
the average is .
Thus Mav Mmean = M and the approximation
is closer for larger number of images.

Jorma Kekalainen

Digital Image Processing

822

Example
We can demonstrate this with the familiar cameraman image.
We first need to create different versions with Gaussian noise,
and then take the average of them.
We shall create 10 versions.
One way is to create an empty three-dimensional array of
depth 10, and fill each level with a noisy image:
>> s=size(c);
>> c_ga10=zeros(s(1),s(2),10);
>> for i=1:10 c_ga10(:,:,i)=imnoise(c,'gaussian'); end

Jorma Kekalainen

Digital Image Processing

823

Example

Jorma Kekalainen

Digital Image Processing

824

829

Example
Suppose we take the (3*3) and (5*5)
averaging filters, and apply them to the noisy
image c_ga.
>> a3=fspecial('average');
>> a5=fspecial('average',[5,5]);
>> cg3=filter2(a3,c_ga);
>> cg5=filter2(a5,c_ga);

Jorma Kekalainen

Digital Image Processing

830

Averaging filtering to remove

Gaussian noise
The results are not really particularly pleasing;
although there has been some noise
reduction, the smeary nature of the resulting
images is unattractive.
Image with additive Gaussian
noise with m=0,v=0.01

Jorma Kekalainen

(3*3) averaging

Digital Image Processing

(5*5) averaging

831

Least squares filtering

Given a degraded image M of some original image M and a
restored version R, what measure can we use to say whether
our restoration has done a good job?
Clearly we would like R to be as close as possible to the
correct image M.
One way of measuring the closeness of R to M is by adding
the squares of all differences:

i, j

ri , j )

i, j

where the sum is taken over all pixels of R and M (which we

assume to be of the same size).
This sum can be taken as a measure of the closeness of R to
M.
Jorma Kekalainen

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

841

Example
% Using smaller gaussian noise
(m=0,v=0.005)
Cga2=imnoise(C,'gaussian',0,0.005);
figure, imshow(Cga2)
title('Image with additive Gaussian
noise with m=0,v=0.005')
Cga2w=wiener2(Cga2,[7,7]);
figure,imshow(Cga2w)
title('Image filtered by 7*7 Wiener
filter')
xlabel('Image with additive Gaussian
noise with m=0,v=0.005')

Jorma Kekalainen

Digital Image Processing

842

The image and its appearance

after Wiener filtering
The result is a clear improvement over the
original noisy image.

Jorma Kekalainen

Digital Image Processing

843

Example: whos command

>> whos
Name

Size

C
256x256
Cga
256x256
Cga10
256x256x10
Cga100
256x256x100
Cga100av
256x256
Cga10av
256x256
Cga2
256x256
Cga2w
256x256
Cga3
256x256
Cga5
256x256
Cgaw3
256x256
Cgaw5
256x256
Cgaw7
256x256
Cgaw9
256x256
Csp
256x256
Csp2
256x256

Jorma Kekalainen

>> fspm3=medfilt2(fsp);% class: uint8

figure, imshow(fspm3)
title('Image filtered by 3*3 median filter')
xlabel('Image with 10% of its pixels corrupted
by binary noise')

Digital Image Processing

848

>> % Cleaning binary (salt & pepper noise)

noise with median filter
% 20% of pixel corrupted by noise
fsp20=imnoise(f,'salt & pepper',0.2);% 20% of
pixel corrupted by noise
fsp20m3=medfilt2(fsp20);% class: uint8
figure, imshow(fsp20)
title('Image with 20% of its pixels corrupted by figure, imshow(fsp20m3)
title('Image filtered by 3*3 median filter')
binary noise')
xlabel('Image with 20% of its pixels corrupted
by binary noise')

(b)

Jorma Kekalainen

Digital Image Processing

849

(c)% Using 5*5 median filter

% Using 3*3 median filter twice

fsp20m3twice=medfilt2(fsp20m3);
figure, imshow(fsp20m3twice)
title('Image filtered twice by 3*3 median filter')
xlabel('Image with 20% of its pixels corrupted
by binary noise')

Jorma Kekalainen

fsp20m5=medfilt2(fsp20,[5,5]);
figure, imshow(fsp20m5)
title('Image filtered by 5*5 median filter')
xlabel('Image with 20% of its pixels corrupted
by binary noise')

Digital Image Processing

850

Note
f=imread('flowers.tif');
figure,imshow(f)
title('Original image ''flowers.tif''')
% 20% of pixel corrupted by noise
fsp20=imnoise(f,'salt & pepper',0.2);% 20% of pixel corrupted by noise
figure, imshow(fsp20)
title('Image with 20% of its pixels corrupted by binary noise')
fsp20m3r=medfilt2(fsp20(:,:,1));% Red component
fsp20m3g=medfilt2(fsp20(:,:,2));
fsp20m3b=medfilt2(fsp20(:,:,1));
fsp20m3=cat(3,fsp20m3r, fsp20m3g, fsp20m3b);% all components back into a
single 3D array for use with imshow
figure, imshow(fsp20m3)
title('Image filtered by 3*3 median filter')
Jorma Kekalainen
Digital Image Processing
851
xlabel('Image with 20% of its pixels
corrupted by binary noise')

Note
f=imread('flowers.tif');
figure,imshow(f)
title('Original image ''flowers.tif''')
% 20% of pixel corrupted by noise
fsp20=imnoise(f,'salt & pepper',0.2);% 20% of pixel corrupted by noise
figure, imshow(fsp20)
title('Image with 20% of its pixels corrupted by binary noise')
a3=fspecial('average');
fsp20a3r=filter2(a3,fsp20(:,:,1));%Red component
fsp20a3g=filter2(a3,fsp20(:,:,2));
fsp20a3b=filter2(a3,fsp20(:,:,3));
fsp20a3=cat(3,fsp20a3r, fsp20a3g, fsp20a3b);% all components back into a
single 3D array for use with imshow
Jorma Kekalainen

Digital Image Processing

852

Exercise
Add Gaussian noise to the grayscale flowers image
with the following parameters:
(a) mean 0, variance 0.01 (default)
(b) mean 0, variance 0.02
(c) mean 0, variance 0.05
(d) mean 0, variance 0.1
In each case, attempt to remove the noise with
average filtering and with Wiener filtering.
Can you produce satisfactory results with the last
two noisy images?
Jorma Kekalainen

Digital Image Processing

856

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

Edges and Differences

Introduction
Edges contain the most useful information in an image.
In an image, an edge is a curve that follows a path of rapid
change in image intensity.
Edges are often associated with the boundaries of objects in a
scene.
We may use edges
to measure the size of objects in an image;
to isolate particular objects from their background;
to recognize or classify objects.

Edge detection is used to identify the edges in an image.

Jorma Kekalainen

Digital Image Processing

859

Edge-finding algorithms
There are a large number of edge-finding
algorithms, and we shall look at some of the most
straightforward of them.
Here, we shall show how to create edge images
using basic filtering methods, and discuss the
Matlab edge function.
The general Matlab command for finding edges is
edge(image,'method',parameters. . . )
where the available parameters depend on the
method used.
Jorma Kekalainen

Digital Image Processing

860

Definition of edge
An edge may be loosely defined as a line of pixels showing an
observable difference.
For example, consider the two blocks of pixels

In the right hand block, there is a clear difference between

the gray values in the second and third columns.
This would be easily discernible in an image.
The human eye can pick out gray differences of this
magnitude with relative ease.
Our aim is to develop methods which will enable us to pick
out the edges of an image.
Jorma Kekalainen
Digital Image Processing

861

Edges
Suppose we follow the gray values of pixels in a line
crossing an edge, as illustrated below.

If we consider the gray values along this line, and

plot their values, we will have something similar to
that a step edge and a ramp edge shown in the
following figure.
Jorma Kekalainen

Digital Image Processing

862

Gray values across edges

A step edge

A ramp edge

A line

Jorma Kekalainen

A roof

Digital Image Processing

863

Differences
If we now plot the differences between each gray
value and its predecessor from the ramp edge, we
would obtain a graph
Differences of the edge function

To see where this graph comes from, suppose that

the values of the ramp edge in the previous figure
are, from left to right:
Jorma Kekalainen

Digital Image Processing

864

Differences and edges

If we form the differences, by subtracting each value
from its successor, we obtain:
which are plotted above.
It appears that the difference tends to enhance
edges, and reduce other components.
So if we could difference the image, we would obtain
the effect we want.

Jorma Kekalainen

Digital Image Processing

865

Different differences
We can define the difference in three separate
ways:
the forward difference:
the backward difference:
the central difference:
Jorma Kekalainen

Digital Image Processing

866

Generalized differences
An image is a function of two variables, so we
must generalize those definitions to include
both the x and y values

where the subscripts in each case indicate the

direction of the difference.

Jorma Kekalainen

Digital Image Processing

867

Finding vertical edges

To see how we might use x

to determine edges in the x direction, consider the function

values around a point (x,y) and using the convention that x
values increase to the right, y values to the bottom:

To find the filter which returns the value x , we just compare

the coefficients of the function values in x with their position
in the array:
Jorma Kekalainen

Digital Image Processing

868

Finding vertical edges

This filter

thus will find vertical edges in an image.

However, the edges in the result can be a bit irregular or
jumpy; this can be overcome by smoothing the result in the
opposite direction by using the filter

Jorma Kekalainen

Digital Image Processing

869

Prewitt filter Px
Both of the previous filters can be applied at once
convolving them.
As a result of convolution we get a combined filter:

>> A=[0 0 0;-1 0 1;0 0 0]

A=
0 0 0
-1 0 1
0 0 0
>> B=[0 1 0;0 1 0;0 1 0]
B=
0 1 0
0 1 0
0 Jorma
1 Kekalainen
0

Digital Image Processing

>> Px=conv2(A,B,'same')
Px =
-1 0 1
-1 0 1
870
-1 0 1

Finding horizontal edges

To find the filter which returns the value y ,
we just compare the coefficients of the function
values in y with their position in the previous
function array.

Jorma Kekalainen

Digital Image Processing

871

Finding horizontal edges

To see how we might use y

to determine edges in the y direction, consider the function

values around a point (x,y) and using the convention that x
values increase to the right, y values to the bottom:

To find the filter which returns the value y , we just compare

the coefficients of the function values in y with their position
in the array:
Jorma Kekalainen

0 -1 0
0 0 0
0 1 0
Digital Image Processing

872

Finding horizontal edges

This filter
0 -1
0 0
0 1

0
0
0

thus will find horizontal edges in an image.

However, the edges in the result can be a bit
irregular or jumpy; this can be overcome by
smoothing the result in the opposite direction by
using the filter
0
1
0
Jorma Kekalainen

0
1
0

Digital Image Processing

873

Prewitt filter Py
Both of the previous filters can be applied at
once convolving them.
As a result of convolution we get a combined
filter:
A =
0 -1
0 0
0 1
B' =
0
1
0

0
0
0

0 0
1 1
Jorma Kekalainen
0 0

Digital Image Processing

>> Py=conv2(A',B','same')
Py =
-1 -1 -1
0 0 0
874
1 1 1

Note
>> A=[0 0 0;-1 0 1;0 0 0]
A=
0 0 0
-1 0 1
0 0 0
>> B=[0 1 0;0 1 0;0 1 0]
B=
0 1 0
0 1 0
0 0
0 1 0
1 1
>> Px=conv2(A,B,'same')
0 0
Px =
-1 0 1
0 0
-1 0 1
1 1
-1 0 1
0 0
>> Py=conv2(A',B','same')
Py =
-1 -1 -1
0 0 0
Digital Image Processing 1
1 1

A=[0 0 0;-1 0 1;0 0 0]

B=[0 1 0;0 1 0;0 1 0]
>> A'
ans =
0 -1
0 0
0 1
0 -1
0 0
0 1

0
0
0
0
0
0

Jorma Kekalainen

>> B'
ans =
0
1
0
0
1
0

875

Prewitt filters
Filter for finding vertical edges

and its companion for finding horizontal edges

are the Prewitt filters for edge detection.

Jorma Kekalainen

Digital Image Processing

876

Note
If we change increasing directions in the function array just
the opposite we correspondingly get matrices flilr(Px) and
flipud(Py)
1
1
1

Py =
-1 -1 -1
0 0 0
1 1 1

>> fliplr(Px)
ans =
1 0 -1
1 0 -1
1 0 -1

>>flipud(Py)
ans =
1 1 1
0 0 0
-1 -1 -1

Px =
-1
-1
-1

0
0
0

Flipud flip matrix up to down and Fliplr flip matrix left to right.
Jorma Kekalainen

Digital Image Processing

877

Note
h = fspecial('prewitt') returns the 3-by-3 filter
h=[ 1 1 1; 0 0 0; -1 -1 -1 ]
that emphasizes horizontal edges by approximating a
>> fspecial('prewitt')
vertical gradient.
ans =
1 1 1
0 0 0
-1 -1 -1

If you need to emphasize vertical edges, transpose

the filter h'.
To find vertical edges, or for x-derivatives, use h'.
Jorma Kekalainen

Digital Image Processing

Example

%Edge detection
ic=imread('ic.tif');
figure,imshow(ic)
title('Original image ic.tif')
%
px=[-1 0 1;-1 0 1;-1 0 1];
icx=filter2(px,ic);
figure,imshow(icx/255)
title('Vertical edges in ic.tif')
%
py=px';
icy=filter2(py,ic);
figure,imshow(icy/255)
title('Horizontal edges in ic.tif')
%
% All the edges with:
edge_p=sqrt(icx.^2+icy.^2);
figure,imshow(edge_p/255)
title('All edges in the gray-scale image ic.tif')

885

Example
We can create a figure containing all the edges with:
edge_p=sqrt(icx.^2+icy.^2);
figure,imshow(edge_p/255)
This is a gray-scale image; a binary image containing
edges only can be produced by thresholding.
edge_t=im2bw(edge_p/255,0.3);

>> whos('edge_p','edge_t')
Name
Size
Bytes Class Attributes
edge_p 256x256
524288 double
Jorma Kekalainen
edge_t 256x256
65536 logicalDigital Image Processing

886

Example

% All the edges with:

edge_p=sqrt(icx.^2+icy.^2);
figure,imshow(edge_p/255)
title('All edges in the grayscale image ic.tif')
edge_t=im2bw(edge_p/255,0.
3);
figure,imshow(edge_t)
title('Binary image
thresholded from ic.tif')
xlabel('Threshold level=0.3')

Jorma Kekalainen

Digital Image Processing

887

Note
We can obtain edges by the Prewitt filters directly by
using the command
>> edge_p=edge(ic,'prewitt');
and the edge function takes care of all the filtering, and
of choosing a suitable threshold level.
The result is shown in the following figure.
Note that the previous figure and the next figure seem
different to each other.
This is because the edge function does some extra
processing besides taking the square root of the sum of
the squares of the filters.
Jorma Kekalainen
Image Processing
Note:
See the help text of theDigital
edge
function for more information.888

Example
%We can obtain edges by the Prewitt filters directly by using the command
edge_p=edge(ic,'prewitt');
figure, imshow(edge_p)
title('Edges of ic.tif directly by using the Prewitt filters')
xlabel('Using command edge(ic,''prewitt'');')

Jorma Kekalainen

Digital Image Processing

889

Sobel filters
The Sobel filters are similar to the Prewitt filters, in that they
apply a smoothing filter in the opposite direction to the
central difference filter.
Now the smoothing takes the form which gives slightly more
prominence to the central pixel
0
1
0

Jorma Kekalainen

0
2
0

0
1
0

0
0
0

Digital Image Processing

1
2
1

0
0
0

890

Sobel filter Sx
As we previously notice in Prewitts case this filter

will find vertical edges in an image.

However, the edges in the result can be a bit irregular or
jumpy; this can be overcome by smoothing the result in the
opposite direction by using the filter
0
0
0
Jorma Kekalainen

1
2
1

0
0
0

In the Sobel filters, the

smoothing gives slightly
more prominence to the
central pixel.

Digital Image Processing

891

Sobel filter Sx
Both of the previous filters can be applied at once
convolving them.
As a result of convolution we get a combined filter:
Sx =

Sx will find vertical edges in an image.

>> A=[0 0 0;-1 0 1;0 0 0]
A=
0 0 0
-1 0 1
Jorma Kekalainen
0 0 0

>> B'=[0 1 0;0 2 0;0 1 0]

B' =
0 1 0
0 2 0Digital Image Processing
0 1 0

>> Sx=conv2(A,B','same')
Sx =
-1 0 1
-2 0 2
892
-1 0 1

Sobel filter Sy
As we previously notice in Prewitts case this filter
0 -1
0 0
0 1

0
0
0

will find vertical edges in an image.

However, the edges in the result can be a bit irregular or
jumpy; this can be overcome by smoothing the result in the
opposite direction by using the filter
0
1
0
Jorma Kekalainen

0
2
0

0
1
0

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

h=
1 2 1
0 0 0
-1 -2 -1

h=h' =
1 0 -1
2 0 -2
897
1 0 -1

Example
edge_r=edge(ic,'roberts');
figure,imshow(edge_r)

edge_s=edge(ic,'sobel');
figure,imshow(edge_s)

Of the three filters, the Sobel filters are probably the best; they provide good edges,
and they
perform reasonably well in Digital
the presence
of noise.
Jorma Kekalainen
Image Processing
898

Example

%Roberts and Sobel filters

edge_r=edge(ic,'roberts');
figure,imshow(edge_r)
title('Edge finding of ic.tif using Roberts cross-gradient filters')
xlabel('Using command edge(ic,''proberts'');')
%
%and
%
edge_s=edge(ic,'sobel');
figure,imshow(edge_s)
title('Edge finding of ic.tif using Sobel filters')
xlabel('Using command edge(ic,''sobel'');')

Jorma Kekalainen

Digital Image Processing

899

Example

Jorma Kekalainen

Digital Image Processing

900

Digital Image Processing

Second Differences

905

Central second differences

To calculate a central second difference, take the backward
difference of a forward difference:

x2 = x x f ( x, y ) = x [ f ( x + 1, y ) f (x, y )]

= [ f ( x + 1, y ) f ( x, y )] [ f ( x, y ) f ( x 1, y )]
= f ( x + 1, y ) 2 f ( x, y ) + f ( x 1, y )
The corresponding filter for second differences x2 in the x
direction can be implemented by the filter:

Jorma Kekalainen

Digital Image Processing

906

Central second differences

To calculate a central second difference, take the backward
difference of a forward difference:

y2 = y y f (x, y ) = y [ f (x, y + 1) f (x, y )]

= [ f ( x, y + 1) f ( x, y )] [ f ( x, y ) f ( x, y 1)]
= f ( x, y + 1) 2 f ( x, y ) + f ( x, y 1)
The corresponding filter for second differences y2 in the y
direction can be implemented by the filter

Jorma Kekalainen

Digital Image Processing

907

Laplacian filter
The corresponding filters for second differences x2 in the x
direction and for second differences y2 in the y direction are

The sum of these two is written

and implemented by the filter, which is known as a discrete
Laplacian.

Jorma Kekalainen

Digital Image Processing

908

Example
Take the pixel values describing a ramp edge

If we form the differences, by subtracting each value from its

successor, we obtain:

Differences of the edge function

Jorma Kekalainen

Digital Image Processing

909

Laplacian
We see that the Laplacian (after taking an
absolute value, or squaring) gives double
edges.
It is also extremely sensitive to noise.
However, the Laplacian does have the
advantage of detecting edges in all directions
equally well.

Jorma Kekalainen

Digital Image Processing

910

Example
Suppose we enter the Matlab commands:
l=fspecial('laplacian',0);% Basic laplacian mask
ic_l=filter2(l,ic);
figure,imshow(mat2gray(ic_l))
The result of which is shown in the following
figure.

Jorma Kekalainen

Digital Image Processing

911

Example
%
ic=imread('ic.tif');
figure,imshow(ic)
title('Original image ic.tif')
%
L=fspecial('laplacian',0);
ic_L=filter2(L,ic);
figure,imshow((ic_L))
title('Filtering by discrete Laplacian')
xlabel('fspecial(''laplacian'',0)')
%

Jorma Kekalainen

Digital Image Processing

912

Result after filtering with a

discrete laplacian
Although the result is
adequate, it is very
messy when compared
to the results of the
Prewitt and Sobel
methods discussed
earlier.

Jorma Kekalainen

Digital Image Processing

913

Other Laplacian masks

Other Laplacian masks can be used, e.g.,:

In Matlab, Laplacians of all sorts can be generated using the

fspecial function, in the form fspecial('laplacian',alpha) which
produces the Laplacian

If the parameter alpha (which is optional) is omitted, it is

assumed to be 0.2.
Jorma Kekalainen
Digital Image Processing
Note:
The value 0 gives the Laplacian
developed earlier.

914

fspecial('laplacian', alpha)
fspecial('laplacian', alpha) returns a 3-by-3
filter approximating the shape of the twodimensional Laplacian operator.
The parameter alpha controls the shape of the
Laplacian and must be in the range 0.0 to 1.0.
The default value for alpha is 0.2.

Jorma Kekalainen

Digital Image Processing

915

fspecial('laplacian', alpha)

>> fspecial('laplacian',1.0)

ans =

0.5000
0 0.5000

0 -2.0000
0

Jorma
Kekalainen

0.3333 0.3333 0.3333

>> fspecial('laplacian',0.4)
ans =
0.2857 0.4286 0.2857
0.4286 -2.8571 0.4286
0.2857 0.4286 0.2857
>> fspecial('laplacian',0.3)
ans =
0.2308 0.5385 0.2308
0.5385 -3.0769 0.5385
0.2308 0.5385 0.2308
>> fspecial('laplacian',0.2)
ans =
0.1667 0.6667 0.1667
0.6667 -3.3333 0.6667
0.1667 0.6667 0.1667

>> fspecial('laplacian',0.1)
ans =
0.0909 0.8182 0.0909
0.8182 -3.6364 0.8182
0.0909 0.8182 0.0909

>> fspecial('laplacian',0.0)
ans =
0 1 0
1 -4 1
0 1 0

Digital Image Processing

default value
alpha=0.2

916

Zero crossings
A more appropriate use for the Laplacian is to find the
position of edges by locating zero crossings.
If we look at figure below, we see that the position of the
edge is given by the place where the value of the filter takes
on a zero value.
In general, these are places where the result of the filter
changes sign.

919

Example
To give an indication of the
way zero-crossings work,
look at the edge plots and
their second differences.
In each case the zerocrossing is circled.
The important point is to
note that across any edge
there can be only one zerocrossing.
Thus an image formed from
zero-crossings has the
potential to be very handy.
Jorma Kekalainen

Edges and second differences

Digital Image Processing

920

Locating zero crossings in an

image
Consider a small image (a), and the result after
filtering with a Laplacian mask in figure (b).
Zero crossings are colored with red.
(a) Small image
50
50
50
50
50
50
50
50
50
50

50
50
50
50
50
50
50
50
50
50

50
50
200
200
200
200
50
50
50
50

(b) After laplace filtering

-100 -50 -50 -50 -50 -50 -50 -50 -50 -100
50 50 50 50 50 50 50
-50 0 150 150 150 150 150 150 0 -50
50 50 50 50 50 50 50
200 200 200 200 200 50 50 -50 150 -300 -150 -150 -150 -150 -300 150 -50
200 200 200 200 200 50 50 -50 150 -150 0 0 0 0 -150 150 -50
200 200 200 200 200 50 50 -50 150 -150 0 0 0 0 -150 150 -50
200 200 200 200 200 50 50 -50 150 -300 -150 0 0 0 -150 150 -50
-50 0 150 300 -150 0 0 -150 150 -50
50 200 200 200 200 50 50
-50 0 0 150 -300 -150 -150 -300 150 -50
50 200 200 200 200 50 50
-50 0 0 0 150 150 150 150 0 -50
50 50 50 50 50 50 50
-100 -50 -50 -50 -50 -50 -50 -50 -50 -100
50 50 50 50 50 50 50

923

Marr-Hildreth method
This method was designed to provide an edge detection method to
be as close as possible to biological vision.
The first two steps (smooting and convolving) can be combined
into one, to produce a Laplacian of Gaussian (LoG) or 'log' filter.
These filters can be created with the fspecial function.
If no extra parameters are provided to the zerocross edge option,
then the filter is chosen to be the LoG filter found by
fspecial('log',13,2)
This means that the following command:
edge(ic,'log');
produces exactly the same result as the commands:
log=fspecial('log',13,2);
edge(ic,'zerocross',log);
Jorma Kekalainen

Digital Image Processing

924

fspecial('log', hsize, sigma)

h = fspecial('log', hsize, sigma) returns a
rotationally symmetric Laplacian of Gaussian
(LoG) filter of size hsize with standard
deviation sigma (positive).
hsize can be a vector specifying the number of
rows and columns in h, or it can be a scalar, in
which case h is a square matrix.
The default value for hsize is [5 5] and 0.5 for
sigma.
Jorma Kekalainen

Digital Image Processing

925

Marr-Hildreth method
In fact the LoG and
zerocross options
implement the same edge
finding method; the
difference being that the
zerocross option allows
us to specify our own
filter.
The result after applying
an LoG filter and finding
its zero crossings is given
in figure beside
Jorma Kekalainen

Using an LoG filter first and

finding then zero crossings

Digital Image Processing

926

Marr-Hildreth method
log=fspecial('log',13,2);
edge(ic,'zerocross',log);

Jorma Kekalainen

Digital Image Processing

edge(ic,'log');

927

Edge detection
In an image, an edge is a curve that follows a path of rapid
change in image intensity.
Edges are often associated with the boundaries of objects in a
scene.
Edge detection is used to identify the edges in an image.
To find edges, we can use the edge function.
This function looks for places in the image where the intensity
changes rapidly, using one of these two criteria:
Places where the first derivative of the intensity is larger in magnitude
than some threshold
Places where the second derivative of the intensity has a zero crossing

Jorma Kekalainen

Digital Image Processing

928

Note
There are a a vast number of different edge detection algorithms
and methods.
As to which method is the best; like so many image processing
tasks, this is highly subjective.
The choice of an suitable edge detector will depend very much on
the nature of the image, the amount (and type) of noise, and the
use for which the edges are to be put.
The edge function provides a number of derivative estimators.
The most powerful edge-detection method that Matlabs edge
function provides is the Canny method.
The Canny method differs from the other edge-detection methods
in that it uses two different thresholds (to detect strong and weak
edges), and includes the weak edges in the output only if they are
connected to strong edges.
This method is therefore less likely than the others to be fooled by
noise, and more likely to detect true weak edges.
Jorma Kekalainen

Digital Image Processing

929

Example
I = imread('coins.png');

figure,imshow(I)
>> BW1 = edge(I,'sobel');
BW2 = edge(I,'canny');
figure;
imshowpair(BW1,BW2,'montage')
title('Sobel Filter
Canny Filter');

Jorma Kekalainen

Digital Image Processing

930

Exercise
Construct the next matrix image using matrix operations
50
50
50
50
50
50
50
50
50
50

50
50
50
50
50
50
50
50
50
50

50
50
200
200
200
200
50
50
50
50

50 50 50 50 50 50 50
50 50 50 50 50 50 50
200 200 200 200 200 50 50
200 200 200 200 200 50 50
200 200 200 200 200 50 50
200 200 200 200 200 50 50
50 200 200 200 200 50 50
50 200 200 200 200 50 50
50 50 50 50 50 50 50
50 50 50 50 50 50 50

Filter the previous image with fspecial('laplacian',0) and

construct the corresponding zero crossing chart.
Jorma Kekalainen

Digital Image Processing

931

Solution
>> B=50*ones(10)
B=
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50
50 50 50 50

50
50
50
50
50
50
50
50
50
50

>> B(3:8,3:8)=200*ones(6)
B=
50 50 50 50 50 50 50 50 50
50 50 50 50 50 50 50 50 50
50 50 200 200 200 200 200 200
50 50 200 200 200 200 200 200
50 50 200 200 200 200 200 200
50 50 200 200 200 200 200 200
50 50 200 200 200 200 200 200
50 50 200 200 200 200 200 200
50 50 50 50 50 50 50 50 50
50 50 50 50 50 50 50 50 50

>> B(7:8,3:4)=50*ones(2)
B=
50 50 50 50 50 50 50 50 50 50
50 50 50 50 50 50 50 50 50 50
50 50 200 200 200 200 200 200 50 50
50 50 200 200 200 200 200 200 50 50
50 50 200 200 200 200 200 200 50 50
50 50 200 200 200 200 200 200 50 50
50 50 50 50 200 200 200 200 50 50
50 50 50 50 200 200 200 200 50 50
50 50 50 50 50 50 50 50 50 50
Jorma Kekalainen
Digital Image Processing
50 50 50 50 50 50 50 50 50 50

50
50
50
50
50
50
50
50
50
50

50
50
50
50
50
50

B=100*ones(10)
B(3:8,3:8)=200*ones(6)
B(7:8,3:4)=100*ones(2)
l=fspecial('laplacian',0);
B_L=filter2(l,B)
932

Solution
l=fspecial('laplacian',0);
B_L=filter2(l,B)
B_L =
-100 -50 -50 -50 -50 -50 -50 -50 -50 -100
-50 0 150 150 150 150 150 150 0 -50
-50 150 -300 -150 -150 -150 -150 -300 150 -50
-50 150 -150 0 0 0 0 -150 150 -50
-50 150 -150 0 0 0 0 -150 150 -50
-50 150 -300 -150 0 0 0 -150 150 -50
-50 0 150 300 -150 0 0 -150 150 -50
-50 0 0 150 -300 -150 -150 -300 150 -50
-50 0 0 0 150 150 150 150 0 -50
-100 -50 -50 -50 -50 -50 -50 -50 -50 -100

Jorma Kekalainen

Digital Image Processing

933

940

Unsharp masking filter

We can in fact perform the filtering and subtracting operation
in one command, using the linearity of the filter, and that the
3*3 filter

is the identity filter.

Hence unsharp masking can be implemented by a filter of the
form

where k is a constant chosen to provide the best result.

Solution
Verify with Matlab the previous equality

>> B=-3*1/9*ones(3)
B=
-0.3333 -0.3333 -0.3333
-0.3333 -0.3333 -0.3333
>> C=1/3*[-1,-1,-1;-1,11,-1;-1,-1,-1]
-0.3333 -0.3333 -0.3333
C=
>> A+B
-0.3333 -0.3333 -0.3333
ans =
-0.3333 3.6667 -0.3333
-0.3333 -0.3333 -0.3333
-0.3333 -0.3333 -0.3333
-0.3333 3.6667 -0.3333
Jorma Kekalainen
Digital Image Processing
-0.3333 -0.3333 -0.3333

945

Example
Figure below was created using the Matlab commands
p=imread('cameraman.tif');
u=fspecial('unsharp',0.5);
pu=filter2(u,p);
imshow(p),figure,imshow(pu/255)
Original

Jorma Kekalainen

Figure after unsharp

masking, appears sharper
and cleaner than the
original.

After unsharp masking

Digital Image Processing

Note: Although we have

used averaging filters above,
we can use any low pass
946
filter for unsharp masking.

High boost filtering

Allied to unsharp masking filters are the high boost filters,
which are obtained by
high boost = A(original)-(low pass)
where A is an amplication factor.
If A=1, then the high boost filter becomes an ordinary high
pass filter.
If we take as the low pass filter the 3*3 averaging filter, then a
high boost filter will have the form
where z > 8

Jorma Kekalainen

Digital Image Processing

947

950

Note
We can also write the high boost formula above as

Best results for high boost filtering are obtained if we multiply

the equation by a factor so that the filter values sum to 1;
this requires A- =1 or =1/(A-1).
So a general unsharp masking formula is

Jorma Kekalainen

Digital Image Processing

951

High boost filters

A different version of the high boost filter is

where for best results A taken so that 3/5 A5/6.

If we take A=3/5 , the formula becomes

If we take A=5/6

Jorma Kekalainen

Digital Image Processing

952

Special high boost filters

Using the identity and averaging filters, we can obtain high
boost filters by:
>>id=[0 0 0;0 1 0;0 0 0];
>>f=fspecial('average');
>>hb1=3*id-2*f
hb1 =
-0.2222 -0.2222 -0.2222
-0.2222 2.7778 -0.2222
-0.2222 -0.2222 -0.2222
>> hb2=1.25*id-0.25*f
hb2 =
-0.0278 -0.0278 -0.0278
-0.0278 1.2222 -0.0278
Jorma
Kekalainen
Digital Image Processing
-0.0278
-0.0278 -0.0278

953

Special high boost filters

id=[0 0 0;0 1 0;0 0 0]
f=fspecial('average')
id =
0 0 0
0 1 0
0 0 0
f=
0.1111 0.1111 0.1111
0.1111 0.1111 0.1111
0.1111 0.1111 0.1111
Jorma Kekalainen

hb1=3*id-2*f
hb2=1.25*id-0.25*f
hb1 =
-0.2222 -0.2222 -0.2222
-0.2222 2.7778 -0.2222
-0.2222 -0.2222 -0.2222
hb2 =
-0.0278 -0.0278 -0.0278
-0.0278 1.2222 -0.0278
-0.0278 -0.0278 -0.0278

Digital Image Processing

954

Example
If each of the filters hb1 and hb2 are applied to an image with filter2, the
result will have enhanced edges.
The images were obtained with
x1=filter2(hb1,x);
x2=filter2(hb1,x);
imshow(x2/255)
imshow(x1/255)
(a) High boost filtering with hb1

Jorma Kekalainen

(b) High boost filtering with hb2

Digital Image Processing

962

Example
Consider the discrete
sequence

Expressing a discrete function

as the sum of sines.

If we compare IDFT to DFT we will see that
there are really only two differences:
1. there is no scaling factor 1/N.
2. the sign inside the exponential function has
been changed to positive.

Jorma Kekalainen

Digital Image Processing

966

Note
The formulas given in equations

are not used by all authors.

The main change is the position of the scaling factor 1/N.
Some authors put it in front of the sum in the forward formula.
Some authors put it in front of the sum in the inverse formula.
Others put a factor of 1/(N) in front of both sums.
The point is the sums by themselves would produce a result (after both
forward and inverse transforms) which is too large by a factor of N.
So somewhere in the forward-inverse formulas a corresponding 1/N must
exist; it doesn't really matter where.

Jorma Kekalainen

Digital Image Processing

967

Note
Perhaps, more familiar forms
2
DFT{x( n)} X( k ) x( n) exp j kn , k=0,1, ,N-1
N
n=0
N 1

1 N 1
2
x( n) = IDFT{X( k )} X( k ) exp j kn , n=0,1, ,N-1
N k =0
N

x(n) X(k)
Jorma Kekalainen

Digital Image Processing

968

Example
Express DFT of the sequence
1 , n = 0,1,2,6,7
x( n) =
3 n 5
0 ,

as compact as possible both with the help of

the complex exponentials and real cosine
functions. Calculate the both expressions in
Matlabs command window.

Jorma Kekalainen

Digital Image Processing

969

Example
Lets substitute the given sequence x
1 , n = 0,1,2,6,7
x( n) =
3 n 5
0 ,

and N=8 to the DFT formula

2
X( k ) = x( n) exp j kn

N
n=0
N 1

, k = 0,1,L , N 1

3
7
X( k ) = 1 + exp j k + exp j k + exp j k + exp j k
4
2

2
4

3
7
= 1 + exp j k + exp j k + exp j k + exp j k

4 2
2
4

= 1 + exp j k + exp j k + exp j k + exp j k
4 2
2
4
Jorma Kekalainen

= 1 + 2 cos k + 2 cos k , k = 0,1,L ,7
4
2Image
Processing
Digital

970

974

function [X]=muunnos1(x)
function [X]=muunnos1(x)
N=length(x);
P=log10(N)/log10(2);
J=1;
for I=1:N-1;
if I<J;
A=x(J);x(J)=x(I);x(I)=A;end;
K=N/2;
while K<J; J=JK;K=K/2;end;
J=J+K;
end;
Jorma Kekalainen

for L=1:P;
B=2^L;
B1=B/2;
V=1;
W=cos(pi/B1)-i*sin(pi/B1);
for J=1:B1;
for I=J:B:N;
C=I+B1;
A=x(C)*V;
x(C)=x(I)-A;
x(I)=x(I)+A;
end;
V=V*W;
end;
end;

979

Solution
>> xx=[1,1,1,0,0,0,1,1]
xx =
1 1 1 0 0 0 1 1
>> ifft(xx)
ans =
Columns 1 through 7
0.6250 0.3018 -0.1250 -0.0518 0.1250 -0.0518 -0.1250
Column 8
0.3018
>> muunnos2(xx)
ans =
Columns 1 through 3
0.6250 + 0.0000i 0.3018 - 0.0000i -0.1250 + 0.0000i
Columns 4 through 6
-0.0518 + 0.0000i 0.1250 + 0.0000i -0.0518 + 0.0000i
Columns 7 through 8
-0.1250 + 0.0000i 0.3018 - 0.0000i
Jorma Kekalainen

Digital Image Processing

980

Solution
>> ifft(xx')
ans =

0.6250

0.3018
-0.1250
-0.0518

0.1250
-0.0518
-0.1250

0.3018
Jorma Kekalainen

>> muunnos2(xx')
ans =
0.6250 + 0.0000i
0.3018 - 0.0000i
-0.1250 + 0.0000i
-0.0518 + 0.0000i
0.1250 + 0.0000i
-0.0518 + 0.0000i
-0.1250 + 0.0000i
0.3018 - 0.0000
Digital Image Processing

981

Exercise
Combine the functions of the previous
exercises so that one and the same function
performs FFT and IFFT, for example, with
commands like styles muunnos3('fft',x) or
muunnos3('ifft',x)

Jorma Kekalainen

Digital Image Processing

982

function [x]=muunnos3(type,x)
for L=1:P;
B=2^L;
B1=B/2;
V=1;
W=cos(pi/B1)-i*sin(pi/B1);
for J=1:B1;
for I=J:B:N;
C=I+B1;
A=x(C)*V;
x(C)=x(I)-A;
x(I)=x(I)+A;
end;
V=V*W;
end;
end;
if strcmp(type,'ifft');x=x'/N;end% IFFT

function [x]=muunnos3(type,x)
% FFT and IFFT algorithm
% type='fft' or 'ifft'
if strcmp(type,'ifft');x=x';end% Conjugate
N=length(x);
P=log10(N)/log10(2);
J=1;
for I=1:N-1;
if I<J; A=x(J);x(J)=x(I);x(I)=A;end;
K=N/2;
while K<J; J=J-K;K=K/2;end;
J=J+K;
end;
Jorma
Digital Image Processing
% Kekalainen

983

fft(x') vs. muunnos3('fft',x')

>> x =[1
x=
1 1
>> fft(x')
ans =
5.0000
2.4142
-1.0000
-0.4142
1.0000
-0.4142
-1.0000
2.4142
Jorma Kekalainen

>> muunnos3('fft',x')
ans =
5.0000 + 0.0000i
2.4142 + 0.0000i
-1.0000 + 0.0000i
-0.4142 - 0.0000i
1.0000 + 0.0000i
-0.4142 - 0.0000i
-1.0000 + 0.0000i
2.4142 + 0.0000i
Digital Image Processing

984

ifft(XX) vs. muunnos3('ifft',XX)

XX =
5.0000
2.4142
-1.0000
-0.4142
1.0000
-0.4142
-1.0000
2.4142

Jorma Kekalainen

>> ifft(XX)
ans =
1.0000
1.0000
1.0000
0
0.0000
0
1.0000
1.0000

>> muunnos3('ifft',XX)
ans =
1.0000 + 0.0000i
1.0000 - 0.0000i
1.0000 + 0.0000i
0.0000 + 0.0000i
0.0000 + 0.0000i
0.0000 + 0.0000i
1.0000 + 0.0000i
1.0000 - 0.0000i

Digital Image Processing

985

Note
>> a=rand(1,8)-0.5+j*(rand(1,8)-0.5)
a=
Columns 1 through 3
0.3147 + 0.4575i 0.4058 + 0.4649i -0.3730 - 0.3424i
Columns 4 through 6
0.4134 + 0.4706i 0.1324 + 0.4572i -0.4025 - 0.0146i
Columns 7 through 8
-0.2215 + 0.3003i 0.0469 - 0.3581i
>> a'
ans =
0.3147 - 0.4575i
0.4058 - 0.4649i
-0.3730 + 0.3424i
0.4134 - 0.4706i
0.1324 - 0.4572i
-0.4025 + 0.0146i
-0.2215 - 0.3003i
0.0469 + 0.3581i
Jorma Kekalainen
Digital Image Processing

986

fft(a) vs. muunnos3('fft',a)

a=rand(1,8)-0.5+j*(rand(1,8)-0.5)
>> fft(a)
ans =
Columns 1 through 3
0.3162 + 1.4353i 0.7771 - 0.9257i 1.3794 + 1.4137i
Columns 4 through 6
1.4377 - 0.7349i -0.6110 + 0.3098i -1.6977 + 1.2294i
Columns 7 through 8
0.7038 + 0.4999i 0.2124 + 0.4326i
>> muunnos3('fft',a)
ans =
Columns 1 through 3
0.3162 + 1.4353i 0.7771 - 0.9257i 1.3794 + 1.4137i
Columns 4 through 6
1.4377 - 0.7349i -0.6110 + 0.3098i -1.6977 + 1.2294i
Columns 7 through 8
Jorma Kekalainen
Digital Image Processing
0.7038 + 0.4999i 0.2124 + 0.4326i

987

fft(a') vs. muunnos3('fft',a')

>> fft(a')
ans =
0.3162 - 1.4353i
0.2124 - 0.4326i
0.7038 - 0.4999i
-1.6977 - 1.2294i
-0.6110 - 0.3098i
1.4377 + 0.7349i
1.3794 - 1.4137i
0.7771 + 0.9257i

Jorma Kekalainen

Digital Image Processing

>> muunnos3('fft',a')
ans =
0.3162 - 1.4353i
0.2124 - 0.4326i
0.7038 - 0.4999i
-1.6977 - 1.2294i
-0.6110 - 0.3098i
1.4377 + 0.7349i
1.3794 - 1.4137i
0.7771 + 0.9257i

988

ifft(a) vs. muunnos3('ifft',a)

>> ifft(a)
ans =
Columns 1 through 3
0.0395 + 0.1794i 0.0265 + 0.0541i 0.0880 + 0.0625i
Columns 4 through 6
-0.2122 + 0.1537i -0.0764 + 0.0387i 0.1797 - 0.0919i
Columns 7 through 8
0.1724 + 0.1767i 0.0971 - 0.1157i
>> muunnos3('ifft',a)
ans =
Columns 1 through 3
0.0395 + 0.1794i 0.0265 + 0.0541i 0.0880 + 0.0625i
Columns 4 through 6
-0.2122 + 0.1537i -0.0764 + 0.0387i 0.1797 - 0.0919i
Columns 7 through 8
0.1724 + 0.1767i 0.0971 - 0.1157i
Jorma Kekalainen

Digital Image Processing

989

ifft(a') vs. muunnos3('ifft',a')

>> ifft(a')
ans =
0.0395 - 0.1794i
0.0971 + 0.1157i
0.1724 - 0.1767i
0.1797 + 0.0919i
-0.0764 - 0.0387i
-0.2122 - 0.1537i
0.0880 - 0.0625i
0.0265 - 0.0541i

Jorma Kekalainen

>> muunnos3('ifft',a')
ans =
0.0395 - 0.1794i
0.0971 + 0.1157i
0.1724 - 0.1767i
0.1797 + 0.0919i
-0.0764 - 0.0387i
-0.2122 - 0.1537i
0.0880 - 0.0625i
0.0265 - 0.0541i

Digital Image Processing

990

Verification of muunnos3
a=rand(1,256)-0.5+j*(rand(1,256)-0.5);
X1=[];X2=[];
X1=[];X2=[];
X1=muunnos3('fft',a);
X2=fft(a);
max(abs(X2-X1))
ans =
3.2610e-14
figure
plot(real(X2-X1))
title('real(fft-muunnos3)')
grid
figure
plot(imag(X2-X1))
title('imag(fft-muunnos3)')
grid
Jorma Kekalainen

x1=[];
x2=[];
x1=muunnos3('ifft',a);
x2=ifft(a);
max(abs(x2-x1))
ans =
1.1857e-16
figure
plot(real(x2-x1))
title('real(ifft-muunnos3)')
grid
figure
plot(imag(x2-x1))
title('imag(ifft-muunnos3)')
grid
Digital Image Processing

991

Verification of muunnos3
a=rand(1,256)-0.5+j*(rand(1,256)-0.5);

Jorma Kekalainen

Digital Image Processing

992

Comparison of results
>> x'
ans =
1
1
1
0
0
0
1
1

x = Jorma
1 Kekalainen
1 1

>> muunnos1(x')
ans =
5.0000 + 0.0000i
2.4142 + 0.0000i
-1.0000 + 0.0000i
-0.4142 - 0.0000i
1.0000 + 0.0000i
-0.4142 - 0.0000i
-1.0000 + 0.0000i
2.4142 + 0.0000i

>> fft(x')
ans =
5.0000
2.4142
-1.0000
-0.4142
1.0000
-0.4142
-1.0000
2.4142
0

1 Digital1 Image Processing

>> dftv(x')
ans =
5.0000 + 0.0000i
2.4142 + 0.0000i
-1.0000 - 0.0000i
-0.4142 + 0.0000i
1.0000 + 0.0000i
-0.4142 - 0.0000i
-1.0000 - 0.0000i
2.4142 + 0.0000i
993

Comparison of results
>> a
a=
2
3
4
5
6
7
8
1

>> fft(a)
ans =
36.0000 + 0.0000i
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i
4.0000 + 0.0000i
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i

Jorma Kekalainen

>> muunnos1(a)
ans =
36.0000 + 0.0000i
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i
4.0000 + 0.0000i
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i

Digital Image Processing

>> dftv(a)
ans =
36.0000 + 0.0000i
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i
4.0000 + 0.0000i
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i
994

Simple DF transformer

998

PF=fft(pf') and QG=fft(qg')

PF =
1.0e+02 *
1.0800 + 0.0000i
-0.1200 + 0.2897i
-0.1200 + 0.1200i
-0.1200 + 0.0497i
-0.1200 + 0.0000i
-0.1200 - 0.0497i
-0.1200 - 0.1200i
-0.1200 - 0.2897i
Jorma Kekalainen

QG =
1.0e+02 *
2.8800 + 0.0000i
-0.3200 + 0.7725i
-0.3200 + 0.3200i
-0.3200 + 0.1325i
-0.3200 + 0.0000i
-0.3200 - 0.1325i
-0.3200 - 0.3200i
-0.3200 - 0.7725i
Digital Image Processing

999

H1=PF+QG vs. H
>> H-H1
H1 =
H=
ans =
1.0e+02 *
1.0e+02 *
3.9600 + 0.0000i
3.9600 + 0.0000i 1.0e-13 *
0.0000 + 0.0000i
-0.4400 + 1.0623i -0.4400 + 1.0623i
0.0000 + 0.1421i
-0.4400 + 0.4400i -0.4400 + 0.4400i 0.0000 + 0.0000i
-0.4400 + 0.1823i -0.4400 + 0.1823i 0.0000 + 0.0000i
-0.4400 + 0.0000i -0.4400 + 0.0000i 0.0000 + 0.0000i
-0.4400 - 0.1823i -0.4400 - 0.1823i 0.0000 + 0.0000i
-0.4400 - 0.4400i -0.4400 - 0.4400i 0.0000 + 0.0000i
-0.4400 - 1.0623i -0.4400 - 1.0623i 0.0000 - 0.1421i
Jorma Kekalainen

Digital Image Processing

max(abs(H-H1))
= 1.4211e-14

1000

H1=PF+QG vs. pfft(f')+qfft(g')

>> H1
H1 =
1.0e+02 *
3.9600 + 0.0000i
-0.4400 + 1.0623i
-0.4400 + 0.4400i
-0.4400 + 0.1823i
-0.4400 + 0.0000i
-0.4400 - 0.1823i
-0.4400 - 0.4400i
-0.4400 - 1.0623i
Jorma Kekalainen

>> 3*fft(f')+4*fft(g')
ans =
1.0e+02 *
3.9600 + 0.0000i
-0.4400 + 1.0623i
-0.4400 + 0.4400i
-0.4400 + 0.1823i
-0.4400 + 0.0000i
-0.4400 - 0.1823i
-0.4400 - 0.4400i
-0.4400 - 1.0623i
Digital Image Processing

1001

Properties of the one-dimensional

DFT: Shifting
Suppose we multiply each element xn of a
vector x by (-1)n .
In other words, we change the sign of every
second element.
Let the resulting vector be denoted x1.
Then the DFT X1 of x1 is equal to the DFT X of
x with the swapping of the left and right
halves.
Jorma Kekalainen

Digital Image Processing

1002

Example

>> x = [2 3 4 5 6 7 8 1]
x=
2 3 4 5 6 7
>> X=fft(x')
X=
36.0000 + 0.0000i
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i
4.0000 + 0.0000i
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i

>> x1=(-1).^[0:7].*x
x1 =
2 -3 4 -5 6 -7
>> X1=fft(x1')
X1 =
4.0000 + 0.0000i
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i
36.0000 + 0.0000i
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i

8 -1

Jorma Kekalainen
1003
Note:
The first four elements of X areDigital
theImage
lastProcessing
four elements of X1, and vice versa.

Example
>> x = [2 3 4 5 6 7 8 1];
>> X=fft(x')
36.0000
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i
4.0000
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i

>> x1=(-1).^[0:7].*x
x1 =
2 -3 4 -5 6 -7 8 -1
>> X1=fft(x1')
X1 =
4.0000
1.6569 + 4.0000i
-4.0000 + 4.0000i
-9.6569 - 4.0000i
36.0000
-9.6569 + 4.0000i
-4.0000 - 4.0000i
1.6569 - 4.0000i

Jorma Kekalainen
1004
Note:
The first four elements of X areDigital
theImage
lastProcessing
four elements of X1, and vice versa.

Conjugate symmetry
If x is real, and of length N , then its DFT X satisfies the
condition that
is the complex conjugate of XN-k , for all k=1,2, , N-1.
So in our example of length 8, we have
In this case we also have X4 = conj(X4), which means X4 must
X0=4.0000
be real.
X1=1.6569 + 4.0000i
In fact if N is even, then XN/2 will be real,
X2=-4.0000 + 4.0000i
X3=-9.6569 - 4.0000i
as we can see above.
X =36.0000
4

arithmetic of DFT

Jorma Kekalainen

Digital Image Processing

1008

Exercise
How length the vector (2n) must be if the
increase in speed with FFT compared to the
direct method is at least 1000 times?

Jorma Kekalainen

Digital Image Processing

1009

The two-dimensional DFT

In two dimensions, the DFT takes a matrix as input, and
returns another matrix, of the same size, as output.
If the original matrix values are f(x,y), where x an y are the
indices, then the output matrix values are F(u,v).
The matrix F
is the Fourier transform of f.
Then the original matrix f
is the inverse Fourier transform of F.
Jorma Kekalainen

Digital Image Processing

1010

are not used by all authors.

The main change is the position of the scaling factor 1/(MN).
Some authors put it in front of the sums in the forward formula.
Others put a factor of 1/(MN) in front of both sums.
The point is the sums by themselves would produce a result (after both
forward and inverse transforms) which is too large by a factor of MN.
So somewhere in the forward-inverse formulas a corresponding 1/(MN)
Jormamust
Kekalainen
Digital Image
Processing
1015
exist; it doesn't really matter
where.

Properties of the two dimensional

Fourier transform
All the properties of the one-dimensional DFT transfer
into two dimensions.
First we notice that the forward and inverse transforms
are very similar, with the exception of the scale factor
1/(MN) in the inverse transform, and the negative sign
in the exponent of the forward transform (similarity
property).
This means that nearly the same algorithm, only very
slightly adjusted, can be used for both the forward an
inverse transforms.
But there are some further properties, which are of
particular use for image processing.
Jorma Kekalainen

Digital Image Processing

1016

DFT as a spatial filter

Note that the values

are independent of the values f or F.

This means that they can be calculated in advance, and only then
put into the formulas above.
It also means that every value F(u,v) is obtained by multiplying
every value of f(x,y) by a fixed value, and adding up all the results.
But this is precisely what a linear spatial filter does: it multiplies all
elements under a mask with fixed values, and adds them all up.
Thus we can consider the DFT as a linear spatial filter which is as big
as the image.

Jorma Kekalainen

To deal with the problem of edges (image

borders), we assume that the image is tiled
(duplicated) in all directions, so that the mask
Digital Image Processing
always has image
values to use.

1017

Separability
Notice that the Fourier transform filter elements can be
expressed as products:

The first product value

depends only on x and u, and is independent of y and v.
Conversely, the second product value

depends only on y and v, and is independent of x and u.

Jorma Kekalainen

Digital Image Processing

1018

Separability
This means that we can break down our long formulas above
to simpler formulas that work on single rows or columns e.g.:

If we replace x and u with y and v we obtain the

corresponding formulas for the DFT of matrix columns.
These formulas define the one-dimensional DFT of a vector, or
simply the DFT.
Jorma Kekalainen

Digital Image Processing

1019

Calculating a 2-D DFT

The 2-D DFT can be calculated by using this property of
separability.
To obtain the 2-D DFT of a matrix, we first calculate the DFT of
all the rows, and then calculate the DFT of all the columns of
the result.
Since a product is independent of the order, we can equally
well calculate a 2-D DFT by calculating the DFT of all the
columns first, then calculating the DFT of all the rows of the
result.
Original
image
Jorma Kekalainen

DFT of
each row
Digital Image Processing

DFT of
each
column
1020

Linearity
An important property of the DFT is its linearity; the DFT of a
sum is equal to the sum of the individual DFT's, and the same
goes for scalar multiplication:

where k is scalar, and f and g are matrices.

This property is of great use in dealing with image
degradation such as noise which can be modeled as a sum:

d=f+n
where f is the original image; n is the noise and d is the
degraded image.
Jorma Kekalainen

Digital Image Processing

1021

This term is equal to the sum of all terms in the original

matrix.
The value F(0, 0) of the DFT is called the DC coefficient.
Jorma Kekalainen

Digital Image Processing

1027

Shifting a DFT
For purposes of display, it is convenient to have the DC
coefficient in the centre of the matrix.
This will happen if all elements f(x,y) in the matrix are
multiplied by (-1)x+y before the transform.
Figure below demonstrates how the matrix is shifted by this
method.
In each diagram the DC coefficient is the top left hand
element of submatrix A and is shown as a black square.
FFT

Jorma Kekalainen

After shifting

Digital Image Processing

1028

Conjugate symmetry
An analysis of the Fourier transform definition
leads to a symmetry property; if we make the
substitutions u=-u and v=-v in the equation

we find that
For integers
p and q

This is the conjugate symmetry property of

DFT.

Jorma Kekalainen

Digital Image Processing

1029

Conjugate symmetry in the DFT

Conjugate symmetry means that
half of the transform is a mirror
image of the conjugate of the
other half.
We can think of the top and
bottom halves, or the left and
right halves, being mirror images
of the conjugates of each other.
Figure demonstrates this
symmetry in a shifted DFT.
The black square shows the
position of the DC coefficient.
The symmetry means that its
information is given in just half
of a transform, and the other
half is redundant.

Jorma Kekalainen

Digital Image Processing

1030

Displaying transforms
Having obtained the Fourier transform F(u, v) of an
image f(x, y) we would like to see what it looks like.
As the elements F(u, v) are complex numbers, we cant
view them directly, but we can view their magnitude
|F(u, v)|.
Since these will be numbers of type double, generally
with large range, we have two approaches
1. find the maximum value m of |F(u, v)| (this will be the
DC coefficient), and use imshow to view |F(u, v)|/m,
2. use mat2gray to view |F(u, v)| directly.
Jorma Kekalainen

Example
Suppose we take a constant matrix f(x, y) = 1.
Going back to the idea of a sum of corrugations, we
see that in fact no corrugations are required to form
a constant.
Thus we would hope that the DFT consists of a DC
coefficient and zeroes everywhere else.
We will use the ones function, which produces an
(n x n) matrix consisting of 1s, where n is an input to
the function.
Jorma Kekalainen

Digital Image Processing

1034

Example
>> a=ones(8);
>> fft2(a)
The result is indeed as we expected:
ans =
64 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
Note that the DC coefficient is indeed the sum of all the matrix values.

Jorma Kekalainen

Digital Image Processing

1035

Example
>> a=ones(8)
a=

1 1 1

1
1
1
1
1
1
1
1

>> fft2=fft2(a)
fft2 =

64 0 0 0

0 0 0 0

0
0
0
0
0
0
0
0

Note:
Jorma Kekalainen

Digital Image Processing

1036

Example
Now we'll take a matrix consisting of a single corrugation:
>> a = [100 200; 100 200];
>> a = repmat(a,4,4)
ans =

Jorma Kekalainen

Digital Image Processing

1037

0
0
0
0
0
0
0
0

1040

Example
We will take here a single step edge:
>> a = [zeros(8,4) ones(8,4)]
a=

Jorma Kekalainen

Digital Image Processing

1041

Example: Shifted DFT

a = [zeros(8,4) ones(8,4)]
af=fftshift(fft2(a));

round(abs(af))
%
a=

0 0 0 0 1 1 1 1

0 0 0 0 1 1 1 1
ans =

0 0 0 0 0 0 0 0

0 9 0 21 32 21 0 9

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0
Jorma Kekalainen
0 0 0 0 0 0 0 0

Digital Image Processing

1042

Example: Shifted DFT

Lets perform the Fourier transform with a shift, to place the
DC coefficient in the centre, and since it contains some
complex values, for simplicity we shall just show the rounded
absolute values:
The DC coefficient is the sum of all values
of a; the other values may be considered
>> af=fftshift(fft2(a));
to be the coefficients of the necessary sine
>> round(abs(af))
functions required to from an edge.
ans =

Jorma Kekalainen

Digital Image Processing

The mirroring of
values about the
DC coefficient is
a consequence
of the symmetry
of the DFT.1043

Example
Lets create a few simple images, and see what the Fourier
transform produces.
The first example is a single edge
>> a=[zeros(256,128) ones(256,128)];
This is displayed
Then we shall take its DFT,
and shift it:
>> af=fftshift(fft2(a));

Digital Image Processing

1047

fftshow(f,type) function
function [ ] = fftshow(f,type)
if nargin<2,
%Function to display a Fourier transform
type='log';
%fftshow(f,'log') and fftshow(f,'abs')
end
%
if (type=='log')
% Usage: FFTSHOW(F,TYPE)
fl = log(1+abs(f));
%
fm = max(fl(:));
% Displays the fft matrix f using imshow,
where %TYPE must be one of 'abs' or 'log'.
imshow(im2uint8(fl/fm))
%If TYPE='abs', then then abs(f) is displayed;
elseif (type=='abs')
%if TYPE='log' then log(1+abs(f)) is displayed.
fa=abs(f);
%If TYPE is omitted, then 'log' is chosen as a
fm=max(fa(:));
%default.
imshow(fa/fm)
%
else
% Example:
error('TYPE must be abs or log.');
% c=imread('image.tif');
% cf=fftshift(fft2(c));
end
% fftshow(cf,'abs')
Jorma
Digital Image Processing
1048
% Kekalainen

Using fftshow(f,type)
Then for example
fftshow(af,'log')
will show the logarithm of the absolute
values of the transform log(1+abs(f)),
and
fftshow(af,'abs')
will show the absolute values of the
transform abs(f) without any scaling.
An example command sequence is
shown on the right.

Jorma Kekalainen

Digital Image Processing

1052

Example: Box

Jorma Kekalainen

Digital Image Processing

1053

Example

A
rotated
box and
its DFT

Now we shall look at a box

rotated 45
>> [x,y]=meshgrid(1:256,1:256);
>> b=(x+y<329)&(x+y>182)&(xy>-67)&(x-y<73);
>> imshow(b)
>> bf=fftshift(fft2(b));
>> figure,fftshow(bf)
Note that the transform of
the rotated box is the rotated
transform of the original box.

Jorma Kekalainen

Digital Image Processing

1054

Note: meshgrid
[X,Y] = meshgrid(xgv,ygv)
replicates the grid vectors xgv
and ygv to produce a full grid.

>> [x,y]=meshgrid(1:5,1:5)
x=
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
y=
1 1 1 1 1
2 2 2 2 2
3 3 3 3 3
4 4 4 4 4
Jorma Kekalainen
5 5 5 5 5

>> [x,y]=meshgrid(-4:3,-4:3)
x=
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
y=
-4 -4 -4 -4 -4 -4 -4
-3 -3 -3 -3 -3 -3 -3
-2 -2 -2 -2 -2 -2 -2
-1 -1 -1 -1 -1 -1 -1
0 0 0 0 0 0 0
1 1 1 1 1 1 1
2 2 2 2 2 2 2
Digital Image
3 Processing
3 3 3 3 3 3

3
3
3
3
3
3
3
3
-4
-3
-2
-1
0
1
2
3

1055

Note

Jorma Kekalainen

Digital Image Processing

1056

Example
We will create a small circle,
[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z<15);
and then transform and display it:
cf=fft2shift(fft2(z));
>> fftshow(cf,'log')

Jorma Kekalainen

Digital Image Processing

1057

Note
>> x=1:5
x=
1 2 3

[X,Y] = meshgrid(xgv,ygv)
replicates the grid vectors xgv
and ygv to produce a full grid.
>> [x,y]=meshgrid(1:5,1:5)
x=
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
1 2 3 4 5
y=
1 1 1 1 1
2 2 2 2 2
3 3 3 3 3
4 4 4 4 4
5 5 5 5 5

Jorma Kekalainen

>> z=sqrt(x.^2+y.^2)
z=
1.4142 2.2361 3.1623
2.2361 2.8284 3.6056
3.1623 3.6056 4.2426
4.1231 4.4721 5.0000
5.0990 5.3852 5.8310

Digital Image Processing

4.1231
4.4721
5.0000
5.6569
6.4031

5.0990
5.3852
5.8310
6.4031
7.0711

1058

Note
>> [x,y]=meshgrid(-4:3,-4:3)
x=
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2
y=
-4 -4 -4 -4 -4 -4 -4
-3 -3 -3 -3 -3 -3 -3
-2 -2 -2 -2 -2 -2 -2
-1 -1 -1 -1 -1 -1 -1
0 0 0 0 0 0 0
1 1 1 1 1 1 1
2 2 2 2 2 2 2
3 3 3 3 3 3 3
Jorma Kekalainen

3
3
3
3
3 >> z=sqrt(x.^2+y.^2)
3 z=
5.6569 5.0000 4.4721
3
5.0000 4.2426 3.6056
3
4.4721 3.6056 2.8284
-4 4.1231 3.1623 2.2361
-3 4.0000 3.0000 2.0000
-2 4.1231 3.1623 2.2361
-1 4.4721 3.6056 2.8284
5.0000 4.2426 3.6056
0
1
2
3
Digital Image Processing

4.1231
3.1623
2.2361
1.4142
1.0000
1.4142
2.2361
3.1623

4.0000 4.1231 4.4721 5.0000

3.0000 3.1623 3.6056 4.2426
2.0000 2.2361 2.8284 3.6056
1.0000 1.4142 2.2361 3.1623
0 1.0000 2.0000 3.0000
1.0000 1.4142 2.2361 3.1623
2.0000 2.2361 2.8284 3.6056
3.0000 3.1623 3.6056 4.2426

1059

DFT of the small circle

[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z<15);
figure, imshow(c)
title('Small circle (c<15)')
%
cf=fftshift(fft2(c));
figure, fftshow(cf,'abs')
title('DFT of the small circle')
xlabel('fftshow(cf,''abs'')')
figure, fftshow(cf,'log')
title('DFT of the small circle')
xlabel('fftshow(cf,''log'')')

Jorma Kekalainen

1063

Digital Image Processing

Jorma Kekalainen

Digital Image Processing

Frequency Domain
Filtering

Filtering in the frequency domain

We have seen previously that one of the
reasons for the use of the Fourier transform in
image processing is due to the convolution
theorem: a spatial convolution can be
performed by element-wise multiplication of
the Fourier transform by a suitable filter
matrix.
Here we shall explore some filtering by this
method.
Jorma Kekalainen

Digital Image Processing

1066

One way to do this is to multiply the transform by an
ideal low-pass matrix, which is a binary matrix m
defined by:

Then the inverse Fourier transform of the elementwise product of F and m is the result we require:

Jorma Kekalainen

Digital Image Processing

1070

Note: The circle c displayed previously is just such a matrix with D=15.

Example
Let's see what happens if we apply this filter to an image. First
we obtain an image and its DFT.
>> cm=imread('cameraman.tif');
>> cf=fftshift(fft2(cm));
>> figure,fftshow(cf,'log')
The cameraman image and its DFT are shown below.

Jorma Kekalainen

Digital Image Processing

1071

Ideal filtering on the DFT

Now we can perform a low pass filter by multiplying the
transform matrix by the circle matrix (recall that dot asterisk is
the Matlab syntax for element-wise multiplication of two
matrices):
>> cfl=cf.*c;
>> figure,fftshow(cfl,'log')
and this is shown beside.

Jorma Kekalainen

Digital Image Processing

1072

Example

Jorma Kekalainen

Digital Image Processing

1073

Applying ideal low pass filtering

Now we can take the inverse transform and display the result:
>> cfli=ifft2(cfl);
>> figure,fftshow(cfli,'abs')
After inversion
and this is shown beside.

Note the ringing about the

edges in this image. This is a
direct result of the sharp cutoff
of the circle. The ringing is
Jorma Kekalainen
Digital Image Processing
transferred
to the image.

1074

Jorma Kekalainen

Digital Image Processing

1075

Comment
Note that even though cfli is obviously a matrix of
real numbers, we are still using fftshow to display
it.
This is because the ifft2 and fft2 functions being
numeric, will not produce mathematically perfect
results, but rather very close numeric
approximations.
So using fftshow with the 'abs' option rounds out
any errors obtained during the transform and its
inverse.
Jorma Kekalainen

Digital Image Processing

1076

Example
C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
figure,fftshow(Cf,'log')
title('DFT of the cameraman image')
xlabel('fftshow(Cf,''log'')')
%
%Lowpass filtering
[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z<15);
Cfl=Cf.*c;% Lowpass filtering(symbol l)
figure,fftshow(Cfl,'log')
title('DFT of the cameraman image multiplied by an ideal low-pass matrix c')
xlabel('fftshow(Cfl,''log'')')
%
%Now we can take the inverse transform and display the result:
Cfli=ifft2(Cfl);
figure,fftshow(Cfli,'abs')
title('IDFT [the cameraman image multiplied by an ideal low-pass matrix c]')
xlabel('fftshow(Cfli,''abs'')')
Jorma
Kekalainen
Digital Image Processing

1077

DFT of a very small circle (c<5)

We would expect that the smaller the circle, the more blurred the image, and the
larger the circle; the less blurred.
%Effect of a very small circle (c<5)
[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z<5);
figure, imshow(c)
title('Very small circle (c<5)')
%
cf=fftshift(fft2(c));
figure, fftshow(cf,'log')
title('DFT of the very small circle(c<5)')
xlabel('fftshow(cf,''log'')')
%without function
%figure, imshow(mat2gray(log(1+abs(cf))))
%title('DFT of the very small circle (c<5)')

Jorma Kekalainen

Digital Image Processing

1078

Effect of a very small circle (c<5)

C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
figure,fftshow(Cf,'log')
title('DFT of the cameraman image')
xlabel('fftshow(Cf,''log'')')
%Lowpass filtering
Cfl=Cf.*c;% Lowpass filtering(symbol l)
figure,fftshow(Cfl,'log')
title('DFT of of the cameraman image multiplied by an ideal low-pass matrix c<5')
xlabel('fftshow(Cfl,''log'')')
%
%Now we can take the inverse transform and display the result:
Cfli=ifft2(Cfl);
figure,fftshow(Cfli,'abs')
title('IDFT the cameraman image multiplied by an ideal low-pass matrix c(<5)')
xlabel('fftshow(Cfli,''abs'')')

Jorma Kekalainen

Note
We would expect that the smaller the circle, the more blurred
the image, and the larger the circle; the less blurred.
Figure below demonstrates this, using cutoffs of 5 and 30.
Notice that ringing is still present, and clearly visible in figure (b).
(a) Cutoff 5

Jorma Kekalainen

%Experiments with gentle cutoffs

promt='Give cutoff value r=';
r=input(promt)
[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z<r);
%Circle with gentle cutoff or spatially blurred circle
b=1./(1+(z./r).^2);
%
figure,imshow(b)
title(['Circle c<',num2str(r),' with a gentle cutoff'])
%
bf=fftshift(fft2(b));
figure, fftshow(bf,'log')
title(['DFT of the circle c<',num2str(r),' with a
gentle cutoff'])
xlabel('fftshow(cf,log)')
%without function
%figure, imshow(mat2gray(log(1+abs(bf))))
%title('DFT of the circle c<30) with a gentle cutoff')
%

Jorma Kekalainen

C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
figure,fftshow(Cf,'log')
title('DFT of the cameraman image')
%Lowpass filtering
Cfl=Cf.*b;% Lowpass filtering(symbol l)
figure,fftshow(Cfl,'log')
title(['DFT of of the cameraman image multiplied
by a blurred low-pass matrix c<',num2str(r)])
%
%Now we can take the inverse transform and
display the result:
%
Cfli=ifft2(Cfl);
figure,fftshow(Cfli,'abs')
title(['IDFT the cameraman image multiplied by a
blurred low-pass matrix c<',num2str(r)])

Digital Image Processing

1086

Digital Image Processing

High Pass Filtering

High pass filtering

Just as we can perform low pass filtering by
keeping the center values of the DFT and
eliminating the others, so high pass filtering
can be performed by the opposite: eliminating
center values and keeping the others.
This can be done with a minor modification of
the preceding method of low pass filtering.

Jorma Kekalainen

Digital Image Processing

1088

High pass filtering

First we create the circle:
>> [x,y]=meshgrid(-128:127,-128:127);
>> z=sqrt(x.^2+y.^2);
>> c=(z>15);
and multiply it by the DFT of the image:
>> cfh=cf.*c;
>> figure,fftshow(cfh,'log')

Jorma Kekalainen

Digital Image Processing

The DFT after

high pass filtering

1089

IDFT
The inverse DFT can be easily produced and displayed:
>> cfhi=ifft2(cfh);
The resulting image
>> figure,fftshow(cfhi,'abs')
and this is shown beside.

As with low pass filtering, the

size of the circle influences the
information available to the
inverse DFT, and hence the
final result.
Jorma Kekalainen

Digital Image Processing

1090

High pass filtered (c>15) image

Jorma Kekalainen

Digital Image Processing

1091

Example
%High pass filtering
%First we create the filter circle:
[x,y]=meshgrid(-128:127,-128:127);
z=sqrt(x.^2+y.^2);
c=(z>15);% High pass filtering
figure, imshow(c)
title('Ideal high pass filter (c>15)')
C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
%and the multiply it by the DFT of the image:
Cfh=Cf.*c;
figure,fftshow(Cfh,'log')
title('Product of an ideal high pass filter (c>15) and DFT of the image')
%
% The inverse DFT can be easily produced and displayed:
Cfhi=ifft2(Cfh);
figure,fftshow(Cfhi,'abs')
Kekalainen
Image Processing
Jormatitle('IDFT
of the ideal high passDigital
filtered
(c>15) image')

1092

Ideal high pass filtering with

different cutoffs
The following figures show some results of ideal high pass
filtering with different cutoffs.
If we have small cutoff, such as in the next figure, we are
only removing a small amount of the transform.
We would thus expect that only the lowest frequencies of
the image would be removed; then there is some grayscale
detail in the final image, but large areas of low frequency
are close to zero.
If the cutoff is large, then more information is removed
from the transform, leaving only the highest frequencies.
We also see this in the following figures; only the edges of
the image remain.

Jorma Kekalainen

Digital Image Processing

1093

Examples
Cutoffs: c>15, c>30, c>60

Cutoffs: c>1, c>2, c>5

Jorma Kekalainen

Digital Image Processing

1094

c>50

c>100

c>120
Jorma Kekalainen

Digital Image Processing

1095

Digital Image Processing

Butterworth Filtering

Butterworth filtering
Ideal filtering simply cuts off the Fourier
transform at some distance from the center.
This is very easy to implement, as we have
seen, but has the disadvantage of introducing
unwanted effects: ringing, into the result.
One way of avoiding this is to use as a filter
matrix a circle with a less sharp cutoff.
A popular choice is to use Butterworth filters.
Jorma Kekalainen

Digital Image Processing

1097

Butterworth filters
Butterworth filters are based on the following functions for
low pass filters and for high pass filters :

where the parameter n is called the order of the filter and D is

cutoff radius.
The size of n dictates the sharpness of the cutoff.
Butterworth filters with n = 2

Digital Image Processing

1101

Example
Applying a Butterworth low
pass filter to the DFT of the
cameraman image we get
results shown beside and
the next slices.
Note that there is no sharp
cutoff as seen previously
with the ideal low pass
filter; also that the outer
parts of the transform are
not equal to zero, although
they are dimmed
considerably.
Jorma Kekalainen

DFT after Butterworth

low pass filtering

Digital Image Processing

1102

Note
Compare the transform after multiplying with a Butterworth filter with
the original transform
We see that the Butterworth filter causes an attenuation of values away
from the center, but they don't become suddenly zero, as with the ideal
low pass filter.

Jorma Kekalainen

Digital Image Processing

1103

Example
Performing the inverse
transform and
displaying it as we have
done previously
produces the resulting
image beside.
This is a blurred image,
but now the ringing is
completely absent.

Jorma Kekalainen

Resulting image

Digital Image Processing

1104

Compare

Jorma Kekalainen

Digital Image Processing

1105

Exercise
Study effects of Butterworth low pass filter
with different order (n) and cutoff radius (D)
on cameraman.tif image

Jorma Kekalainen

Digital Image Processing

1106

Solution

1117

Solution: Bu high pass filters

n=1

(
)
f
x
=
2n
low
x

1
f (x ) =
2n
low
D
1+

=1, kun

f low ( x ) + f high (x ) =

1
x
1+
D

1
D
1+
x
2n

Jorma Kekalainen

D
x
1+ +1+
x
D
=
x 2n D 2n
1 + 1 +
D x
2n

D
x
D
x
1+ +1+
2+ +
x
D
x
D =1
=
=
2n
2n
2n
2n
2n
2n
x
D
x
D
D
x

1+ + +
2+ +
x Image1
D4
x4
D
Digital
2
4
3
x
D 1121
Processing
4
=1

Solution: Low pass Butterworth

filter function
function lout=lbutter(im,d,n)
% LBUTTER(IM,D,N) creates a low-pass Butterworth filter
% of the same size as image IM, with cutoff D, and order N
%
% Use e.g.:
% x=imread('cameraman.tif');
% l=lbutter(x,25,2);
%
height=size(im,1);
width=size(im,2);
[x,y]=meshgrid(-floor(width/2):floor((width-1)/2),-floor(height/2):
...
floor((height-1)/2));
lout=1./(1+((x.^2+y.^2)/d^2).^n);

Jorma Kekalainen

Digital Image Processing

1122

Solution: High pass Butterworth

filter function
function hout=hbutter(im,d,n)
% HBUTTER(IM,D,N) creates a high-pass
%Butterworth filter of the same size as image IM,
%with cutoff D, and order N
%
% Use e.g.:
% x=imread('image.tif');
% l=hbutter(x,25,2);
%
hout=1-lbutter(im,d,n);
Jorma Kekalainen

Digital Image Processing

1123

Solution: Low or high pass Bufilter function

function [ out] = butter(im,d,n,type)

% A function to generate a low pass or high pass Butterworth filter
%function out=butter(im,d,n,type)
% BUTTER(IM,D,N,TYPE) creates a low-pass or high pass Butterworth filter
height=size(im,1);
width=size(im,2);
[x,y]=meshgrid(-floor(width/2):floor((width-1)/2),-floor(height/2): ...
floor((height-1)/2));
%
if strcmp(type,'low')
out=1./(1+((x.^2+y.^2)/d^2).^n);
elseif strcmp(type,'high')
out=1-1./(1+((x.^2+y.^2)/d^2).^n);
else
error('TYPE must be ''low'' or ''high''')
end

1127

Example
% Applying Butterworth low pass filter
C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
figure,fftshow(Cf,'log')
title('DFT of the cameraman image')
bl=lbutter(Cf,15,1);
Cfbl=Cf.*bl;
figure,fftshow(Cfbl,'log')
title('DFT of the cameraman image
multiplied by Bu-lpf (D=15,n=1)')
%inverting and displaying the result:
Cfbli=ifft2(Cfbl);
figure,fftshow(Cfbli,'abs')
title('IDFT')

Jorma Kekalainen

%Applying Butterworth high pass filter

C=imread('cameraman.tif');
figure, imshow(C)
title('Original image - cameraman.tif')
Cf=fftshift(fft2(C));
figure,fftshow(Cf,'log')
title('DFT of the cameraman image')
bh=hbutter(Cf,15,1);
Cfbh=Cf.*bh;
figure,fftshow(Cfbh,'log')
title('DFT of the cameraman image
multiplied by Bu-hpf (D=15, n=1)')
%inverting and displaying the result:
Cfbhi=ifft2(Cfbh);
figure,fftshow(Cfbhi,'abs')
title('Bu-highpass filtered (D=15, n=1)
image ')

1131

Digital Image Processing

Gaussian Filtering in
Frequency Domain

Gaussian filtering
We have discussed Gaussian filters earlier, and we saw that
they could be used for low pass filtering in the spatial
domain.
However, we can also use Gaussian filters in the frequency
domain.
As with ideal and Butterworth filters, the implementation is
very simple: create a Gaussian filter, multiply it by the
image transform, and invert the result.
Since Gaussian filters have the very nice mathematical
property that a Fourier transform of a Gaussian is a
Gaussian, we should get exactly the same results as when
using a linear Gaussian spatial filter.

Jorma Kekalainen

Digital Image Processing

1133

Gaussian filter
Gaussian filters may be considered to be the most smooth of all the filters
we have discussed so far, with ideal filters the least smooth, and
Butterworth filters in the middle.
We can create Gaussian filters using the fspecial function, and apply them
to our transform.
C=imread('cameraman.tif');
Cf=fftshift(fft2(C));
h = fspecial('gaussian', hsize, sigma)
g1=mat2gray(fspecial('gaussian',256,10)); returns a rotationally symmetric
Gaussian lowpass filter of size hsize
Cg1=Cf.*g1;
with standard deviation sigma
fftshow(Cg1,'log')
(positive). hsize can be a vector
g2=mat2gray(fspecial('gaussian',256,30)); specifying the number of
rows and columns in h, or it can be a
Cg2=Cf.*g2;
scalar, in which case h is a square
figure,fftshow(Cg2,'log')
Jorma Kekalainen

matrix.
The default value for hsize is [3 3]; the
Digital Image Processing
1134
default value for sigma is 0.5.

Note
Note the use of the mat2gray function.
The fspecial function on its own produces a low pass Gaussian filter with a
very small maximum:
>> g=fspecial('gaussian',256,10);
format long, max(g(:))
ans =

0.001587575526792
>> g=fspecial('gaussian',256,30);
format long, max(g(:)), format
ans =

filter in the frequency domain
=10

=30

Note: As with ideal and Butterworth filters, the wider the high pass
filter, the more of the transform we are reducing, and the less of the
Jorma Kekalainen
Digital Image Processing
1139
original
image will appear in the
result.

Gaussian hp filtered (size=256,

std=10 and std=30) DFT-image

Jorma Kekalainen

Image with periodic noise, and its

transform
This can be viewed directly with imshow, and it is shown
below .
The extra two spikes away from the center
We can produce its DFT: correspond to the noise just added. In general
>> cpf=fftshift(fft2(cp)); the smaller the period of the noise, the further
from the center the two spikes will be.

Jorma Kekalainen

Digital Image Processing

This is because a
small period
corresponds to a
high frequency
(large change over
a small distance),
and is therefore
further away from
the center of the
1144
shifted transform.

Removing noise spikes

We will now remove these extra spikes, and invert the result.
If we use data cursor and move it around the image, we find that the
spikes have row, column values of (156,170) and (102,88).
These have the same distance from the center: 49.0918.
We can check this by
>>z=sqrt(x.^2+y.^2);
>> z(156,170)
>> z(102,88)
ans =49.0918
ans =49.0918
There are two methods we can
use to eliminate the spikes
Band reject filtering
Jorma Kekalainen
Notch
filtering

Digital Image Processing

1145

Band reject filtering

We create a filter consisting of ones with a ring of zeroes; the
zeroes lying at a radius of 49 from the center:
br=(z < 47 | z > 51);
where z is the matrix consisting of distances from the origin.
This particular ring will have a thickness large enough to cover the
spikes.
Then as before, we multiply this by the transform:
cpfbr=cpf.*br
The result is that the spikes have been blocked out by this filter.
Taking the inverse transform produces the image
Cpfbri=ifft2(Cpfbr);
figure, fftshow(Cpfbri,'log')

Digital Image Processing

1149

Digital Image Processing

Inverse Filtering

Inverse filtering
We have seen that we can perform filtering in the Fourier
domain by multiplying the DFT of an image by the DFT of a
filter: this is a direct use of the convolution theorem.
We thus have
where X is the DFT of the image; F is the DFT of the filter, and
Y is the DFT of the result.
If we are given Y and F, then we should be able to recover the
DFT of the original image X simply by dividing by F:

Jorma Kekalainen

Digital Image Processing

1151

Example
Suppose, for example we take the cameraman image
cameraman.tif, and blur it using a low-pass Butterworth filter:
C=imread('cameraman.tif');
Cf=fftshift(fft2(C));
b=lbutter(C,15,2);
Cb=Cf.*b;
Cba=abs(ifft2(Cb));
Cba=uint8(255*mat2gray(Cba));
imshow(Cba)
The result is shown beside.
Jorma Kekalainen

Digital Image Processing

1152

Example
We can attempt to recover the
original image by dividing by the
filter:
C1=fftshift(fft2(Cba))./b;
C1a=abs(ifft2(C1));
imshow(mat2gray(C1a))
The result is shown beside and
there is no improvement!

1156

Example

C=imread('cameraman.tif');
Cf=fftshift(fft2(C));
% Blurring the image
b=lbutter(C,15,2);
Cb=Cf.*b;
Cba=abs(ifft2(Cb));
Cba=uint8(255*mat2gray(Cba));
figure, imshow(Cba)
title('Image blurred by Bu-lpf (D=15, n=2)')
%
%We can attempt to recover the original image by dividing by the filter:
%
C1=fftshift(fft2(Cba))./b;
C1a=abs(ifft2(C1));
figure, imshow(mat2gray(C1a))
title('Recovery attempt of the original image at inverse filtering')

Jorma Kekalainen

Digital Image Processing

1157

Example
We can at randomly test on the ability of Butterworth
low pass filter with different cutoff radius and order to
eliminate very low values.
promt1='Give filter order n=';
n=input(promt1)
promt2='Give cutoff radius D=';
D=input(promt2)
Cbf=fftshift(fft2(Cba));
C1=(Cbf./b).*lbutter(C,D,n);
C1a=abs(ifft2(C1));
figure, imshow(mat2gray(C1a))
title(['Inverse filtering using Bu-lpf (D=',num2str(D),',
n=',num2str(n),') to eliminate
zeros'])
Jorma Kekalainen
Digital Image Processing

1158

Inverse filtering using

constrained division
We can try the second method; to implement
it we simply make all values of the filter which
are too small equal to 1 e.g.:
d=0.01;
b=lbutter(C,15,2);b(find(b<d))=1;
C1=fftshift(fft2(Cba))./b;
C1a=abs(ifft2(C1));
imshow(mat2gray(C1a))
Jorma Kekalainen

Digital Image Processing

fspecial('motion',len,theta)
m= fspecial('motion',len,theta) returns a filter to
approximate, once convolved with an image, the
linear motion of a camera by len pixels, with an
angle of theta degrees in a counterclockwise
direction.
The filter becomes a vector for horizontal and
vertical motions.
The default len is 9 and the default theta is 0,
which corresponds to a horizontal motion of nine
pixels.
Jorma Kekalainen

Digital Image Processing

1165

Note
fspecial(type) creates a two-dimensional filter of the specified
type. type is a string having one of these values.
Value

Description

average
disk
gaussian
laplacian

Averaging filter
Circular averaging filter (pillbox)
Gaussian lowpass filter
Approximates the two-dimensional Laplacian
operator
Laplacian of Gaussian filter
Approximates the linear motion of a camera
Prewitt horizontal edge-emphasizing filter
Sobel horizontal edge-emphasizing filter

log
Motion
Prewitt
Sobel

Jorma Kekalainen

Digital Image Processing

1166

Attempts at removing motion blur

To deblur the image, we need to divide its transform by the
transform corresponding to the blur filter.
This means that we first must create a matrix corresponding
to the transform of the blur:
m2=zeros(256,256);
m2(1,1:11)=m;
mf=fft2(m2);
Now we can attempt to divide by this transform.
bmi=ifft2(fft2(bm)./mf);
fftshow(bmi,'abs')
and the result is shown in the following figure.
Jorma Kekalainen

Digital Image Processing

1167

Attempts at removing motion blur

As with inverse
filtering, the result
is not particularly
good, because the
values close to zero
in the matrix mf
have tended to
dominate the
result.

Jorma Kekalainen

Straight division

Digital Image Processing

1168

Attempts at removing motion blur

As above, we can constrain the division by only dividing by
values which are above a certain threshold e.g.:
d=0.02;
mf=fft2(m2);mf(find(abs(mf)<d))=1;
bmi=ifft2(fft2(bm)./mf);
imshow(mat2gray(abs(bmi))*2)
where the last multiplication by 2 just brightens the result,
which is shown in the following figure.

Jorma Kekalainen

Digital Image Processing

1169

Attempts at removing motion blur

The writing,
especially in the
centre of the image,
is now quite legible.

Jorma Kekalainen

Constrained division

Digital Image Processing

1170

Compare
Original and
blurred

Jorma Kekalainen

Straight and constrained division

Digital Image Processing

1171

Example

%Suppose we take an image and blur it by a small amount.

bc=imread('board.tif');
bg=im2uint8(rgb2gray(bc));
b=bg(100:355,30:285);
figure, imshow(b)
title('Take the color image and make a grayscale version of data type uint8')
xlabel('and pick out a square subimage')
m=fspecial('motion',11,0);% Blurring filter
bm=imfilter(b,m);
figure, imshow(bm)
title('Blurred image with filter fspecial(''motion'',11,0)')
%To deblur the image, we need to divide its transform by the transform corresponding to the blur filter
m2=zeros(256,256);
m2(1,1:11)=m;%
mf=fft2(m2);% First create a matrix corresponding to the transform of the blur
%Now we can attempt to divide by this transform.
bmi=ifft2(fft2(bm)./mf);
figure, fftshow(bmi,'abs')
title('Recovery attempt of the original image through straight division')
d=0.02;% Discrimination threshold
mf=fft2(m2);mf(find(abs(mf)<d))=1;
bmi=ifft2(fft2(bm)./mf);
figure, imshow(mat2gray(abs(bmi))*2);% The last multiplication by 2 just brightens the result
title(['Constrained division using threshold d = ',num2str(d)])

1175

Colormap
Colormaps are associated with axes and
figures.
Use the colormap function to change the
colormap for a particular axes or figure.
E.g., create a surface plot and use mymap as
the colormap for the figure
surf(peaks)
colormap(mymap)
Jorma Kekalainen

Digital Image Processing

1176

Example

mixing
The additive combination of
red, green and blue lights.

The subtractive combination of

yellow, purple and cyan pigments.

Note: The mixing of pigments is subtractive because it depends on the lights reflected
by those pigments (i.e., on the wavelengths that are not absorbed or subtracted by
Jorma Kekalainen
Digital Image Processing
1185
them)

Absorption of light by
the cones in the human eye

Jorma Kekalainen

Digital Image Processing

1186

Primary and secondary colors of

light (Additive color mixing)
Here, secondary colors are mixtures of two primary colors.
yellow = red + green
cyan = green + blue
magenta = red + blue
CRT
LCD
plasma

Jorma Kekalainen

Digital Image Processing

1187

Primary and secondary colors of

pigments (Subtractive color mixing)
A primary color of pigment absorbs one primary color of light
and reflects the others.
red = yellow + magenta
green = cyan + yellow
blue = magenta + cyan
Painting colors

Color printing is a mixture

of additive and subtractive
color mixing
Jorma Kekalainen

Digital Image Processing

1188

Color matching experiment

The amounts of red, green, and blue which make
up a given color can be determined by a color
matching experiment.
In such an experiment, people are asked to match
a given color (a color source) with different
amounts of the additive primaries red, green and
blue.
Such an experiment was performed for the first
time in 1931 by the CIE (Commission
Internationale d'Eclairage), an organization
responsible for color standards.
Jorma Kekalainen

Digital Image Processing

1189

RGB color matching functions

(CIE, 1964)
Note that for some
wavelengths, various
of the red, green or
blue values are
negative.
This is a physical
impossibility, but it
can be interpreted by
adding the primary
beam to the color
source, to maintain a
color match.

Jorma Kekalainen

Digital Image Processing

1190

XYZ RGB
To remove negative values from color information, the CIE
introduced the XYZ color model.
The values of X, Y and Z can be obtained from the
corresponding R, G and B values by a linear transformation:

The inverse transformation is easily obtained by inverting the

matrix:

Jorma Kekalainen

Digital Image Processing

1191

Exercise
Show with the help of Mablab that the
previous coefficient matrices of the color
model pair XYZRGB are the inverse
matrices.

Jorma Kekalainen

Digital Image Processing

1192

(1)
x
=

X +Y + Z

(2)
y
=

X +Y + Z

z = X + Y + Z (3)

x + y + z = 1 (4 )
Dividing equation (1) by equation (2 )

X x
=
Y
y

x
Y
y

Dividing equation (3) by equation (2 )

Jorma Kekalainen

Z z
=
Y y

z
Z= Y
y

z =1 x y

Digital Image Processing

1 x y
Y
y
1203

Exercise
Show that from equations
x+y+z=1

we obtain the normalized tristimulus values x, y, and

z.
Jorma Kekalainen

Digital Image Processing

1204

Solution
> eq4:=x+y+z=1;

eq4 := x + y + z = 1

> eq5:=X=(x/y)*Y;
eq5 := X =

> eq6:=Z=(1-x-y)*Y/y;
eq6 := Z =

xY
y

( 1 x y) Y
y

> solve({eq4,eq5,eq6},[x,y,z]);
X
Y
Z
x =
,
y
=
,
z
=

X+Y+Z
X+Y+Z
X + Y + Z

>
Jorma Kekalainen

Digital Image Processing

1205

In this model, each color is represented as three
values R, G and B , indicating the amounts of red,
green and blue which make up the color.
This model was originally used for computer
displays where three independent light source
are for the red, green and blue component of
each color.
We may imagine all the colors sitting inside a
color cube.
Jorma Kekalainen

Digital Image Processing

1211

RGB color cube

Backsides

Jorma Kekalainen

R=0

G=0

B=0

Digital Image Processing

1212

RGB color cube

The colors along the black-white diagonal, shown in the
diagram as a dotted line, are the points of the space where
all the R, G, B values are equal.
They are the different intensities of gray.
RGB is the standard for the display of colors on computer
monitors and TV sets.
But it is not a very good way of describing colors.
How, for example, would you define light brown using RGB?
Note also from the previous RGB color matching functions
that some colors require negative values of R, G or B.
These colors are not realizable on a computer monitor or
TV set, on which only positive values are possible.
Jorma Kekalainen

Digital Image Processing

1213

RGB gamut
The set of colors that can be represented
using a particular color space is called its
gamut (spectrum, range of colors, color
space)
The colors corresponding to positive values
form the RGB gamut; in general a color gamut
consists of all the colors realizable with a
particular color model.
Note: In color reproduction, including computer graphics and photography,
the gamut, or color gamut, is a certain complete subset of colors. The most
common usage refers to the subset of colors which can be accurately
represented in a given circumstance, such as within a given color space or by a
Jorma Kekalainen
Digital Image Processing
1214
certain
output device.

RGB gamut
To define the gamut, we shall create a
100*100*3 array, and to each point (i,j) in the
array, associate an XYZ triple defined by
(i/100, j/100, 1- i/100 - j/100).
We can then compute the corresponding RGB
triple, and if any of the RGB values are
negative, make the output value white.
This is done with the few commands.
Jorma Kekalainen

Digital Image Processing

1215

Computing the RGB gamut

load OmaCIExyz64.mat; % CIE data

x2r=[3.063 -1.393 -0.476;-0.969 1.876 0.042;0.068 -0.229 1.069];
for i=1:100,
for j=1:100,
cg(i,j,:)=x2r*[j/100 i/100 1-i/100-j/100]';
if min(cg(i,j,:))<0,
cg(i,j,:)=[1 1 1];
end;
end;
end;
figure, imshow(cg),line(x',y'),axis square,axis xy

title('RGB gamut')
Jorma Kekalainen

Digital Image Processing

1216

RGB gamut

Jorma Kekalainen

Digital Image Processing

1217

Chromaticity diagram and gamut

load OmaCIExyz64.mat; % CIE data

xyz=OmaCIExyz64(:,2:4);
xy=xyz./(sum(xyz')'*[1 1 1]);
x=100*xy(:,1)';
y=100*xy(:,2)';
figure,plot([x x(1)],[y y(1)]),axis square
hold on
xlabel('X coordinate'),ylabel('Y coordinate')
%
x2r=[3.063 -1.393 -0.476;-0.969 1.876 0.042;0.068 -0.229 1.069];
cg=zeros(100,100,3);
for i=1:100,
for j=1:100,
cg(i,j,:)=x2r*[j/100 i/100 1-i/100-j/100]';
if min(cg(i,j,:))<0,
cg(i,j,:)=[1 1 1];
end;
end;
end;
imshow(cg),line([x x(1)],[y y(1)]),axis square,axis xy,axis on,axis tight
title('Chromaticity diagram and RGB gamut')

Jorma Kekalainen

Digital Image Processing

1218

Chromaticity diagram and gamut

Jorma Kekalainen

Digital Image Processing

1219

A typical gamut
The grayed-out horseshoe
shape is the entire range of
possible chromaticities,
displayed in the CIE 1931
chromaticity
diagram format.
The colored triangle is the
gamut available to a typical
computer monitor; it does
not cover the entire space.
The corners of the triangle
are the primary colors for
this gamut; in the case of
a CRT, they depend on the
colors of the phosphors of
the monitor.
Jorma Kekalainen

Digital Image Processing

1220

Hue, Saturation, Value (HSV)

HSV color model terms have the following
meanings:
Hue: The true color attribute (red, green, blue,
orange, yellow, and so on).
Saturation: The amount by which the color has
been diluted with white.
The more white in the color, the lower the saturation.
So a deep red has high saturation, and a light red (a
pinkish color) has low saturation.

Value: The degree of brightness: a bright color

has high intensity; a dark color has low intensity.
Jorma Kekalainen

Digital Image Processing

1221

HSV (Hue, Saturation, Value)

This model is a quite natural method of
describing colors.
Since the intensity (value) is independent
of the color information, this is a very
useful model for image processing.
We can visualize this model as a cone.
Any point on the surface represents a
purely saturated color.
The saturation is thus given as the
relative distance to the surface from the
central axis of the structure.
Hue is defined to be the angle
measurement from a pre-determined
axis.
Jorma Kekalainen
Digital Image Processing

1222

Illustration of the HSV color

Conversion between RGB and HSV

Suppose a color is specified by its RGB values.
If all the three values are equal, then the color
will be a gray scale; that is, an intensity of white.
Such a color, containing just white, will thus have
a saturation of zero.
Conversely, if the RGB values are very different,
we would expect the resulting color to have a
high saturation.
In particular, if one or two of the RGB values are
zero, the saturation will be one, the highest
possible value.
Jorma Kekalainen

Digital Image Processing

1228

Conversion between RGB and HSV

Hue is defined as the fraction around the circle starting from
red, which thus has a hue of zero.
Reading around the circle produces the following hues:
Color

Hue

Red

Yellow

1/6

Green

1/3

Cyan

1/2

Blue

H=(4+(0.1-0.3)/0.4)/6= 0.5833
The result is
(R,G,B)=(0.1,0.3,0.5) (0.5833,0.8,0.5)=(H,S,V)
Jorma Kekalainen

Digital Image Processing

1232

Example

(R,G,B)=(0.1,0.3,0.5) (0.5833,0.8,0.5)=(H,S,V)

Checking:
>> hsv=rgb2hsv([0.1,0.3,0.5])
hsv =
0.5833 0.8000 0.5000
Checking also other way: >> rgb=hsv2rgb(hsv)
rgb =
0.1000 0.3000 0.5000
Jorma Kekalainen

1236

YIQ
This color space was originally used for TV/video in America
and other countries where NTSC was the video standard.
In this scheme Y is the luminance (this corresponds roughly
with intensity), and I and Q carry the color information
(chrominance).
The conversion between RGB is straightforward:

and

Jorma Kekalainen
Digital Image Processing
The
two conversion matrices are of course
inverses of each other.

1237

Exercise
Using Matlab show that the two previous
coefficient matrices are inverses of each other.

Jorma Kekalainen

Digital Image Processing

1238

Solution

>> %[YIQ]=[C]*[RGB]
C=[0.299,0.587,0.114;0.596,-0.274,-0.322;0.211,-0.523,0.312]
C=
0.2990 0.5870 0.1140
0.5960 -0.2740 -0.3220
0.2110 -0.5230 0.3120
>> invC=inv(C)
invC =
1.0000 0.9562 0.6214
A=
1.0000 -0.2727 -0.6468
0.2990 0.5870 0.1140
1.0000 -1.1037 1.7006
0.5960 -0.2740 -0.3220
>> inv(invC)
0.2110 -0.5230 0.3120
ans =
>> inv(A)
0.2990 0.5870 0.1140
0.5960 -0.2740 -0.3220
ans =
0.2110 -0.5230 0.3120
1.0000 0.9562 0.6214
>>

>> rgb2ntsc([1,1,1])
ans =
1 0 0
>> rgb2ntsc([0.5,0.5,0.5])
ans =
1241
0.5000
0
0

Note

Jorma Kekalainen

>> rgb2ntsc([1,1,1])
ans =
1 0 0
>> rgb2ntsc([0.5,0.5,0.5])
ans =
0.5000
0
0
>> ntsc2rgb([0.5,0.5,0.5])
ans =
1.0000 0.0314 0.6197
>> ntsc2rgb([1,1,1])
ans =
1.0000 0.0314 0.6197
>> ntsc2rgb([1,0,0])
ans =
1 1 1
>> ntsc2rgb([0.5,0,0])
ans =
0.5000 0.5000
0.5000
Digital Image Processing

1242

rgb2ntsc
Convert RGB color values to NTSC color space
Syntax:
yiqmap = rgb2ntsc(rgbmap)
YIQ = rgb2ntsc(RGB)

Description:
yiqmap = rgb2ntsc(rgbmap) converts the m-by-3 RGB
values in rgbmap to NTSC color space.
yiqmap is an m-by-3 matrix that contains the NTSC
luminance (Y) and chrominance (I and Q) color
components as columns that are equivalent to the colors
in the RGB colormap.
YIQ = rgb2ntsc(RGB) converts the truecolor image RGB to
the equivalent NTSC image YIQ.
Jorma Kekalainen

Digital Image Processing

1243

rgb2ntsc
Class Support
RGB can be of class uint8, uint16, int16, single, or double.
rgbmap can be double.
The output is double.

Note:
In the NTSC color space, the luminance is the grayscale
signal used to display pictures on monochrome (black and
white) televisions.
The other components carry the hue and saturation
information.
rgb2ntsc defines the NTSC components using

Jorma Kekalainen

Digital Image Processing

1244

ntsc2rgb
Convert NTSC values to RGB color space
Syntax:
rgbmap = ntsc2rgb(yiqmap)
RGB = ntsc2rgb(YIQ)

Description:
rgbmap = ntsc2rgb(yiqmap) converts the m-by-3 NTSC
(television) color values in yiqmap to RGB color space.
If yiqmap is m-by-3 and contains the NTSC luminance (Y) and
chrominance (I and Q) color components as columns, then
rgbmap is an m-by-3 matrix that contains the red, green, and
blue values equivalent to those colors.
Both rgbmap and yiqmap contain intensities in the range 0 to
1.0.
The intensity 0 corresponds to the absence of the component,
while the intensity 1.0 corresponds to full saturation of the
component.
Jorma Kekalainen
Digital Image Processing
1245

ntsc2rgb
RGB = ntsc2rgb(YIQ) converts the NTSC image YIQ
to the equivalent truecolor image RGB.
ntsc2rgb computes the RGB values from the NTSC
components using

Class Support:
The input image or colormap must be of class double.
The output is of class double.
Jorma Kekalainen

Digital Image Processing

1246

Examples

Convert RGB image to NTSC and back.

RGB = imread('board.tif');
NTSC = rgb2ntsc(RGB);
RGB2 = ntsc2rgb(NTSC);
figure,imshow(RGB)
title('RGB')
figure,imshow(NTSC)
title('NTSC')
figure,imshow(RGB2)
title('RGB2')

Jorma Kekalainen

Digital Image Processing

1247

Example

Jorma Kekalainen

Digital Image Processing

1248

components

%Suppose we read in an RGB image:

f=imread('flowers.tif');
size(f)
%We can isolate each color component by the colon operator:
f(:,:,1); %The first, or red component
f(:,:,2); %The second, or green component
f(:,:,3); %The third, or blue component
%These can all be viewed with imshow:
imshow(f)
figure,imshow(f(:,:,1))
figure,imshow(f(:,:,2))
figure,imshow(f(:,:,3))

Jorma Kekalainen

Digital Image Processing

1252

RGB color image and its

Color image
components
Note how the colors with particular hues show up
with high intensities in their respective components.
For the bright flowers the red component shows a very high
intensity. The green and blue components show much lower
intensities. Similarly the green leaves show slightly higher
intensity in the green component than the other two.

Red component

Jorma Kekalainen

Green component

Digital Image Processing

Blue component

1253

RGB HSV
We can convert RGB to HSV and view the
components again:
fh=rgb2hsv(f);
figure, imshow(fh)
figure,imshow(fh(:,:,1))
figure,imshow(fh(:,:,2))
figure,imshow(fh(:,:,3))
Jorma Kekalainen

Digital Image Processing

1254

RGB HSV

Hue

Saturation

Jorma Kekalainen

Digital Image Processing

Value

1255

RGB YIQ
We can do precisely the same thing for the
YIQ color space:
fn=rgb2ntsc(f);
figure,imshow(fn)
figure,imshow(fn(:,:,1))
figure,imshow(fn(:,:,2))
figure,imshow(fn(:,:,3))
Jorma Kekalainen

Digital Image Processing

1256

YIQ components

Jorma Kekalainen

Digital Image Processing

1257

Note
The National Television Systems Committee (NTSC) has
defined a color space known as YIQ.
One of the main advantages of this format is that
grayscale information is separated from color data, so
the same signal can be used for both color and black
and white sets.
In the NTSC color space, image data consists of three
components: luminance (Y), hue (I), and saturation
(Q).
The first component, luminance, represents grayscale
information, while the last two components make
up chrominance (color information).
Jorma Kekalainen

Digital Image Processing

1258

Note
The function rgb2ntsc converts colormaps of RGB images to the
NTSC color space and ntsc2rgb performs the reverse operation.
For example, these commands convert an RGB image to NTSC
format.
RGB = imread('peppers.png');
YIQ = rgb2ntsc(RGB);

Because luminance is one of the components of the NTSC format,

the RGB to NTSC conversion is also useful for isolating the gray level
information in an image.
In fact, the toolbox functions rgb2gray and ind2gray use the
rgb2ntsc function to extract the grayscale information from a color
image.
For example, these commands are equivalent to calling rgb2gray.
YIQ = rgb2ntsc(RGB);
I = YIQ(:,:,1);
Note:
InKekalainen
the YIQ color space, I is Digital
oneImage
of the
two color components, not the
Jorma
Processing
1259
grayscale component.

Example

RGB = imread('peppers.png');
figure,imshow(RGB)
title('RGB')
YIQ = rgb2ntsc(RGB);
figure,imshow(YIQ)
title('YIQ')
I = YIQ(:,:,1);
figure,imshow(I)
title('Y component')
figure,imshow(rgb2gray(RGB))
title('rgb2gray(RGB)')

Jorma Kekalainen

Digital Image Processing

1260

Example

Digital Image Processing

1265

Intensity slicing
In this method, we break up the image into various gray level
ranges.
We simply assign a different color to each range.
For example:
We can consider this as a mapping

Digital Image Processing

1269

Jorma Kekalainen

Digital Image Processing

1276

Example

[x,map]=imread('palikat.png');
figure,imshow(x,map)
title('Original image')
b=uint8(ind2gray(x,map));
figure,imshow(b)
title('Image converted grayscale')
figure, imshow(b,colormap(jet(256)))
title('Image colormapped ''colormap(jet(256))''')
b16=grayslice(b,16);
figure,imshow(b16,colormap(vga))
title('Image colormapped ''colormap(vga)''')
mycolormap=[0 0 1;1 0 1;0 1 0;1 0 0];
b4=grayslice(b,4);
figure,imshow(b4,mycolormap)
title('Image colormapped ''mycolormap''')

Jorma Kekalainen

Digital Image Processing

1277

Jorma Kekalainen

Digital Image Processing

1278

imfinfo('palikat.png')
>> imfinfo('palikat.png')
ans =

Filename: 'C:\Users\ktkekjo\Documents\MATLAB\palikat.png'

FileModDate: '19-Aug-2015 01:49:20'

FileSize: 39932

Format: 'png'

FormatVersion: []

Width: 307

Height: 308

BitDepth: 8

ColorType: 'indexed'

FormatSignature: [137 80 78 71 13 10 26 10]

Colormap: [256x3 double]

Histogram: []

Digital Image Processing

1283

Example
h = ones(7,7)/49
h=
0.0204 0.0204
0.0204 0.0204
0.0204 0.0204
0.0204 0.0204
0.0204 0.0204
0.0204 0.0204
0.0204 0.0204

Original image

Jorma Kekalainen

0.0204
0.0204
0.0204
0.0204
0.0204
0.0204
0.0204

h = ones(7,7)/49

Digital Image Processing

0.0204
0.0204
0.0204
0.0204
0.0204
0.0204
0.0204

h = ones(30,30)/900

1284

Contrast enhancement
Contrast enhancement is best done by processing
the intensity component.
Suppose we start with the image forest.tif, which
is an indexed color image, and convert it to a
truecolour (RGB) image.
[x,map]=imread('forest.tif');
c=ind2rgb(x,map);

Now we have to convert from RGB to YIQ, so as

to be able to isolate the intensity component:
cn=rgb2ntsc(c);
Jorma Kekalainen

Digital Image Processing

1285

RGB YIQ

Jorma Kekalainen

Digital Image Processing

1286

Histogram equalization to the

intensity component
Now we apply histogram Intensity processing
equalization to the
intensity component, and
convert back to RGB for
display:
cn(:,:,1)=histeq(cn(:,:,1));
c2=ntsc2rgb(cn);
imshow(c2)

The result is shown beside.

Jorma Kekalainen

Whether this is an improvement is

debatable, but contrast has enhanced and
Digital Image Processing
the green color has nearly washed1287
out

Intensity components before and

after histogram equalization

Jorma Kekalainen

Digital Image Processing

1288

imfinfo('forest.tif')

>> imfinfo('forest.tif')

ans =

Filename: [1x64 char]

FileModDate: '04-joulu-2000 14:57:58'

FileSize: 124888

Format: 'tif'

FormatVersion: []

Width: 447

Height: 301

BitDepth: 8

ColorType: 'indexed'

FormatSignature: [73 73 42 0]

ByteOrder: 'little-endian'

NewSubFileType: 0

BitsPerSample: 8

Compression: 'PackBits'

PhotometricInterpretation: 'RGB Palette'

StripOffsets: [1x17 double]

SamplesPerPixel: 1

RowsPerStrip: 18

StripByteCounts: [1x17 double]

XResolution: 72

YResolution: 72

ResolutionUnit: 'Inch'
Jorma Kekalainen Colormap: [256x3 double]
Digital Image Processing

1289

Example
But suppose we try to apply histogram equalization to each of
the RGB components:
cr=histeq(c(:,:,1));
cg=histeq(c(:,:,2));
cb=histeq(c(:,:,3));

Now we have to put them all back into a single 3 dimensional

array for use with imshow.
The cat function is what we want:
c3=cat(3,cr,cg,cb);
imshow(c3)

Note: cat function concatenates arrays along specified dimension

C = cat(dim, A1, A2, A3, A4, ...)
Digital Image Processing
1290
The Jorma
firstKekalainen
variable to cat is the dimension
along which we want our arrays to be joined.

1293

Example
Original RGB image

Colors are considerably

washed out

Jorma Kekalainen

Digital Image Processing

Some false colors have

been introduced

1294

Spatial filtering
It depends on the filter as to which schema we use.
For a low pass filter, say a blurring filter, we can apply the
filter to each RGB component:
Low pass filtering
a15=fspecial('average',15);
cr=filter2(a15,c(:,:,1));
cg=filter2(a15,c(:,:,2));
cb=filter2(a15,c(:,:,3));
blur=cat(3,cr,cg,cb);
figure,imshow(blur)
The result is shown beside.
Jorma Kekalainen

Digital Image Processing

1295

Digital Image Processing

1299

Example

Jorma Kekalainen

Digital Image Processing

1300

Jorma Kekalainen

Digital Image Processing

1301

Conclusion: Spatial filtering of a

color image
In general, we will obtain reasonable results
using the intensity component only.
Although we can apply a filter to each of the
RGB components, as we did for the example
above, we cannot be guaranteed a good
result.
The problem is that any filter will change the
values of the pixels, and this may introduce
unwanted colors.
Jorma Kekalainen

Digital Image Processing

1302

Noise reduction
Now we use the image pears.png:
p=imread('pears.png');
We can add noise, and look at the noisy image, and its RGB
components:
pn=imnoise(p,'salt & pepper');
%About 5% of pixel corrupted (default value)
figure,imshow(pn)
figure,imshow(pn(:,:,1))
figure,imshow(pn(:,:,2))
figure,imshow(pn(:,:,3))
Jorma Kekalainen

Digital Image Processing

1303

Original and noisy image

Jorma Kekalainen

Digital Image Processing

1304

Noise on a color image

Salt & pepper
noise

Red
component

Green
component

Blue
component

Jorma Kekalainen

Digital Image Processing

1305

Denoising each RGB component

We can apply median filtering to each of the
RGB components:
prm=medfilt2(pn(:,:,1));
pgm=medfilt2(pn(:,:,2));
pbm=medfilt2(pn(:,:,3));
pm=cat(3,prm,pgm,pbm);
figure,imshow(pm)

The result is beside.

Jorma Kekalainen

Digital Image Processing

1306

Jorma Kekalainen

Digital Image Processing

1310

Reduction of heavy noise heavy noise

Jorma Kekalainen

Digital Image Processing

1311

Example: Reduction of heavy noise

p=imread('pears.png');

figure, imshow(p)

title('Original image')

d=0.2;% Noise increased to d*100%

pn=imnoise(tw,'salt & pepper',d);

figure,imshow(tn)

title(['Noisy image (',num2str(d*100),')% from pixel corrupted'])

figure,imshow(pn(:,:,1))

title('Red component')

figure,imshow(pn(:,:,2))

title('Green component')

figure,imshow(pn(:,:,3))

title('Blue component')

% Apply median filtering to each of the RGB components

prm=medfilt2(pn(:,:,1));

pgm=medfilt2(pn(:,:,2));

pbm=medfilt2(pn(:,:,3));

pm=cat(3,prm,pgm,pbm);

figure,imshow(pm)

title('Median filtered each of the RGB components')

% We remove the noise from Y only

pnn=rgb2ntsc(pn);

pnn(:,:,1)=medfilt2(pnn(:,:,1));

pm2=ntsc2rgb(pnn);
Jormafigure,imshow(pm2)
Kekalainen
Digital Image Processing

title('Median filtered luminance (Y) component only')

1312

Note
f=imread('flowers.tif');
figure,imshow(f)
title('Original image ''flowers.tif''')
% 20% of pixel corrupted by noise
fsp20=imnoise(f,'salt & pepper',0.2);% 20% of pixel corrupted by noise
figure, imshow(fsp20)
title('Image with 20% of its pixels corrupted by binary noise')
fsp20m3r=medfilt2(fsp20(:,:,1));% Red component
fsp20m3g=medfilt2(fsp20(:,:,2));
fsp20m3b=medfilt2(fsp20(:,:,1));
fsp20m3=cat(3,fsp20m3r, fsp20m3g, fsp20m3b);% all components back into
a single 3D array for use with imshow
figure, imshow(fsp20m3)
title('Image filtered by 3*3 median filter')
Jorma Kekalainen
Digital Image Processing
1313
xlabel('Image with 20% of its pixels
corrupted by binary noise')

Note
f=imread('flowers.tif');
figure,imshow(f)
title('Original image ''flowers.tif''')
% 20% of pixel corrupted by noise
fsp20=imnoise(f,'salt & pepper',0.2);% 20% of pixel corrupted by noise
figure, imshow(fsp20)
title('Image with 20% of its pixels corrupted by binary noise')
a3=fspecial('average');
fsp20a3r=filter2(a3,fsp20(:,:,1));%Red component
fsp20a3g=filter2(a3,fsp20(:,:,2));
fsp20a3b=filter2(a3,fsp20(:,:,3));
fsp20a3=cat(3,fsp20a3r, fsp20a3g, fsp20a3b);% all components back into a
single 3D array for use with imshow
Jorma Kekalainen

Digital Image Processing

Applying the edge function to the

intensity component only
To implement the first
method, we start with the
rgb2gray function:

fe1: Edges after rgb2gray

fg=rgb2gray(f);
fe1=edge(fg);
figure,imshow(fe1)

Recall that edge with no

parameters implements
Sobel edge detection.
The result is shown beside.

Jorma Kekalainen

Digital Image Processing

1319

Applying the edge function to

each of the RGB components
For the second method, we can join the
results with the logical or:
f1=edge(f(:,:,1));
f2=edge(f(:,:,2));
f3=edge(f(:,:,3));
fe2=f1 | f2 | f3;
figure,imshow(fe2)

Result is shown beside.

Jorma Kekalainen

Digital Image Processing

1320

Example

% Edge detection (Sobel method)

% An edge image will be a binary image containing the edges of the input image.
% We can go about obtaining an edge image in two ways
% To implement the first method, we start with the rgb2gray function
f=imread('pep.jpg');
figure,imshow(f)
title('Original')
fg=rgb2gray(f);% First method: Take the intensity component only,
fe1=edge(fg);% and apply the edge function to it
figure,imshow(fe1)
title('Edge applied to the intensity component only taking first rgb2gray conversion')
% Note edge with no parameters implements Sobel edge detection
% Second method: Apply the edge function to each of the RGB components
f1=edge(f(:,:,1));
f2=edge(f(:,:,2));
f3=edge(f(:,:,3));
fe2=f1 | f2 | f3;% Join the results with the logical "or"
figure,imshow(fe2)
title('Edge function is applied to each of the RGB components, and joined the results')

Jorma Kekalainen

Digital Image Processing

1321

Comment
The edge image fe2 is a
much more complete
edge image than fe1.

Jorma Kekalainen

Digital Image Processing

1322

Example

% Edge detection (Canny method)

% An edge image will be a binary image containing the edges of the input image.
% We can go about obtaining an edge image in two ways
% To implement the first method, we start with the rgb2gray function
f=imread('pep.jpg');
figure,imshow(f)
title('Original')
fg=rgb2gray(f);% First method: Take the intensity component only,
fe1=edge(fg, 'canny');% and apply the edge function to it
figure,imshow(fe1)
title('Edge applied to the intensity component only taking first rgb2gray conversion')
% Note edge with no parameters implements Sobel edge detection
% Second method: Apply the edge function to each of the RGB components
f1=edge(f(:,:,1), 'canny');
f2=edge(f(:,:,2), 'canny');
f3=edge(f(:,:,3), 'canny');
fe2=f1 | f2 | f3;% Join the results with the logical "or"
figure,imshow(fe2)
title('Edge function is applied to each of the RGB components, and joined the results')

Jorma Kekalainen

Digital Image Processing

1323

As a reference: Canny method

Jorma Kekalainen

Digital Image Processing

1324

Exam Up to This
Effective Image File
Compression Methods

JSS 5855-11-2016
100% (1)
JSS 5855-11-2016
28 pages
The Film Developing Cookbook by Bill Troop, Steve Anchell
100% (1)
The Film Developing Cookbook by Bill Troop, Steve Anchell
547 pages
Basics of Image Processing and Analysis
No ratings yet
Basics of Image Processing and Analysis
175 pages
Digital Image Processing Solution of Homework
No ratings yet
Digital Image Processing Solution of Homework
7 pages
Telecommunication Systems by Jorma Kekalainen
No ratings yet
Telecommunication Systems by Jorma Kekalainen
520 pages
Dimond Iii: Image Quality and Dose Management For Digital Radiography
No ratings yet
Dimond Iii: Image Quality and Dose Management For Digital Radiography
137 pages
Image Segmentation: Unlocking Insights through Pixel Precision
From Everand
Image Segmentation: Unlocking Insights through Pixel Precision
Fouad Sabry
No ratings yet
Digital Image Processing by Annapurna Mishra 327ed4
No ratings yet
Digital Image Processing by Annapurna Mishra 327ed4
208 pages
EC2029-Digital Image Processing Two Marks Questions and Answers - New PDF
No ratings yet
EC2029-Digital Image Processing Two Marks Questions and Answers - New PDF
20 pages
Image Compression
No ratings yet
Image Compression
14 pages
Deep-Learning Based Lossless Image Coding
No ratings yet
Deep-Learning Based Lossless Image Coding
14 pages
Image Processing
No ratings yet
Image Processing
39 pages
Image Enhancement Techniques
No ratings yet
Image Enhancement Techniques
15 pages
Eng4Bf3 Medical Image Processing: Image Enhancement in Frequency Domain
No ratings yet
Eng4Bf3 Medical Image Processing: Image Enhancement in Frequency Domain
59 pages
Image Compression Using DCT
100% (1)
Image Compression Using DCT
10 pages
Image Enhancement Frequency Domain
100% (2)
Image Enhancement Frequency Domain
42 pages
Histogram Specification
No ratings yet
Histogram Specification
6 pages
Assignment-1 Digital Image Processing
No ratings yet
Assignment-1 Digital Image Processing
8 pages
Image Processing Frequency
No ratings yet
Image Processing Frequency
55 pages
Image Enhancement
No ratings yet
Image Enhancement
89 pages
Image Restoration
91% (11)
Image Restoration
91 pages
Image Descriptor
100% (1)
Image Descriptor
53 pages
Image Processing QB
100% (1)
Image Processing QB
29 pages
Image and Video Processing Klu
No ratings yet
Image and Video Processing Klu
1 page
3d Visualization PDF
No ratings yet
3d Visualization PDF
24 pages
National Institute of Technology Rourkela: Answer All Questions. Figures in The Right Hand Margin Indicate Marks
No ratings yet
National Institute of Technology Rourkela: Answer All Questions. Figures in The Right Hand Margin Indicate Marks
3 pages
Digital Image Processing
No ratings yet
Digital Image Processing
129 pages
Question Bank
No ratings yet
Question Bank
37 pages
AI Neuromorphic-Computing
0% (1)
AI Neuromorphic-Computing
40 pages
Frequency Domain Filters
No ratings yet
Frequency Domain Filters
43 pages
DIP02-Image Enhancement-Point Processing
No ratings yet
DIP02-Image Enhancement-Point Processing
34 pages
Ec8093 Dip - Question Bank With Answers
No ratings yet
Ec8093 Dip - Question Bank With Answers
189 pages
Image Segmentation in Digital Image Processing
No ratings yet
Image Segmentation in Digital Image Processing
71 pages
Image Restoration
No ratings yet
Image Restoration
59 pages
Digital Image Processing: Image Restoration: Noise Removal
No ratings yet
Digital Image Processing: Image Restoration: Noise Removal
31 pages
Assignment ENT 413
No ratings yet
Assignment ENT 413
3 pages
Unit 5 Degradation and Restoration Noise Filter
No ratings yet
Unit 5 Degradation and Restoration Noise Filter
108 pages
Image Processing & Computer Vision With MATLAB 2013
No ratings yet
Image Processing & Computer Vision With MATLAB 2013
71 pages
Digital Image Processing Notes
No ratings yet
Digital Image Processing Notes
94 pages
Digital Image Processing - Lecture Notes
0% (1)
Digital Image Processing - Lecture Notes
32 pages
Digital Image Processing: Elements of Visual Perception
No ratings yet
Digital Image Processing: Elements of Visual Perception
31 pages
15EC72 - Digital Image Processing 2018-19 PDF
No ratings yet
15EC72 - Digital Image Processing 2018-19 PDF
62 pages
Image Processing 4-ImageEnhancement (PointProcessing)
No ratings yet
Image Processing 4-ImageEnhancement (PointProcessing)
42 pages
FFT Algorithm Implement in C
No ratings yet
FFT Algorithm Implement in C
9 pages
Applied Digital Imaging Image Processing For Embedded Devices by Sebastiano Battiato, Arcangelo Ranieri Bruna, Giuseppe Messina and Giovanni Puglisi PDF
100% (1)
Applied Digital Imaging Image Processing For Embedded Devices by Sebastiano Battiato, Arcangelo Ranieri Bruna, Giuseppe Messina and Giovanni Puglisi PDF
388 pages
Chapter 3. Intensity Transformations and Spatial Filtering
No ratings yet
Chapter 3. Intensity Transformations and Spatial Filtering
69 pages
Ch5 Restoration
No ratings yet
Ch5 Restoration
56 pages
ECE - DIP - Unit 4
No ratings yet
ECE - DIP - Unit 4
43 pages
Image Processing-Chapter 1
No ratings yet
Image Processing-Chapter 1
8 pages
Chapter 6. Image Segmentation
No ratings yet
Chapter 6. Image Segmentation
83 pages
Wavelet Transform
100% (1)
Wavelet Transform
41 pages
PIP Unit 3 Image Segmentation
No ratings yet
PIP Unit 3 Image Segmentation
54 pages
Computer Graphics Viva Question
No ratings yet
Computer Graphics Viva Question
8 pages
Image Processing 2nd Edition WWW EBooksWorld Ir PDF
100% (2)
Image Processing 2nd Edition WWW EBooksWorld Ir PDF
816 pages
Image Enhancement-Spatial Filtering From: Digital Image Processing, Chapter 3
No ratings yet
Image Enhancement-Spatial Filtering From: Digital Image Processing, Chapter 3
56 pages
Digital Image Processing: Huffman Coding Example
No ratings yet
Digital Image Processing: Huffman Coding Example
3 pages
Digital Image Processing File
No ratings yet
Digital Image Processing File
26 pages
DIP Questions
100% (1)
DIP Questions
5 pages
Ec 1009 - Digital Image Processing
75% (4)
Ec 1009 - Digital Image Processing
30 pages
An Introduction To Image Compression
No ratings yet
An Introduction To Image Compression
58 pages
Image Processing
79% (14)
Image Processing
38 pages
Digital Image Processing
50% (2)
Digital Image Processing
1 page
AI-UNIT-2 PPT
No ratings yet
AI-UNIT-2 PPT
135 pages
Digital Image Processing Fundamentals
88% (8)
Digital Image Processing Fundamentals
43 pages
Telecommunications Engineering II by Jorma Kekalainen
No ratings yet
Telecommunications Engineering II by Jorma Kekalainen
608 pages
Telecommunications Engineering I - Probability Concepts and Stochastic Processes
No ratings yet
Telecommunications Engineering I - Probability Concepts and Stochastic Processes
68 pages
Telecommunications Engineering I - SNR in Analog Communication Systems and Digital Transmission
No ratings yet
Telecommunications Engineering I - SNR in Analog Communication Systems and Digital Transmission
46 pages
Wireless Techniques - Wireless Channel and Radio Wave Propagation
No ratings yet
Wireless Techniques - Wireless Channel and Radio Wave Propagation
251 pages
Telecommunications Engineering I - Fourier Representations and Time-Frequency Domain Relations
No ratings yet
Telecommunications Engineering I - Fourier Representations and Time-Frequency Domain Relations
116 pages
Wireless Techniques - Mobility in IP and Cellular Networks
No ratings yet
Wireless Techniques - Mobility in IP and Cellular Networks
48 pages
Wireless Techniques - Link Layer and Multiple Access
No ratings yet
Wireless Techniques - Link Layer and Multiple Access
119 pages
Wireless Techniques - Introduction To Wireless Networks
No ratings yet
Wireless Techniques - Introduction To Wireless Networks
58 pages
Telecommunication Network Design - Design of Optical Fiber Systems
100% (2)
Telecommunication Network Design - Design of Optical Fiber Systems
37 pages
Switching Systems - Switch Structures
No ratings yet
Switching Systems - Switch Structures
50 pages
Telecommunication Network Design by Jorma Kekalainen
No ratings yet
Telecommunication Network Design by Jorma Kekalainen
574 pages
Telecommunication Network Design - Lecture Weeks 1 and 2
100% (1)
Telecommunication Network Design - Lecture Weeks 1 and 2
37 pages
Telecommunication Network Design - Microwave Link Design
100% (1)
Telecommunication Network Design - Microwave Link Design
86 pages
Digital Image Processing - Lecture Weeks 21 and 22
No ratings yet
Digital Image Processing - Lecture Weeks 21 and 22
55 pages
Digital Image Processing - Lecture Weeks 19 and 20
No ratings yet
Digital Image Processing - Lecture Weeks 19 and 20
54 pages
Digital Image Processing - Lecture Weeks 11 and 12
No ratings yet
Digital Image Processing - Lecture Weeks 11 and 12
60 pages
5273 Ti2655
No ratings yet
5273 Ti2655
4 pages
Winlens 2008
No ratings yet
Winlens 2008
27 pages
Zernike Wavefront Aberrations
No ratings yet
Zernike Wavefront Aberrations
26 pages
Lens and Sensor Compatibility by ScorpionVision
No ratings yet
Lens and Sensor Compatibility by ScorpionVision
10 pages
Article MTF 2008 EN PDF
No ratings yet
Article MTF 2008 EN PDF
33 pages
Catphan412 424manual
No ratings yet
Catphan412 424manual
39 pages
Kawashima 2013
No ratings yet
Kawashima 2013
8 pages
Joseph Goodman - Introduction To Fourier Optics-W. H. Freeman (2017)
No ratings yet
Joseph Goodman - Introduction To Fourier Optics-W. H. Freeman (2017)
545 pages
Digital Image Processing
No ratings yet
Digital Image Processing
24 pages
Tutorial On ISO 10110 Optical Drawing Standards
No ratings yet
Tutorial On ISO 10110 Optical Drawing Standards
18 pages
John G. Webster Editor 09.circuits and Systems For Video Technology
No ratings yet
John G. Webster Editor 09.circuits and Systems For Video Technology
210 pages
MTG Objective NCERT Chemistry
No ratings yet
MTG Objective NCERT Chemistry
7 pages
Panasonic Lumix G 14mm f:2.5 ASPH - Review: Lens Test Report - Analysis
No ratings yet
Panasonic Lumix G 14mm f:2.5 ASPH - Review: Lens Test Report - Analysis
6 pages
Veraview X800 F150 Brochure L-1349 0418 v6 B
No ratings yet
Veraview X800 F150 Brochure L-1349 0418 v6 B
18 pages
VT1730 PDF
No ratings yet
VT1730 PDF
161 pages
Cat Phan 700 Manual
No ratings yet
Cat Phan 700 Manual
42 pages
Matlab Fonction Image Processing
No ratings yet
Matlab Fonction Image Processing
5 pages
Opd Scan III Vse 6p 5
No ratings yet
Opd Scan III Vse 6p 5
6 pages
Comparison of The Optical Behaviour of Five Different Multifocal Diffractive Intraocular Lenses in A Model Eye
No ratings yet
Comparison of The Optical Behaviour of Five Different Multifocal Diffractive Intraocular Lenses in A Model Eye
12 pages
Sui2017 PDF
No ratings yet
Sui2017 PDF
8 pages
Fundamentals of medical imaging 2nd Edition Paul Suetens download pdf
100% (1)
Fundamentals of medical imaging 2nd Edition Paul Suetens download pdf
67 pages
ZEISS Dimension 2 12 Datasheet
No ratings yet
ZEISS Dimension 2 12 Datasheet
5 pages
CGIV 2008 Art00099 Alicia-Fernández-Oliveras
No ratings yet
CGIV 2008 Art00099 Alicia-Fernández-Oliveras
6 pages
Intermediate Optical Design Michael J. Kidger All Chapters Instant Download
100% (12)
Intermediate Optical Design Michael J. Kidger All Chapters Instant Download
85 pages
10 1259@dmfr 20190229
No ratings yet
10 1259@dmfr 20190229
17 pages
Glistening Formation in A New Hydrophobic Acrylic Intraocular Lens
No ratings yet
Glistening Formation in A New Hydrophobic Acrylic Intraocular Lens
7 pages
Vario Sonnar 3 5 4 5 28 70mm e
No ratings yet
Vario Sonnar 3 5 4 5 28 70mm e
4 pages