Introduction To DIP

Image Processing and the
Applications
Rajashekhara
2/22/2009 1
What is an Image?
a representation,
likeness, or imitation
of an object or thing
a vivid or graphic
description
something introduced
to represent
something else
2/22/2009 2
Where are we?
Display/Printing?
Computer
Imaging?
Vision?
Digital Image
Processing
Computer Biological
Graphics? Vision?
2/22/2009 3
Computer vision Vs computer
graphics
Transformation(Computer Graphics)
Image or
3D information
Display
Extraction(Computer Vision)
2/22/2009 4
Computer vision, Computer
graphics, Image processing
Computer vision estimates 3D data from one or
more 2D images.
Computer graphics generates 2D/3D images
from the 3D (from the mathematical
functions) of an object.
Computer vision and Computer graphics are
inverse operations each other. They both use
image processing which is there fore regarded
as low level (or basic) operation for computer
vision and computer graphics.
Note that Computer Vision and Computer
graphics and Image processing are normally
considered as three overlapping areas but
none them are subset of the other
2/22/2009 5
Computer Vision Means
Machine Vision
Robot Vision
Scene Analysis
Image Understanding
Image Analysis
2/22/2009 6
Image processing Means
Image processing refers to a set of

computational techniques which accept
images as input. The results of the processing
can be new images or information extracted
from the input images. Video is just time
sequence of images called frames. All image
processing techniques can be applied to
frames. Image processing has many
applications
2/22/2009 7
Why Image processing?
Why?
– Coding/compression
– Enhancement, restoration, reconstruction
– Analysis, detection, recognition, understanding
– Visualization
2/22/2009 8
What do we do?
Image Processing/
Manipulation
Digital Image
Processing
Image Analysis/ Image Coding/

Interpretation Communication
2/22/2009 9
Digital Image
2/22/2009 10
What is an Image?
a visual representation
of objects, their
components,
properties,
relationships,
Mapping 3D scene on
to a 2D plane.
2/22/2009 11
What is Digital
Image ?
A digital image
contains a fixed
number of rows and
columns of integer
numbers. Each integer
number is called pixel,
picture elements,
representing
brightness at the 34 58 98
points of the image. R 13 25 39
2/22/2009 12
88 47 17
Digital Image
Digital image = a multidimensional
array of numbers (such as intensity image)
or vectors (such as color image)
Each component in the image ⎡10 10 16 28⎤

⎢ 9 ⎡656 70 56 ⎥ 43⎤
26 37 ⎥ 78
called pixel associates with ⎢ ⎢32 ⎡99 54 70 ⎥ 67 ⎤
96 56
⎢
⎢15 25⎢6013 902296⎥ ⎥ 67⎥
the pixel value (a single number in ⎢ ⎢ 21 ⎢ 54 47 ⎥ 42⎥ ⎥
2/22/2009
the case of intensity images or a ⎢32 ⎢ 15⎢8587 8539⎥4313⎥ 92⎥
⎢54 ⎢ 65 65 39⎥ ⎥
vector in the case of color images). ⎢32 65 87 99⎥
Digital Image Types :
Binary Image
Binary image or black and white image
Each pixel contains one bit :
1 represent white
0 represents black
Binary data
⎡0 0 0 0⎤
⎢0 0 0 0⎥
⎢ ⎥
⎢1 1 1 1⎥
2/22/2009 14
⎢ ⎥
⎢1 1 1 1⎥
Intensity or Gray image
Intensity image or monochrome image
each pixel corresponds to light intensity
normally represented in gray scale (gray
level).
Gray scale values

⎡10 10 16 28⎤
⎢ 9 6 26 37⎥
⎢ ⎥
2/22/2009 ⎢15 25 13 22⎥ 15
⎢ ⎥
⎢32 15 87 39⎥
RGB image
Color image or RGB image:
each pixel contains a vector
representing red, green and
blue components.
RGB components
⎡10 10 16 28⎤
⎢ 9 ⎡656 70 56 ⎥ 43⎤
26 37 ⎥ 78
⎢ ⎢32 ⎡99 54 70 ⎥ 67 ⎤
96 56
⎢
⎢15 25⎢6013 902296⎥ ⎥ 67⎥
⎢ ⎢ 21 ⎢ 54 47 ⎥ 42⎥ ⎥
⎢32 ⎢ 15⎢8587 8539⎥43 ⎥ 92⎥
⎢54 ⎢ 65 65 16 39⎥
2/22/2009 ⎥
⎢32 65 87 99⎥
Index Image
Index image
Each pixel contains index number
pointing to a color in a color table
Color Table
Index Red Green Blue
No. component component component
1 0.1 0.5 0.3

2 1.0 0.0 0.0
⎡1 4 9 ⎤
⎢6 4 7 ⎥ 3 0.0 1.0 0.0
⎢ ⎥
⎢⎢6 5 2⎥⎥ 4 0.5 0.5 0.5
5 0.2 0.8 0.9
2/22/2009 Index value … … … … 17
Human Vision & Image
Visualization
In the beginning…
we’ll have a look at the human eye
2/22/2009 18
Cross section of Human
Eye
2/22/2009 19
Visual perception :
Human eye
1. The lens contains 60-70% water, 6% of fat.
2. The iris diaphragm controls amount of light that enters the eye.
3. Light receptors in the retina
- About 6-7 millions cones for bright light vision called photopic
- Density of cones is about 150,000 elements/mm2.
- Cones involve in color vision.
- Cones are concentrated in fovea about 1.5x1.5 mm2.
- About 75-150 millions rods for dim light vision called scotopic
- Rods are sensitive to low level of light and are not involved
color vision.
4. Blind spot is the region of emergence of the optic nerve from the eye.
2/22/2009 20 20
Electromagnetic
Spectrum
The whole electromagnetic spectrum is used by “imagers”
The human eye is sensible to electromagnetic waves in the
‘visible spectrum’ :
Electromagnetic Spectrum radio
frequency
microwave
visible (SAR)
gamma
cosmic rays
X-Rays UV
rays IR
-4 -2 2 4 6 8 10 12
10 10 1 10 10 10 10 10 10
wavelength (Angstroms)
-10
1 Å = 10 m
2/22/2009 21
The human eye is sensible to
electromagnetic waves in the ‘visible
spectrum’ , which is around a wavelength of
0.000001 m = 0.001 mm
2/22/2009 22
The human eye
•IIs able to perceive electromagnetic waves in a

certain spectrum
•IIs able to distinguish between wavelengths in this

spectrum (colors)
•HHas a higher density of receptors in the center
•Maps our 3D reality to a 2 dimensional

2/22/2009 23
image !
The retinal model is mathematically hard to
handle (e.g. neighborhood ?)
Easier: 2D array of cells, modelling the
cones/rods
Each cell contains a numerical value (e.g.

2/22/2009 24
between 0-255)
•TThe position of each cell defines the position of
the receptor
•TThe numerical value of the cell represents the

illumination received by the receptor
5 7 1 0 12 4 ………
2/22/2009 25
•WWith this model, we can create GRAYVALUE
images
•VValue = 0: BLACK (no illumination / energy)

•VValue = 255: White (max. illumination / energy)
2/22/2009 26
What is light?
• TThe visible portion of the electromagnetic (EM)
spectrum.
• IIt occurs between wavelengths of approximately 400 and
700 nanometers.
2/22/2009 27
Short wavelengths
• DDifferent wavelengths of radiation have different
properties.
• TThe x-ray region of the spectrum, it carries sufficient
energy to penetrate a significant volume or material.
2/22/2009 28
Long wavelengths
• CCopious quantities of infrared (IR) radiation are emitted
from warm objects (e.g., locate people in total darkness).
2/22/2009 29
Long wavelengths
• “Synthetic aperture radar” (SAR) imaging techniques
use an artificially generated source of microwaves to probe
a scene.
• SAR is unaffected by weather conditions and clouds (e.g.,
has provided us images of the surface of Venus).
2/22/2009 30
Range images
• AAn array of distances to the objects in the scene.
• TThey can be produced by sonar or by using laser
rangefinders.
2/22/2009 31
Sonic images
• PProduced by the reflection of sound waves off an object.
• HHigh sound frequencies are used to improve resolution.
2/22/2009 32
Image formation
Image is two dimensional pattern of
brightness
What to do to get info on 3D world?

-Study image formation process
- Understand how the brightness pattern is
produced.
Two important tasks
- Where the image of some point will
appear?
- How bright the image of some surface will
be ?
2/22/2009 33
A simple model of image formation
The scene is illuminated by a single
source.
The scene reflects radiation towards the
camera.
The camera senses it via chemicals on
film.
Light reaches
surfaces in 3D.
Surfaces reflect.
Sensor element
receives light energy.
Intensity is important.
Angles are important.
Material
2/22/2009
is important. 34
Geometry and physics
– The geometry of image formation,

which determines where in the image
plane the projection of a point in the scene
will be located.
– The physics of light, which determines

the brightness of a point in the image
plane as a function of illumination and
surface properties.
2/22/2009 35
Image Formation
Digital image generation is first step in any

image processing or computer vision
method. The generated image is function
of many parameters like the reflection
characteristics of object surface, sensor
characteristics of the camera, the optical
characteristics of lens, the analog to digital
converter, characteristics of light source,
geometric laws on the basis of which
image is acquired.
2/22/2009 36
Image formation
– The first task is related to the camera

projection which can be either a
perspective projection or an orthographic
projection. The perspective projection is
more general than the orthographic
projection, but requires calculations.
– The second task is related to surface

reflection properties, illumination
conditions, and surface orientation with
respect to camera and light sources.
2/22/2009 37
Geometric camera models
The projection of surface point of 3 dimensional
scene into 2 dimensional image plane can be
described by a perspective or orthographic
projection
Pine hole camera
A camera with zero aperture size
All rays from the 3D scene points always pass
through optical center of the lens ©
2/22/2009 38
Coordinate system
In computer vision, we deal with three kinds
of coordinates systems – Image coordinate
system, Camera coordinate system, and
World coordinate system. Image coordinate is
basically two-dimensional image plane.
Camera coordinate is one, that is adjusted to
the camera. It can be either camera centered
or image centered. In the camera-centered
coordinate system the origin is the focal point
and the optical axis is Z axis. In the image –
centered system the origin is positioned in
the XY image plane. World coordinate system
is general coordinate system with some
reference axes.
2/22/2009 39
Perspective projection set
up Consider pinhole camera model
Projection of a scene point P of the XYZ
space onto the image point P’ in the xy
image plane is perspective projection
The optical axis defined to be
perpendicular from the pinhole C to the
image plane
The distance f between C and the image
plane is focal length
The coordinate system of XYZ space is
defined such that the XY plane is parallel
to the image plane, and origin is at the
pinhole C, then the Z axis lies along the
2/22/2009 40
optical axis
Perspective projection
equations
2/22/2009 41
equations
From the similar triangles (CA’P’) and (CAP
From the two similar triangles (A’B’P’) and

(ABP)
From the last two equations, perspective

projection equations are obtained:
2/22/2009 42
Points go to points
Line go to lines
Planes go to whole image or half planes
Polygons go to polygons
Long focal length -> narrow field of view

Small focal length -> large (wide) field
of view-wide angle cameras
2/22/2009 43
• Produces a view where the object’s size
depends on the distance from the viewer
• An object farther away becomes smaller
2/22/2009 44
Horizon – observer’s eye level
Ground Line – plane on which object rests
Vanishing point – position on horizon where
depth projectors converge
Projection plane – plane upon which object is
projected
2/22/2009 45
Vanishing points
Object edges parallel to projection plane
remain parallel in a perspective projection
Object edges not parallel to projection
plane converge to a single point in a
perspective projection Æ vanishing point
(vp)
2/22/2009 46
Camera with aperture
In practice, the aperture must be larger to
admit more light.
Lenses are placed to in the aperture to focus
the bundle of rays from each scene point
onto the corresponding point in the image
plane
2/22/2009 47
Orthographic projection
Orthographic projection is modeled by rays
parallel to optical axis rather than passing
through the optical center
Suppose that the image of a plane lying at Z=Zo
parallel to the image plane is formed. The
magnification m can be defined as the ratio of
distance between two points in the image to
distance between their corresponding points on
the image plane
2/22/2009 48
2/22/2009 49
For an object located at average distance –Zo
and variations in Z over its visible surface is not
significant compared to –Zo (when distance
between camera and object is very large
relative to the variations in the object depth)
then the image this object will be magnified by
a factor m. For all the visible points of object ,
projection equations are
The scaling factor m is usually set to 1 or –1 for

convenience. Simple projection equations are
x=X, y=Y
2/22/2009 50
Radiometry basics
What determines the brightness of an
image pixel?
Light source
properties
Surface
Exposure shape
Surface reflectance
2/22/2009
Optics properties 51
Radiometry basics
Foreshortening and Solid angle

Measuring light : radiance Incoming Outgoing
Light at surface : interaction between light and
surface
– irradiance = light arriving at surface
– BRDF
– outgoing radiance
Special cases and simplifications : Lambertain,

specular, parametric and non-parametric models
2/22/2009 52
Foreshortening
Two sources that look the same to a receiver must

have same effect on the receiver;
Two receivers that look the same to a source must
receive the same energy.
2/22/2009 53
Solid Angle
By analogy with angle (in radians), the solid
angle subtended by a region at a point is the
area projected on a unit sphere centered at
that point
The solid angle dω subtended by a patch of
area dA is given by:
dA cos θ
dω =
r 2
A
Measured in steradians (sr)

Foreshortening : patches that
look the same, same solid
angle. 2/22/2009 54
Radiometry basics
Radiometry is a branch of physics that deals with
the measurement of the flow and transfer of
radiant energy.
Radiance is the power of light that is emitted from
a unit surface area into some spatial angle; the
corresponding photometric term is brightness.
Irradiance is the amount of energy that an image
capturing device gets per unit of an efficient
sensitive area of the camera. Quantizing it gives
image gray tones.c
2/22/2009 55
Radiometry basics
Radiance (L): energy carried by a ray
– Power per unit area perpendicular to the direction of travel,
per unit solid angle
– Units: Watts per square meter per steradian (W m-2 sr-1)
Irradiance (E): energy arriving at a surface

– Incident power in a given direction per unit area
– Units: W m-2
– For a surface receiving radiance L(x,θ,φ) coming in from dω the
corresponding irradiance is n dω
E (θ , φ ) = L(θ , φ ) cos θ dω
θ
2/22/2009 dA 56
dA cos θ
Radiance –emitted light
Radiance = power traveling at some point in a direction per unit
area perp to direction of travel, per solid angle
unit = watts/(m2sr)
A A cos θ θ
constant along a ray
dω
P P
L(x, θ , φ ) = dA
(dA cos θ )dω
Radiance transfer :
Power received at dA2 at dist r from
emitting area dA1 dA2
r θ2
dA2 cos θ 2 θ1
P1→2 = LdA1 cos θ1 ( ) P1→2 = P2→1
2/22/2009 r 2
dA1 57
dω21
Light at surface : irradiance
Irradiance = unit for light arriving at the surface
φ
dE (x) = L(x, θ , φ ) cos θdω
θ
Total power = integrate irradiance

over all incoming angles
2π π / 2
E ( x) = ∫ ∫ L(x,θ , φ ) cosθ sin θdθdφ x
dω
0 0
2/22/2009 58
Bidirectional reflectance distribution
Model of local reflection that tells how bright a
surface appears when viewed from one direction
when light falls on it from another
Definition: ratio of the radiance in the outgoing
direction to irradiance in the incident direction
surface normal
Le (θ e , φe ) Le (θ e , φe )
ρ (θ i , φi ,θ e , φe ) = =
Ei (θ i , φi ) Li (θ i , φi ) cos θ i dω
Radiance leaving a surface in a particular direction:

add contributions from every incoming direction
∫ ρ (θ , φ ,θ , φ ,)L (θ , φ )cosθ dω
Ω
i i e e i i i i i
2/22/2009 59
Light leaving surface : BRDF
many effects : Assume:
transmitted - glass • surfaces don’t
reflected - mirror fluorescent
? scattered – marble, skin • cool surfaces
x travel along a surface, • light leaving a
leave some other surface due to light
arriving
absorbed - sweaty skin
BRDF = Bi-directional reflectance distribution function
Measures, for a given wavelength, the fraction of incoming irradiance from a
direction ωi in the outgoing direction ωo [Nicodemus 70]
Le (x, θ e , φe )
ρ (x,θ i , φi ,θ e , φe ) =
Li (x, θ i , φi ) cos θ i dω
Reflectance equation : measured radiance
(radiosity = power/unit area leaving surface Correction
, φo ) = ∫ ρ ( x, θ i , φi ,θ o , φo ) L (θ i , φi ) cos(θ i ) dωi
Lo ( x, θ o2/22/2009 60
Ω
Reflection as convolution
equation
e e ∫Ω'
Reflectance L (x, θ , φ ) = ρ (x, θ ' , φ ' ,θ ' , φe' ) L(θ , φ ) cos(θ )dω
o i i e i i i i
= ∫ ρ (x, θ i ' , φi ' ,θe' , φe ' ) L( Rα , β (θ i ' , φi ' )) cos(θ i )dωi

Ω
Reflection behaves like

a convolution in the
angular domain
BRDF – filter
Light - signal
2/22/2009 61
Lambertian BRDF
Emitted radiance constant in all directions
Models – perfect diffuse surfaces : clay, mate paper, …
BRDF = constant = albedo
One light source = dot product normal and light direction
Lo (x) = ρLi (x,θ i , φi ) cos θ i
= ρ (N • L i )
albedo normal light dir
Diffuse reflectance acts like a low pass filter on the
incident illumination.
Lo ( x, θo , φo ) = ∫ ρ L(θi , φi ) cos(θi )dωi
Ω'
2/22/2009 62
BRDF for Lambertian
surface
Image irradiance = 1 * scene

Π
radiance
2/22/2009 63
How to represent
Z
surface ?
Y
r=1
X
ɸ
Surface normal ( n ) – It is directional vector with magnitude

^
unity
Camera in z direction we see hemisphere
Depth representation would be z=z(x,y)
2/22/2009 64
How to represent surface ?
Equation of the sphere (surface) x^2+y^2+z^2=a^2
Z=+sqrt(a^2-x^2-y^2)
If the surface is well behaved
z = z( x, y) = z( xo , yo ) + ∂xT ∇z
= z( xo , yo ) + ( x − xo ) ∂z + ( y − yo) ∂z + ........higherorderterms
∂x ∂y
If the surface is smooth, simply neglect higher order terms

i.e. small neighborhood you can consider it as plane
called planar approximation
z = ( x − xo ) ∂z + ( y − yo) ∂z
= p∂x + q∂y
∂x ∂y
First order approximation of surface, p and q relate to
The gradient of surface
2/22/2009 65
Surface
z
normal
∧
n
∧ y
n −
B
q∂y OA = (∂x,0, p , ∂x ) = (1,0, p )
−
∂y A OB = (0,1, q )
p∂x
x
∂x
Surface normal perpendicular to tangent plane
p =slope of surface in x direction
q =slope of surface in y direction
Cross product of OA & OB vectors gives the surface normal

∧
n = OA × OB
∧ ( − p , − q ,1 )
(p,q) is known, n∧ is known and henc
n = surface normal
2/22/2009 1+ p 2
+ q 2
66
Specular reflection
Smooth specular surfaces Rough specular surfaces
Mirror like surfaces Lobe of directions around
Light reflected along the specular direction
specular direction Microfacets
Some part absorbed
Lobe
Very small – mirror
Small – blurry mirror
2/22/2009 67
Bigger – see only light sources
Very big – fait specularities
Diffuse reflection
Dull, matte surfaces like chalk or latex paint

Microfacets scatter incoming light randomly
Light is reflected equally in all directions: BRDF
is constant
Albedo: fraction of incident irradiance reflected
by the surface
Radiosity: total power leaving the surface per
unit area (regardless of direction)
2/22/2009 68
Radiosity -summary
Radiance Light energy P
L(θ , φ ) =
(dA cos θ )dω
Irradiance Unit incoming

dE (x) = L(x, θ , φ ) cos θdω
light
Total Energy at Ei (x) = ∫ L(x, θ , φ ) cos θdω
Energy surface ω
incoming
Radiosity Unit outgoing Lo (x, θ e , φe ) = ∫ ρ (x, θ i , φi ,θ e , φe ) L(θ i , φi ) cos(θ i )dωi
radiance Ω
Total Energy leaving ⎡ ⎤

energy the surface Eo = ∫ ⎢∫ ρ ( x, θ i , φ θ
,
i e , φ e ) L (θ i , φi ) cos(θ i ) dω i ⎥ cos(θ e ) dωe
2/22/2009 Ωo ⎢⎣Ωi ⎥⎦ 69
leaving
Interaction of light and
matter
What happens when a light ray hits a point on an
object?
– Some of the light gets absorbed
converted to other forms of energy (e.g., heat)
– Some gets transmitted through the object
possibly bent, through “refraction”
– Some gets reflected
possibly in multiple directions at once
– Really complicated things can happen
fluorescence
Let’s consider the case of reflection in detail

– In the most general case, a single incoming ray could be
reflected in all directions. How can we describe the amount of
light reflected in each direction?
2/22/2009 70
Image formation system
Relation between what camera captures and
what the surface reflects
2/22/2009 71
-Consists of a thin lens and an image plane
The diameter of the lens is d and the value of
the focal length is fp. The system is assumed to
be focused,
rays originating from a particular point on the

object meet at single point in the image plane
rays originating from infinitesimal area dAo on
the object are projected into some area dAp in
the image plane and no rays from outside the
area dAo will reach dAp
When a camera captures the image of an

object, the measured gray value is proportional
to the image irradiance which is related to the
reflection properties of the object surface.
2/22/2009 72
How to calculate image irradiance in an image
forming system
- The radiant flux dɸ that is emitted from the
surface patch dAo and passes through the
-
entrance aperture can be calculated by
Where integration is over solid angle occupied

by the entrance aperture as seen from the
surface patch
By assuming that there is no power loss in the
medium, the image area dAp will receive the
same flux dɸ that is emitted from dAo
By definition, the image irradiance is the

incident
2/22/2009
flux per unit area 73
From the previous equations
- Let Ѳr’ be the angle between the surface

normal and the line to the entrance aperture,
and let α be the angle between this line and
the optical axis. The solid angle occupied by the
surface patch dAo and seen from the entrance
aperture equals the solid angle occupied by the
image area dAp :
From the previous equations
2/22/2009 74
If the size of the lens is small relative to the
distance between the lens and the object, then
the values of angle Ѳr’ in the previous integral
can be approximated by Ѳr’ and the reflectance
Lr tends to be constant and can be removed
from the integral which leads to
The solid angle occupied by the lens as seen

from the surface patch is approximately equal
d
to the foreshortened area Π (
2
) cos( α ) divided
2
by the distance Cos f ( α )

o
2/22/2009 75
Finally the expression of the image irradiance is
obtained as
That is, the image irradiance is proportional to

the scene radiance and the factor of
proportionality is a function of the off-axis
angle.
fp
d - >F stop number of camera.
2/22/2009 76
Image formation
⎡π ⎛ d ⎞2 ⎤
E = ⎢ ⎜ ⎟ cos 4 α ⎥ L
⎢ 4 ⎜⎝ f p ⎟⎠ ⎥
⎣ ⎦
Image irradiance is linearly related to scene
radiance
Irradiance is proportional to the area of the
lens and inversely proportional to the squared
distance between the lens and the image plane
The irradiance falls off as the angle between
the viewing ray and the optical axis increases
2/22/2009 77
What happens on Image plane
? (CCD camera plane)
Lens collects light arrays.
Array of small fixed elements replace chemicals of
film.
Each element generates a voltage signal based on
irradiance value
2/22/2009 78
Digitization
Analog images are continuous

representations of color
This is somewhat of a problem for
computers, which like discrete
measurements
2/22/2009 79
Digitization
Digital
Imaging Sample and storage Digital On-line
systems quantize computer buffer Display output
(disk)
object observe digitize store process Refresh

Record
/store
2/22/2009 80
Digital image acquisition
process
2/22/2009 81
Image sampling &
quantization y
Grayscale image
– A grayscale image is a function
I(x,y) of the two spatial coordinates
of the image plane.
– I(x,y) is the intensity of the image at the x
point (x,y) on the image plane.
I(x,y) takes non-negative values 250
assume the image is bounded by a 200
rectangle [0,a]×[0,b] 150
intensity
100
I: [0, a] × [0, b] → [0, inf )
50
I(x,y)
z Color image
0
100
– Can be represented by three functions,
80
R(x,y) for red, G(x,y) for green, and B(x,y) 50 60
for blue. y 40
x
20
rows 0 0 columns
2/22/2009 82
Image sampling & Quantization
The analog signal representing a

continuous image is sampled to produce
discrete values which can be stored by
a computer
The frequency of digital samples greatly
affects the quality of the digital image
2/22/2009 83
Image sampling & Quantization
To create a digital image, we need to

convert continuous sensed data into
digital form.
This involves two processes: sampling
and quantisation
The basic idea behind sampling and
quantization is illustrated in Fig. 3.1.
2/22/2009 84
Image sampling & quantization
Computer handles “discrete” data.
Sampling
– Sample the value of the image at the
nodes of a regular grid on the image
plane.
– A pixel (picture element) at (i, j) is the 255 (white)
image intensity value at grid point
indexed by the integer coordinate (i,
j).
Quantization
– Is a process of transforming a real
valued sampled image to one taking
only a finite number of distinct values.
0 (black)
– Each sampled value in a 256-level
grayscale image is represented by 8
2/22/2009 85
bits.
How sampling works ?
The original analog Measurements are Discrete samples

representation made at equal are taken from the
intervals measurements
2/22/2009 86
Figure 3.1(a) shows a continuous image,
f (x, y), that we want to convert to digital
form.
To convert it to digital form, we have to
sample the function in both coordinates
and in amplitude.
An image may be continuous with respect
to the x- and y-coordinates and also in
amplitude.
2/22/2009 87
Digitizing the coordinate values is called
sampling.
Digitizing the amplitude values is called
quantization.
2/22/2009 88
Fig 3.1 Generating a digital image (a) Continuous image. (b) A scan line from
A to B in the continuous image. (c) Sampling & quantisation. (d) Digital scan
2/22/2009 89
line.
The one-dimensional function shown in
Fig. 3.1(b) is a plot of amplitute (gray
level) values of the continuous image
along the line segment AB in Fig. 3.1(a).
To sample this function, we take equally
spaced samples along line AB, as shown in
Fig. 3.1(c).
Location of each sample is given by a
vertical tick mark in the bottom part of the
figure.
2/22/2009 90
The samples are shown as small white
squares superimposed on the function.
The set of these discrete locations gives
the sampled function.
However, the values of the samples still
span (vertically) a continuous range of
gray-level values.
In order to form a digital function, the
gray-level values also must be converted
(quantized) into discrete quantities.
2/22/2009 91
The right side of Fig. 3.1(c) shows the
gray-level scale divided into eight discrete
levels, ranging from black to white.
The vertical tick marks indicate the
specific value assigned to each of eight
gray levels.
The continuous gray levels are quantized
simply by assigning one of the eight
discrete gray levels to each sample.
2/22/2009 92
The assignment is made depending on the

vertical proximity of a sample to a vertical
tick mark.
The digital samples resulting from both
sampling and quantization are shown in
Fig. 3.1(d) and Fig 3.2 (b).
2/22/2009 93
How to choose the spatial
resolution : Nyquist rate
Original image
Sampled image
1mm
2mm
No detail is lost!
Minimum Nyquist Rate:
Spatial resolution
Period Spatial resolution must be less or equal
(sampling rate)
half of the minimum period of the image
or sampling frequency must be greater or
= Sampling locations
2/22/2009
Equal twice of the maximum frequency. 94
94
Aliased frequencyx (t ) = sin(2πt), 1 f =1
x2 (t ) = sin(12πt ), f =6
1
0.5
-0.5
-1
0 0.5 1 1.5 2 Sampling rate:
5 samples/sec
1
0.5
-0.5
-1
0 0.5 1 1.5 2
Two2/22/2009
different frequencies but the same results ! 95
Fig. 3.2 (a) Continuous image projected onto a sensor array.

2/22/2009 (b) Result of image sampling and quantisation 96
Image digitization
• SSampling means measuring the value of an

image at a finite number of points.
• QQuantization is the representation of the
measured value at the sampled point by an integer.
2/22/2009 97
Image digitization
2/22/2009 98
Fig. 3.3. Coordinate convention used to represent

2/22/2009 digital images 99
Fig. 3.4. A digital image of size M x N
2/22/2009 100
It is advantageous to use a more
traditional matrix notation to denote a
digital image and its elements.
2/22/2009
Fig. 3.5 A digital image 101
The number of bits required to store a
digitised image is
b=MxNxk
Where M & N are the number of rows and
columns, respectively.
The number of gray levels is an integer
power of 2:
L = 2k where k =1,2,…24
It is common practice to refer to the
image as a “k-bit image”
2/22/2009 102
The spatial resolution of an image is the physical
size of a pixel in that image; i.e., the area in the
scene that is represented by a single pixel in
that image. It is smallest discernible detail in an
image. Sampling is the principal factor
determining spatial resolution. Gray level
resolution refers to smallest discernible change
in gray level (often power of 2)
Dense sampling will produce a high resolution
image in which there are many pixels, each of
which represents of a small part of the scene.
Coarse sampling, will produce a low resolution
image in which there are a few pixels, each of
which represents of a relatively large part of the
scene.
2/22/2009 103
Fig. 3.6 Effect of resolution on image interpretation (a) 8x8

2/22/2009 104
image. (b) 32x32 image © 256x256 image
Effect of sampling
256x256
64x64
16x16
2/22/2009 105
Examples of Sampling
256x256 pixels 128x128 pixels
2/22/2009 106
64x64 pixels 32x32 pixels
Effect of spatial
resolution
2/22/2009 107
Effect of spatial
resolution
2/22/2009 108
Can we increase spatial
resolution by interpolation ?
2/22/2009 109
Down sampling is an irreversible process.
Image Sampling
original image sampled by a factor of 2
sampled by a factor of 4 sampled by a factor of 8
2/22/2009 110
Fig.3.7 Effect of quantisation on image interpretation. (a) 4

2/22/2009 111
levels. (b) 16 levels. (c) 256 levels
Effect of Quantization
8 bits / pixel
4 bits / pixel
2 bits / pixel
2/22/2009 112
Effect of quantization
levels
256 levels 128 levels
2/22/2009 113
64 levels 32 levels
Effect of quantization
16 levels 8 levels
In this image,
it is easy to see
false contour.
2/22/2009 4 levels 2 levels 114

Image quantization
• 256 gray levels (8bits/pixel) 32 gray levels (5 bits/pixel) 16 gray levels (4 bits/pixel)
• 8 gray levels (3 bits/pixel) 4 gray levels (2 bits/pixel) 2 gray levels (1 bit/pixel)
2/22/2009 115
Image representation
The result of sampling and quantisation is
a matrix of integer numbers as shown in
Fig.3.3, Fig.3.4. and Fig 3.5.
The values of the coordinates at the origin
are (x,y) = (0,0).
The next coordinate values along the first
row are (x,y) = (0,1).
The notation (0,1) is used to signify the
2nd sample along the 1st row.
2/22/2009 116
Images can be
represented by 2D
functions of the form
f(x,y).
The physical meaning
of the value of f at f(x,y)
spatial coordinates y
(x,y) is determined by
x
the source of the

image.
2/22/2009 117
In a digital image, both the coordinates and
the image value become discrete quantities.
Images can now be represented as 2D
arrays (matrices) of integer values: I[i,j] (or
I[r,c]).
The term gray level is used to describe
monochromatic intensity.
62 79 23 119 120 105 4 0
10 10 9 62 12 78 34 0
10 58 197 46 46 0 0 48
176 135 5 188 191 68 0 49
2 1 1 29 26 37 0 77
0 89 144 147 187 102 62 208
255 252 0 166 123 62 0 31
166 63 127 17 1 0 99 30
2/22/2009 118
How to select the suitable size and
pixel depth of images
The word “suitable” is subjective: depending on “subject”.
Low detail image Medium detail image High detail image

Lena image Cameraman image
To satisfy human mind
1. For images of the same size, the low detail image may need more pixel depth.
2. As an image size increase, fewer gray levels may be needed.
2/22/2009 119
The pixel
Sample location and sample values
combine to make the picture element or
pixel
3 color samples per pixel:
– 1 RED sample
– 1 GREEN sample
– 1 BLUE sample
Information about pixels is stored in a
rectangular pattern and displayed to the
screen in rows called rasters (from
Spalter).
2/22/2009 120
The pixel
Monitor pixels are actually circular light

representations of red, green and blue
phosphors
Pixel density is measured using Dots Per
Inch (DPI)
Pixel size is measured using Dot Pitch
DPI and Dot Pitch have an inverse
relationship ( DPI = Dot Pitch)
2/22/2009 121
Image characteristics
Each pixel is
assigned a numeric
value
(bit depth) that
represents a shade
of gray based on the
attenuation
characteristics of the
volume of tissue
imaged
2/22/2009 122
Pixel depth
TThe number of bits determines the number
of shades of gray the system is capable of
displaying on the digital images.
110- and 12- bit pixel can display 1024 and

4096 shades of gray, respectively.
IIncreasing pixel bit depth improves image

quality
2/22/2009 123
Bit-Depth
Number of bits to represent pixel color
Expression Name Colors

21 2-bit 2
24 4-bit 16
26 6-bit 64
2/22/2009 124
Bit-Depth
Number of bits to represent pixel color
Expression Name Colors

28 8-bit 256
216 16-bit 65, 536
224 24-bit (True About
Color) 16-million
2/22/2009 125
Digital image characteristics
AA digital image is

displayed as a
combination of rows
and columns known as
matrix
The smallest
component of the
matrix is the pixel
(picture element)
The location of the
pixel within the image
matrix corresponds to
an area within the
patient or volume of
tissue referred to as
2/22/2009
voxel 126
Matrix size
For a given
field of view, a
larger matrix
size includes a
greater number
of smaller
pixels.
2/22/2009 127
Color Fundamentals
UUsed heavily in human vision.
VVisible spectrum for humans is 400 nm (blue) to
700 nm (red).
MMachines can “see” much more; e.g., X-rays,
infrared, radio waves.
2/22/2009 128
HVS
CColor perception
Llight hits the retina, which contains photosensitive cells.
TThese cells convert the spectrum into a few discrete
values.
2/22/2009 129
HVS
TThere are two types of
Pphotosensitive cells:
CCones : Cones are sensitive to
ccolored light, but not very sensitive
tto dim light.
RRods : Sensitive to achromatic light.
TCones perceive color using three different types of cones.
Each one is sensitive in a different region of the spectrum.
445 nm (blue), 535 nm (green), 575 nm (Red). Have
different sensitivities. We are more sensitive to green than
red.
(
2/22/2009 130
W
ColorcanFundamentals
WHumans discern thousands of color shades and
intensities compared to about only two dozen shades
of gray.
WWhen a beam of sunlight passes through a glass
prism Emerging beam of light is continuous spectrum
of colors ranging from violet to at one end to red at
the other.
If the light is achromatic its only attribute is its

intensity, or amount. What can be seen on black and
white television set. Gray level refers to scale measure
Intensity that ranges from black, to grays, and finally
to white.
2/22/2009 131
Color fundamentals
Chromatic light spans the electromagnetic spectrum from 400 to
700nm. Quantities to describe quality of chromatic light source:
radiance, luminance, and brightness. Radiance is the total
amount of energy that flows from the light source. Luminance
is the amount of energy perceived by the observer. Brightness
is subjective measure that is practically impossible to measure.
It embodies the achromatic notion of intensity. Human eye
contains three types of cones red, green and blue. Due to the
absorption characteristics of the human eye, colors are seen as
variable combinations of the so called primary colors. R(red),
green(G), and blue(B). Wavelengths of these colors are 700nm,
546.1nm, and 435.8nm, respectively (as per CIE standard).
The primary colors can be added to produce the secondary colors
of light – magneta(red plus blue), cyan(green plus blue), and
yellow(red plus green).
2/22/2009 132
Color fundamentals
The characteristics that distinguish one color from another are
brightness, hue, and saturation. Brightness embodies the notion
of achromatic intensity. Hue is an attribute associated with the
dominant wavelength in a mixture of light waves. Hue is
dominant color perceived by an observer. When we call an
object is red, blue, orange, yellow we are referring to its hue.
Saturation means relative purity or amount of whit light mixed
with a hue. Degree of saturation is inversely proportional to the
amount of white light added. Hue and saturation taken together
are called chromaticity. Color may be characterized by its
brightness and chromaticity. The amount of red, green, blue
needed to form any particular color are referred to as tristimulus
values and denoted X, Y, and Z, respectively. A color is
characterized by its trichromatic coefficients, defined as
X
x =
X + Y + Z
Y
y =
2/22/2009 X + Y + Z 133
Z
z =
X + Y + Z
Color fundamentals
It is noted from these equations x+y+z=1. Another approach for
specifying colors is to use CIE chromaticity diagram (Fig),
which shows color composition as function of x(red) and
y(green). For any value of x and y, the corresponding value of
z(blue) is obtained as z=1-x-y. The point marked in figure has
approximately 62% green and 25% red content. Composition of
blue is 13%. The position of the various spectrum colors – from
voilet 380nm to red at 780nm –are indicated around the
boundary of the tongue shaped chromaticity diagram. At any
point within the boundary represents some mixture of spectrum
colors. The point of equal energy corresponds to equal fractions
of three primary colors: it represents CIE standard for white
light. Any point on the boundary of chromaticity chart is fully
saturated. As we progress towards point of equal energy more
and more white is added to the color and becomes less
saturated. Saturation is zero at the point of equal energy. 134
2/22/2009
CIE Chromaticity
TThe Commission model
Internationale de l’Eclairage
defined 3 standard primaries: X, Y, Z that can be
added to form all visible colors.
Y It was chosen so that its color matching
function matches the sum of the 3 human cone
responses.
⎡ X ⎤ ⎡0.6067 0.1736 0.2001⎤ ⎡ R ⎤

⎢Y ⎥ = ⎢0.2988 0.5868 0.1143⎥ ⎢G ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢⎣ Z ⎥⎦ ⎢⎣0.0000 0.0661 1.1149 ⎥⎦ ⎢⎣ B ⎥⎦
⎡ R⎤ ⎡ 1.9107 − 0.5326 − 0.2883⎤⎡ X ⎤

⎢G⎥ = ⎢− 0.9843 1.9984 − 0.0283⎥⎢ Y ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
2/22/2009
⎢⎣ B⎥⎦ ⎢⎣ 0.0583 − 0.1185 0.8986 ⎥⎦⎢⎣ Z ⎥⎦ 135
CIE Chromaticity model
xx, y, z normalize X, Y, Z
such that
x + y + z = 1.
Aactually only x and y are
needed because
z = 1 - x - y.
Ppure colors are at the
curved boundary.
Wwhite is (1/3, 1/3, 1/3).
2/22/2009 136
Color fundamentals
TThey provide a standard way of specifying a
particular color using a 3D coordinate system.
HHardware oriented
RRGB: additive system (add colors to black) used for
displays.
CCMY: subtractive system used for printing.
YYIQ: used for TV and is good for compression.
IImage processing oriented
HHSI: good for perceptual space for art, psychology and
recognition.
2/22/2009 137
Color fundamentals
• Primary Colors
2/22/2009 138
Color fundamentals
The RGB model, each color appears in its primary spectral
components of red green, and blue. This model is based on
cartesian coordinate system. The color subspace of interest is
the cube. RGB are the three primary colors and secondary
colors cyan, magnet, and yellow are located at the corners of
cube. In this model the gray scale extends from black to white
along the line that joins origin to (1,1,1). All the values of RGB
are assumed to be in the range [0 1]. Image represented in RGB
color consist of three component images, one for each primary
color. When fed to RGB monitor, these three images combine
to produce composite color. The number of pixels used to
represent each pixel in RGB space is called pixel depth. Each
RGB pixel has depth of 24 bits.
2/22/2009 139
Color fundamentals
Secondary colors
(additive synthesis):
2/22/2009 140
Color fundamentals
Secondary colors
– adding primary colors:
R + G + B = black
R + G + B = blue
R + G + B = green
R + G + B = cyan
R + G + B = red
R + G + B = magenta
R + G + B = yellow
2/22/2009 R+G+B= = white
white 141
Color fundamentals
Secondary colors
– weighted adding of primary colors:
0.5 ··R
R + 0.5 ··G
G + 0.5 ··B
B = grey
1.0 ··R
R + 0.2 ··G
G + 0.2 ··B
B = brown
0.5 ··R
R + 1.0 ··G
G + 0.0 ··B
B = lime
1.0 ··R
R + 0.5 ··G
G + 0.0 ··B
B = orange
2/22/2009 142
Color fundamentals
Color images can be represented by

3D Arrays (e.g. 320 x 240 x 3)
2/22/2009 143
Color
AAdditive fundamentals
model. - RGB
AAn image consists of 3 bands, one for each
primary color.
AAppropriate for image displays.
2/22/2009 144
Color fundamentals - CMY
Primary colors
(subtractive synthesis):
2/22/2009 145
Color fundamentals
Cyan, Magneta, and Yellow are the secondary colors of light or

alternatively primary colors of pigments. For example when a
surface coated with cyan pigment is illuminated with white
light, no red light is reflected from the surface. That is Cyan
subtracts red light from reflected white light, which itself is
composed of equal amounts of red, green, blue light.
Most devices that deposit colored pigments on paper, such as color
printers and copiers, require CMY data input or perform an
RGB to CMY conversion internally. This conversion is
performed using the simple operation
2/22/2009 146
CMY model
CCyan-Magenta-Yellow is a subtractive model
which is good to model absorption of colors.
AAppropriate for paper printing. Assumption here is
all the color values are normalized to the raneg [0 1]
⎡ C ⎤ ⎡1⎤ ⎡ R ⎤
⎢ M ⎥ = ⎢1⎥ − ⎢G ⎥
⎢ ⎥ ⎢⎥ ⎢ ⎥
⎢⎣ Y ⎥⎦ ⎢⎣1⎥⎦ ⎢⎣ B ⎥⎦
2/22/2009 147
Color fundamentals - CMYK
Equal amounts of the pigment primaries, cyan, magneta, and

yellow should produce black. In practice, combining these
colors for printing produces a muddy-looking black. So in order
to produce black, a fourth color black is added giving rise to the
CMYK color model. When publishers talk about “four-color
printing” they are referring to the three colors of CMY color
model plus black.
2/22/2009 148
Color fundamentals - HSI
RGB and CMY color models are ideally suitable for hardware
implementations. RGB strongly matches with the fact that
human eye strongly perceptive to red, green, and blue
components. Unfortunately these color models and similar
other models are not well suited for describing colors in terms
that are practical for human interpretation. For example one
does not refer to the color of an object by giving the percentage
of each of the primaries composing its color. In other words we
do not think of color images as being composed of three
primary images that combine to form that single image.
2/22/2009 149
When humans view a color object, we describe it by its hue,
saturation, and brightness. Hue is a color attribute that describes
pure color (pure red, orange, or yellow). Saturation give
measure of the degree to which a pure color is diluted by white
light. Brightness is subjective descriptor that is practically
impossible to measure. It embodies achromatic notion of
intensity and is one of key factors for color sensation. We do
know that the intensity (grey level) is most useful descriptor of
monochromatic images. This quantity is easily measurable and
interpretable. One such model that decouples the intensity
component from the color-carrying information (hue and
saturation) in a color image is HSI. As a result HIS model is an
ideal tool for developing image processing algorithms based on
color descriptions that are natural and intuitive to humans.150
2/22/2009
Color fundamentals
To summarize RGB is ideal for image generation (image capture
by a color camera or image display on monitor screen), but its
use for color description is much more limited. RGB color
image can be viewed as three monochromatic intensity images.
In the RGB model the line joining black and white vertex
represents intensity axis. To determine intensity of any color
point just pass a plane perpendicular to the intensity axis. That
gives us intensity value in the range [0 1]. Saturation of color
increases as a function of distance from the intensity axis.
Saturation of points on the intensity axis is zero. It is length of
the vector from the origin to the point. Note that origin is
defined by intersection of color plane with intensity axis. Hue
can also be determined from the RGB point. It is a plane
formed by three points (black, white , cyan). All the colors
generated by three colors lie in triangle defined by those colors.
Usually hue of some point is determined by an angle from some
reference point. Usually an angle of 0 from the red axis 151
2/22/2009
designates 0 hue and increases counterclockwise from there.

Color fundamentals
UUniform: equal (small) steps give the same perceived
color changes.
HHue is encoded as an angle (0 to 2π).
SSaturation is the distance to the vertical axis (0 to 1).
IIntensity is the height along the vertical axis (0 to 1).
2/22/2009 152
The three important components of the HSI color space are
the vertical intensity axis, the length of the vector to the
color point, and the angle this vector makes with the red
axis.
HTo summarize HSI: Hue, saturation, value are non-linear
functions of RGB. Hue relations are naturally expressed in
a circle.
( R+G+B)
I=
3
min( R, G, B)
S = 1−
I
⎧ 1 / 2[( R − G )+( R − B)] ⎫⎪
−1 ⎪
H = cos ⎨ ⎬ if B<G
[ ]
⎪⎩ ( R − G ) +( R − B )(G − B ) ⎪⎭
2
⎧ 1 / 2[( R − G )+( R − B )] ⎫⎪
−1 ⎪
H = 360 − cos ⎨ ⎬ if B>G153
2/22/2009
[ ]
⎪⎩ ( R − G ) +( R − B )(G − B) ⎪⎭
2
(Left) Image of food originating from a digital camera.

(Center) Saturation value of each pixel decreased 20%.
(Right) Saturation value of each pixel increased 40%.
2/22/2009 154

Color
HHave fundamentals
better -YIQ
compression properties.
LLuminance Y is encoded using more bits than
chrominance values I and Q (humans are more
sensitive to Y than I and Q).
LLuminance used by black/white TVs.
AAll 3 values used by color TVs.
⎡Y ⎤ ⎡0.299 0.587 0.114 ⎤ ⎡ R ⎤

⎢ I ⎥ = ⎢0.596 − 0.275 − 0.321⎥ ⎢G ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
2/22/2009 ⎢⎣Q ⎥⎦ ⎢⎣0.212 − 0.532 0.311 ⎥⎦ ⎢⎣ B ⎥⎦ 155
Color fundamentals - Summary
TTo print (RGB Æ CMY or grayscale)
TTo compress images (RGB Æ YUV)
CColor description (RGB Æ HSI)
CColor information (U,V) can be compressed 4 times
without significant degradation in perceptual quality.
TTo compare images (RGB Æ CIE Lab)
CCIE Lab space is more perceptually uniform.
EEuclidean distance in Lab space hence meaningful.
2/22/2009 156
Storing Images
With the traditional cameras, the film is used both to record and
store the image. With digital cameras, separate devices perform
these two functions. The image is captured by the image sensor,
then stored in the camera on a storage device of some kind. We
look at many of the storage devices currently used.
Removable Vs. Fixed Storage
Older and less expensive cameras have built-in fixed storage that
can’t be removed or increased. This greatly reduces the number of
photos you can take before having to erase to make room for new
ones.
Allmost all newer digital cameras use some form of removable
storage media, usually flash, memory cards, but occasionally small
hard disks, and even CDs, and the variations of the floppy disk.
Whatever its form, removable media let’s you remove one storage
2/22/2009 157
device when it is full and insert another.
Storing Images
The number of images you take is limited only by the number of
storage devices you have and the capacity of each.
The number of images that you can store in a camera depends on a
variety of factors including
(1) The capacity of the storage device (expressed in Megabytes)
(2) The resolution at which pictures are taken
(3) The amount of compression used
Number you can store is important because once you reach the
limit you have no choice but to quit taking pictures or erase
some existing ones to make room for new ones. How much
storage capacity you need depends partly on what you use
camera for.
2/22/2009 158
Storing images
The advantages of removable storage are many. They include the
following:
(1) They are erasable and reusable
(2) They are usually removable, so you can remove one and plug in
another so storage is limited only by the number of devices you
have.
(3) They can be removed from the camera and plugged into the
computer or printer to transfer the images.
2/22/2009 159
Storing Images
Flash Card Storage:
As the popularity of digital cameras and other hand held devices
has increased, so has the need for small, inexpensive memory
devices. The type that is caught on is flash memory which uses
solid state chips to store your image files. Although flash
memory chips are similar to RAM chips that are used inside
your computer there is one important difference. They don’t
require batteries and don’t loose images when power is turned
off. Your photographs are retained indefinitely without any
power to the flash memory components. These chips are
packaged inside a case equipped with electrical connectors and
the sealed unit is called a card.
Flash memory cards consume little power, take up little space, and
are very rugged. They are also very convenient. You can carry
lots of them with you and change them as needed.
2/22/2009 160
Storing images
Until recently, most flash cards have been in the standard PC card
format that is widely used in the network computers. However,
with the growth of the digital camera and other markets, a
number of smaller formats have been introduced. As a result of
this competition, camera support confusing variety of
incompatible flash memory cards including the following :
PC Cards :
CompactFlash:
Smart media
Memory sticks
xD-picture cards
Each of these formats is supported by its own group of companies
and has its own following.
2/22/2009 161
Storing images
PC Cards :
PC Cards have the highest storage capacities but their large size has
led to their being used mainly in professional cameras
CompactFlashCads
They are generally the most advanced flash storage devices for
consumer level digital cameras
CompactFlash Terminology
Compactflash cards and slots that are 3.3mm thick are called
CompactFalsh(abbreviated CF) or CompactFlash Type
1(abbreviated CF-1) and that are 5mm thick are called Type II
Smart media cards
They are smaller than compactflash cards and generally don’t come
2/22/2009 162
with storage capacities quite as high
Storing Images
Sony memory sticks, shaped something like stick of a gum,
currently used mainly in sony products
xD picture cards
The xD picture cards are the smallest of the memory cards and used
in very small cameras. It was developed jointly by Fuji and
Olympus
Memory card storage cases
Cards are easy to misplace and the smaller they are, the easier they
are to lose. If you don’t find a way to store them safely. One
way to keep them safe is to use an inexpensive storage case.
2/22/2009 163
Storing Images
Hard disk storage
One of the current drawbacks of compact memory flash cards is
their limited storage capacity. For high resolution cameras this
is a reall drawback. One solution is high speed, high capacity
hard disk drives. Untill recently these drives were too large and
expensive to be mounted inside cameras, but that changed with
IBM’s introduction of Microdrive hard disk drives. These
drives now owned by Hitachi are smaller in volume lighter in
volume than a roll of film. Infact, they are so small that they
can be plugged into a Type II compact flash slot into digital
camera or flash card reader. The Hitachi Microdrive fits a CF-II
slot and is a marvel of engineering.
2/22/2009 164
Storing Images
Optical storage disks
CDs are used in few cameras and have the advantage that they can
be read in any system with a CD drive. The disks are write once
with archival quality with no danger of important files being
deleted or written over. Sony’s line of Marvicas use CD discs
for storage.
Temporary storage :
Portable digital image storage and viewing devices are advancing
rapidly that’s good because they meet real need. When our
photographing, if our storage device becomes filled with
images, you need to place to temporarily store the images until
images are transferred to main system. One device used for this
is notebook computer. Not only many do have one of these, but
their large screen and ability to run any of software. However, a
notebook computer is not always the ideal temporary storage
2/22/2009 165
device, because of its weight, short battery life, and long start-
up time. Hence the introduction of portable hard drive.
Storing
PC Cards :
Images
FlashTrax from SmartDisk is one of the new multimedia
storage/viewer devices.
To use one of these devices you insert your memory card into a
slot, often using an adapter, and quickly transfer your images.
You can erase your camera’s storage device to make room for
new images and resume shooting. When you get back to your
permanent set up, you copy or move your images from the
intermediate storage device to the system you use for editing,
printing, and distributing them. The speed with which you
transfer depends on connections supported by the device. Most
support USB 2 and some support FireWire. The latest trend is
to incorporate image storage into multipurpose devices. Many
of these devices let you review the stored images on the device
itself or on connected TV. Some also let you to print the images
2/22/2009 166
directly on the printer without using computer.
Storing Images
The ternd is to go even farther and combine digital photos, digital
videos, and MP3 music, in the same device. With a device like
this one will be able to create slide shows with special
transitions, pans, and accompanying music and play them back
anywhere.
One way to eliminate or reduce the need for intermediate storage is
to use a higher capacity storage device in the camera. For
example, some devices store many gigabytes of data, enough to
store hundreds of large photos.
2/22/2009 167
Storing Images
The key questions to ask when considering an intermediate storage
devices are:
(1) What is its storage capacity ? What is the cost per megabyte of
storage?
(2) Does it have slots or adapters for the storage devices you use ?
(3) Does it support image formats you use ? Many support
common image formats like JPEG but not proprietary formats
such as Canon’s RAW and Nikon’s NEF format.
(4) Does it support video and MP3 music playback? Does it
support camera’s movie format if it has one ?
(5) What is the transfer rate and how long does it take to transfer
the images from card to the device?
(6) 2/22/2009
Can it display images on a TV set or be connected directly168to a
printer ?
Storing Images
(7) If it connects to a TV does it have remote control?
(8) Can you view images on devices own screen ?
(9) Are there ways to rotate, zoom in/out and scroll ?
2/22/2009 169
Introduction to double buffering
BitBlt - > It Stands for Bit-Block Transfer. It means that a “block”

of bits, describing a rectangle in an image, is copied in one
operation. Usually the graphics card supports this command in
hardware. There is a function in the Win32 API of this name,
which also occurs in MFC, but the FCL does not provide this
function to you directly. Nevertheless it is essentially packaged for
use as the Graphics.DrawImage method
2/22/2009 170
Introduction to double buffering
Memory DC - > DC means “device context”. This is represented
in the FCL as a Graphics object. So far, the Graphics objects we
have used in our programs usually corresponded to the screen; and
in one lecture, we used a Graphics object that corresponded to the
printer. But it is possible to create a Graphics object that does not
correspond to a physical device. Instead, it just has an area of
RAM (called a buffer) that it writes to instead of writing to video
RAM on the graphics card. When you (for example) draw a line
or fill a rectangle in this Graphics object, nothing changes on the
screen (even if you call Invalidate), since the memory area being
changed by the graphics calls is not actually video RAM and has no
connection with the monitor. This “virtual Graphics object” is
loosely referred to as a memory DC. It is a “device” that exists
only in memory; but usually its pixel format does correspond to a
physical device such as the screen, so that when data is copied from
this2/22/2009
buffer to video memory, it is correctly formatted. 171
2/22/2009 171
Double buffering
We can use the images as offscreen drawing surfaces
according to storing them as pictures
This allows us to render any image, including text and
graphics,
to an offscreen buffer that we can display at a later time.
The advantage of doing this is that the images is seen only
when it is complete
Drawing a complicated image could take several milliseconds
or more
which can be seen by the user as flashing and flickering
This flashing is distracting
9 causes the user to perceive his rendering as slower than actually is
Usage of an offscreen image to reduce flickers is called
double buffering. Because:
the screen is considered a buffer for pixels, and
the offscreen image is the second buffer,
9 where we can prepare pixels for display.
2/22/2009 172
What is double buffering ?
Double buffering - > “Double buffering” refers to the technique of
writing into a memory DC and then BitBlt-ing the memory DC to
the screen.
This works as follows: your program can take its own sweet time
writing to a memory DC, without producing any delay or flicker
on the screen. When the picture is finally complete, the program
can call BitBlt and bang! Suddenly (at the next vertical retrace
interval) the entire contents of the memory DC’s buffer are copied
to the appropriate part of video RAM, and at the next sweep of the
electron gun, the picture appears on the screen. This technique is
known as double buffering. The name is appropriate because there
are two buffers involved: one on the graphics card (video RAM)
and one that is not video RAM, and the second one is a “double”
of the first in the sense that it has the same pixel format.
2/22/2009 173
[Some books reserve this term for a special case, in which the
graphics card has two buffers that are alternately used to refresh the
monitor, eliminating the copying phase. But most books use the
term double buffering for what we have described.]
Whatever is stored in the memory DC will not be visible, unless
and until it gets copied to the DC that corresponds to the screen.
This is done with BitBlt, so that the display happens without
flicker.
2/22/2009 174
Why use double buffering ?
Double buffering can be used whenever the computations needed
to draw the window are time-consuming. Of course, you could
always use space to replace time, by storing the results of those
computations. That is, in essence, what double-buffering does.
The end result of the computations is an array of pixel information
telling what colors to paint the pixels. That’s what the memory DC
stores.
This situation arises all the time in graphics programming. All
three-dimensional graphics programs use double-buffering.
MathXpert uses it for two-dimensional graphics. We will soon
examine a computer graphics program to illustrate the technique.
Another common use of double-buffering is to support animation.
If you want to animate an image, you need to “set a timer” and
then at regular time intervals, use BitBlt to update the screen to
show the image in the next position.
2/22/2009 175
Why use double buffering?
The BitBlt will take place during the vertical retrace interval.
Meantime, between “ticks” of the timer, the next image is being
computed and drawn into the memory DC. In the next lecture, this
technique will be illustrated.
When should the memory DC be created?
Obviously it has to be created when the view is first created. Less
obviously, it has to be created again when the view window is
resized. But the memory DC is the same size as the screen DC,
which is the same size as the client area of the view window.
When this changes, you must change your memory DC
accordingly.
When you create a new memory DC, you must also destroy the
old one, or else memory will soon be used up by all the old
memory DC’s (provided the evil user plays around resizing his
2/22/2009 176
window many times).
It turns out that every window receives a Resize event when it is
being resized, and it also receives a Resize event soon after its
original creation (when its size changes from zero by zero to the
initial size). Therefore, we add a handler for the Resize event and
put the code for creating the memory DC in the Resize message
handler.
2/22/2009 177
Cathode Ray Tube
Adobe Acrobat 7.0

Document
2/22/2009 178
Liquid crystal display
LCD (Liquid Crystal Display) panels are "transmissive" displays, meaning they
aren't their own light source but instead rely on a separate light source and
then let that light pass through the display itself to your eye.
We can start to describe how an LCD panel works by starting with that light
source. The light source is a very thin lamp called a "back light" that sits
directly behind the LCD panel as shown in Figure 1.
Figure 1
2/22/2009 179
LCD
The light from the backlighting then passes through a polarizing filter (a filter
that aligns the light waves in a single direction). From there the now polarized
light then passes through the actual LCD panel itself. The liquid crystal portion
of the panel either allows the polarized light to pass through or blocks the light
from passing through depending on how the liquid crystals are aligned at the
time the light tries to pass through. See Figure 2.
Figure 2
2/22/2009 180
LCD
The liquid crystal portion or the panel is spit up into tiny individual cells that
are each controlled by a tiny transistor to supply current. Three cells side by
side each represent one "pixel" (individual picture element) of the image. An
800 x 600 resolution LCD panel would have 480,000 pixels and each pixel
would have three cells for a total of 1,440,000 individual cells.
Red, green and blue are the primary colors of light. All other colors are made
up of a combination of the primary colors. An LCD panel uses these three
colors to produce color which is why there are three cells per pixel — one cell
each for red, green, and blue.
Once the light is passed through the liquid crystal layer and the final polarizing
filter it then passes through a color filter so that each cell will then represent
one of the three primary colors of light. See Figure 3.
Figure 3
2/22/2009 181
LCD
The three cells per pixel then work in conjunction to produce color. For example, if a
pixel needs to be white, each transistor that controls the three color cells in the pixel
would remain off, thus allowing red, green and blue to pass through. Your eye sees the
combination of the three primary colors, so close in proximity to each other, as white
light.
If the pixel needed to be blue, for and area of an image that was going to be sky, the two
transistors for the red and green cells would turn on, and the transistor for the blue cell
would remain off, thus allowing only blue light to pass through in that pixel.
Pros:
1.LCD displays are very thin. They can be mounted in places traditional CRT
televisions and monitors cannot.
2.Color reproduction is excellent.
3.Contrast is good, although not great.
4.Pixel structure is very small, which creates a very smooth image.
5.Durable technology.
6.No burn-in issues.
Cons:
1.Very expensive technology per square inch of viewing area.
2.Black levels and details in dark scenes are not as strong as those in competing
technologies.
3.Dead pixels can be an issue, although quality has improved as the technology has
matured.
2/22/2009 182
4.Sizes above 40" are cost prohibitive.
LCD
Is an LCD Panel right for you? It depends on your needs. Below is a list of
common scenarios where an LCD panel provides the best performance,
followed by a list of scenarios that might suggest the need to use a different
technology.
Scenarios where an LCD flat panel will perform well:
1. Any application that will require a screen of less than 42" diagonal.
2. Installations that require the monitor/television to be built into a wall or
cabinetry, and require a diagonal image of less than a 42".
3. Pre-made entertainment centers and bedroom armoires.
4. Any application that requires wall mounting and requires a diagonal image
of less than 42".
Scenarios where another technology might be more effective:
1. Any application that requires a large screen — larger than 40" diagonal. LCD
displays get cost prohibitive for sizes above 40". If you opt to select an LCD
panel of over 40", be prepared to pay.
2. Applications where the best possible image quality is needed. A CRT is still
going to give the best shadow detail and color.
2/22/2009 183
3. Tight budgets; CRT technology will be much less expensive per viewing area.
Printers
Inch
Type of measurement equal to 25.4 millimeters or 2.54
centimeters.
Measurement
When referring to computers, a measurement is the
process of determining a dimension, capacity, or quantity
of an object or the duration of a task. By using
measurements an individual can help ensure that an
object is capable of fitting within an area, a storage
medium is capable of storing the necessary files, a task
will complete in the required time, or how fast a product
is when compared to another product. Below is a listing of
different types of computer measurements you may
encounter while working with a computer or in the
computer
2/22/2009
field. 184
Printers
PPI
Short for Pixels Per Inch, PPI is the number of pixels per inch a pixel image
is made up of. The more pixels per inch the image contains, the higher
quality the image will be.
Term that comes from the words Picture Element (PEL). A pixel is the
smallest portion of an image or display that a computer is capable of
printing or displaying. You can get a better understanding of what a pixel
is when zooming into an image, as seen in the example to the right.
As you can see in this example, the character image in this picture has been
zoomed into at 1600%. Each of the blocks seen in this example is a single
pixel of this image. Everything on the computer display looks similar to this
when zoomed in upon; the same is true with printed images, which are
created by several little dots that are measured in DPI.
Pixel image
A type of computer graphic that is composed entirely of pixels.
2/22/2009 185
Printers
There seems to be a lot of confusion about what PPI
means (apart from the fact that it means Pixels Per Inch
of course). This article is for beginners in computer
graphics and digital photography.
Dots Per Inch usually means the maximum dots a printer

can print per inch. Roughly speaking the more DPI the
higher quality the print will be. DPI is for printers, PPI is
for printed images. But I don't think there is an official
definition of the difference.
PPI and DPI are sometimes but not always the same, I'll
assume for simplicity in this article that they are.
However, until you print an image the PPI number is
meaningless.
2/22/2009 186
Printers
Until you print an image the PPI number is meaningless.
Imagine, for simplicity's sake, that the image below,
when printed on your printer is one inch (or 2.54 cm)
square:
If you count the pixels (blocks, dots) you'll find that there
are 10 across the width of an image. If this was printed
at the size of 1 inch it would be a 10 PPI image. Here is a
50 PPI version of the same image:
2/22/2009 187
Printers
You need not count the pixels, believe me the square
above is a 50 by 50 pixel image, and if printed so that it
covered exactly 1 square inch it would be a 50 PPI image.
Now lets look at a 150 PPI image:
PPI is simply how many pixels are printed per inch of

paper. You may not be able to see the pixels (because
your eyes or your printer is not that high a quality). The
above images are approximations but you get the idea.
I don't care what the "image information" on your camera
says, or the what PPI reading on your paint program
says, only when you print can you really say what the PPI
is. And the same image will have different PPI when
printed2/22/2009
at different sizes. 188
Printers
DPI (dots per inch) is a measurement of printer
resolution, though it is commonly applied, somewhat
inappropriately, to monitors, scanners and even digital
cameras.
For printers, the DPI specification indicates the number of
dots per inch that the printer is capable of achieving to
form text or graphics on the printed page. The higher the
DPI, the more refined the text or image will appear. To
save ink, a low DPI is often used for draft copies or
routine paperwork. This setting might be 300 or even 150
DPI. High resolution starts at 600 DPI for standard
printers, and can far exceed that for color printers
designed for turning out digital photography or other
high-resolution images.
2/22/2009 189
Printers
In the case of monitors, DPI refers to the number of
pixels present per inch of display screen. The technically
correct term is "PPI" or pixels per inch, but DPI is
commonly used instead. A display setting of 1280 x 1024
has 1.3 million DPI, while a setting of 800 x 600 has
480,000, or less than half the resolution of the higher
setting. With fewer dots per inch, the picture will not
have the clarity that can be achieved with a higher DPI
saturation. This is because displays create images by
using pixels. Each dot or pixel reflects a certain color and
brightness. The greater the DPI, the more detailed the
picture can be. Higher DPI also requires more memory
and can take longer to 'paint' images, depending on the
system's video card, processor and other components.
2/22/2009 190
Printers
Scanners also operate at different resolutions. Scan time
will increase with higher DPI settings, as the scanner
must collect and store more data. However, the greater
the DPI, or requested resolution, the richer the resulting
image. A high DPI setting mimics the original image in a
truer fashion than lower DPI settings are capable of
doing. If the image is to be enlarged, a high DPI setting
is necessary. Otherwise the enlarged picture will look
"blocky" or blurry because the software lacks information
to fill in the extra space when the image is enlarged.
Instead it "blows up" each pixel to "smear" it over a wider
area. Technically again, the more correct term in this
application is sampled PPI, but DPI is more often used.
2/22/2009 191
Printers
Digital cameras have their own specifications in terms of
megapixels and resolution, but DPI is often mentioned in
this context as well. Since DPI in all cases refers to the
output image, a digital camera capable of the most basic
current standards of resolution —- 3.0 megapixels and
better —- will output an image capable of taking
advantage of a very high DPI setting on the printer.
However, if your printer is only capable of 600 DPI, the
extra resolution of the camera will be lost in the printing
process. When buying or upgrading components it is
therefore critical that each product is capable of
supporting the highest standards of any interfacing
product.
2/22/2009 192
Printers
Print quality
The quality of the hard copy produced by a computer printer. Below is a listing of some of
the more common reasons why the print quality may differ.
1. Type of printer - Each type of printer has its own capabilities of printing. With
standard printers, dot matrix is commonly the lowest quality printer, ink jet printers are
commonly average quality, and laser printers are commonly the best quality.
2. Low DPI - Printer has a low DPI.
3. Print mode - The mode that the hard copy was produced may also affect the overall
quality of the print. For example, if the mode was draft quality, the printer will print
faster, but will be a lower quality.
4. Available toner or ink - If the printer is low on toner or ink the quality can be
dramatically decreased.
5. Dirty or malfunctioning printer - If the printer is dirty or is malfunctioning this can
also affect the quality of the print.
6. Image quality - It is important to realize that when printing a computer graphic, the
quality may not be what you expect because of any of the below reasons.
•Printer does not have enough colors to produce the colors in the image. For example,
some printers may only have four available inks where others may have six or more
available inks. See process color.
•The image is a low quality or low resolution image.
2/22/2009 193
•Image is too small and/or has too many colors in a small area.
Printers
Most people have used printers at some stage for printing
documents but few are aware of how it works. Printed
documents are arguably the best way to save data. There
are two types of basic printers Impact and Non-impact.
Impact printers, as the very name implies means that the
printing mechanism touches the paper for creating an image.
Impact printers were used in early 70s and 80s. In Dot Matrix
printers a series of small pins is used to strike on a ribbon
coated with ink to transfer the image on the paper.
Other Impact Printers like Character printers are basically
computerized typewriters. They have a series of bars or a
ball with actual characters on them, which strike on the ink
ribbon to transfer the characters on the paper. At a time only
one character can be printed. Daisy Wheel printers use a
2/22/2009 194
plastic or metal wheel.
Printers
These types of printers have limited usage though because
they are limited to printing only characters or one type of
font and not the graphics.
There are Line printers where a chain of characters or pins,
print an entire line, which makes them pretty fast, but the
print quality is not so good. Thermal printers are nothing but
printers used in calculators and fax machines. They are
inexpensive to use. Thermal printers work by pushing
heated pins against special heat sensitive paper.
More efficient and advanced printers have come out now
which use new Non-impact Technology.
Non-impact printers are those where the printing
mechanism does not come into the contact of paper at all.
This makes them quieter in operation in comparison to the
impact
2/22/2009 printers. 195
Printers
In mid 1980s Inkjet printers were introduced. These have
been the most widely used and popular printers so far.
Colour printing got revolutionized after inkjet printers were
invented. An Inkjet printer's head has tiny nozzles, which
place extremely tiny droplets of ink on the paper to create an
image. These dots are so small that even the diameter of
human hair is bigger. These dots are placed precisely and
can be up to the resolution of 1440 x 720 per inch. Different
combinations of ink cartridges can be used for these
printers.
2/22/2009 196
Printers
How an Inkjet printer works
The print head in this printer scans the page horizontally
back and forth and another motor assembly rolls the paper
vertically in strips and thus a strip is printed at a time. Only
half a second is taken to print a strip. Inkjet printers were
very popular because of their ability to colour print. Most
inkjets use Thermal Technology. Plain copier paper can be
used in these printers unlike thermal paper used for fax
machines. Heat is used to fire ink onto the paper through the
print head. Some print heads can have up to 300 nozzles.
Heat resistant and water based ink is used for these
printers.
2/22/2009 197
Printers
The latest and fastest printers are Laser Printers. They use the
principal of static electricity for printing it as in photocopiers.
The principle of static electricity is that it can be built on an
insulated object. Oppositely charged atoms of objects (positive
and negative) are attracted to each other and cling together.
For example, pieces of nylon material clinging to your body, or
the static you get after brushing hair. A laser printer uses this
same principle to glue ink on the paper.
2/22/2009 198
Printers
How Laser Printer works
Unlike the printers before, Laser printers use toner, static
electricity and heat to create an image on the paper. Toner
is dry ink. It contains colour and plastic particles. The toner
passes through the fuser in the computer and the resulting
heat binds it to any type of paper. Printing with laser printers
is fast and non-smudge and the quality is excellent because
of the high resolution that it can achieve with 300 dots per
inch to almost 1200 dpi at the higher end.
2/22/2009 199
Printers
Basic components of a laser printer are fuser, photoreceptor
drum assembly, developer roller, laser scanning unit, toner
hopper, corona wire and a discharge lamp. The laser beam
creates an image on the drum and wherever it hits, it
changes the electrical charge like positive or negative. The
drum then is rolled on the toner. Toner is picked up by
charged portion of the drum and gets transferred to the
paper after passing through the fuser. Fuser heats up the
paper to amalgamate ink and plastic in toner to create an
image. Laser printers are called "page printers" because
entire page is transferred to the drum before printing. Any
type of paper can be used in these printers. Laser printers
popularized DTP or Desk Top Publishing for it can print any
number of fonts and any graphics..
2/22/2009 200
Printers
This is how the computer and printer operate to print
When we want to print something we simply press the
command "Print". This information is sent to either RAM of
the printer or the RAM of the computer depending upon the
type of printer we have. The process of printing then starts.
While the printing is going on, our computer can still perform
a variety of operations. Jobs are put in a buffer or a special
area in RAM or Random Access Memory and the printer
pulls them off at its own pace. We can also line up our
printing jobs this way. This way of simultaneously performing
functions is called spooling. Our computer and the printer
are thus in constant communication.
2/22/2009 201
Printing Images
In image processing, there are overlapping terms that tend to get
interchanged. Especially for image and print resolution: dpi (dots per
inch), ppi (pixel or points per inch), lpi (lines per inch). In addition to
this, the resolution of an image is stated by its dimensions in pixels or in
inches (at a certain ppi or dpi resolution). Yes, we can understand if your
head is swimming. Let’s understand this:
When an image is captured using either a camera or a scanner, the result

is a digital image consisting of rows – known as arrays – of different
picture elements that are called pixels. This array has a horizontal and
vertical dimension. The horizontal size of the array is defined by the
number of pixels in one single row (say 1,280) and the number of rows
(say 1,024), giving the image a horizontal orientation. That picture would
have a “resolution” of “1,024 x 1,280 pixels”.
2/22/2009 202
Printing images
The size of the image displayed is dependent o the number of pixels the
monitor displays per inch. The “pixel per inch” resolutions (ppi) of
monitors vary, and are usually in the range of 72 ppi to 120 ppi (the
latter, lager 21.4” monitors). In most cases, however, with monitors the
resolution is given as the number of pixels horizontally and vertically
(e.g.1,0240 x 1,280 or 1,280 x 1,600). So the “size” of an image very
much depends on how many pixels are displayed per inch. Thus, we
come to a resolution given in ‘pixels per inch’ or ppi for short.
With LCD monitors, their ppi resolution is fixed and can’t be adjusted (at
least not without a loss of display quality). With CRT monitors you have
more flexibility (we won’t go into this further0.
When an image is printed, its physical size depends upon how many
image pixels we put down on paper, but also how an individual image
pixel is laid down on the paper.
2/22/2009 203
How image pixels produced by printer dots?
There are only a few printing technologies where a printer can directly produce a
continuous color range within an individual image pixel printed. Most other types of
printers reproduce the color of a pixel in an image by approximating the color by an n x
n matrix of fine dots using a specific pattern and a certain combination of the basic
colors available to the printer.
If we want to reproduce a pixel of an image on paper, we not only have to place a

physical printer’s ‘dot’ on paper, but also have to give that ‘dot’ the tonal value of the
original pixel. With bitonal images, that is easy. If the pixel value is o, you lay down a
black printed dot, and if the pixel is 1, you omit the dot. However, if the pixel has gray
value (say 128 out of 256), and you print with a black-and-white laser printer (just to
make the explanation a bit simpler), we must find different way. This technique is
called rasterization or dithering.
To simulate different tonal values (let’s just stick to black-and-white for the moment), a
number of printed dots are placed in a certain pattern on the paper to reproduce a single
pixel of the image. In a low-resolution solution, we could use matrix of 3 printed dots
by 3 printed dots per pixel.
2/22/2009 204
Using more printed dots per image pixel allows for more different tonal
values. With a pattern of 6 x 6 dots, you get 37 tonal grades, (which is
sufficient). For a better differentiation let’s call the matrix of printer dots
representing a pixel of the image a raster cell.
Now we see why a printer’s “dot per inch” (dpi) resolution has to be
much higher than the resolution has to be much higher than the
resolution of a display (where a single dot on a screen may be used to
reproduce a single pixel in an images, as the individual screen dot (also
called a pixel) may have different tonal (or brightness) values.
When you print with a device using relatively low resolution for
grayscale or colored images, you must make a trade-off between a high
resolution image (having as many “raster cells per inch” as possible) and
larger raster cells providing greater tonal value per cell.
2/22/2009 205
The image impression may be improved when the printer is able
to vary the size of its dots. This is done on some laser printers, as
well as with some of today’s photo inkjet printers. If the dot size
can be varied (also called modulated), fewer numbers of dots (n x
n) are needed to create a certain number of different tonal values,
(which results in a finer raster). You may achieve more tonal
values from a fixed raster cell size.
There are several different ways (patterns) to place single printed

dots in a raster cell, and the pattern for this dithering is partly a
secret of the printer driver. The dithering dot pattern is less visible
and more photo-like, when the pattern is not the same for all
raster cells having the same tonal values, but is modified from
raster cell to raster cell in some random way.
2/22/2009 206
Linear Systems
2/22/2009 207
Linear Space Invariant
System
2/22/2009 208
Linear Space Invariant
System
2/22/2009 209
This Property holds
2/22/2009 210
Convolution in 1
Dimension
Let’s look at some examples of convolution integrals,

∞
f (x) = g(x) ⊗ h(x) = ∫ g(x')h(x − x' )dx'
−∞
So there are four steps in calculating a convolution integral:
#1. Fold h(x’) about the line x’=0

#2. Displace h(x’) by x
#3. Multiply h(x-x’) * g(x’)
#4. Integrate
2/22/2009 211
Math of Convolution
n
g ( x ) = h * f ( x ) = ∑ h( n) f ( x − n)
−n
1 2 1
2/22/2009
h(-1)=1 h(0)=2 h(1)=1 212
Convolution (1D)
Filter coefficients (mask, kernel, template, window)
Filter Input Signal/Image-row

1 2 1
1 1 2 2 1 1 2 2 1 1
Output Signal/Image-row
Filter Response
5
2/22/2009 213
4
Math of 2D
Convolution/Correlation
n m
Convolution g ( x, y ) = h * f ( x, y ) = ∑∑ h(m, n) f ( x − m, y − n)
−n −m
n m
Correlation g ( x, y ) = h ο f ( x, y ) = ∑∑ h(m, n) f ( x + m, y + n)
−n −m
2/22/2009 214
Correlation (1D)
This process is called
Correlation!! 1 2 1
1 1 2 2 1 1 2 2 1 1
5 7 7 5 5 7 7 5
4 4 4 4 4 4 4 4
2/22/2009 215
Correlation Vs
Convolution
n
1 2 1 g ( x ) = h ο f ( x ) = ∑ h( n) f ( x + n)
−n
Correlation
1 1 2 2 1 1 2 2 1 1
n
1 2 1 g ( x ) = h * f ( x ) = ∑ h( n) f ( x − n)
−n
Convolution
1 1 2 2 1 1 2 2 1 1
In image processing we use CORRELATION
but (nearly) always call it CONVOLUTION!!!!!
Note: When the filter is symmetric: correlation = convolution!
2/22/2009 216
Correlation on images
Process of moving a filter
mask over the image and
compute sum of products at
each location. In convolution 1 1 1 1
filter is first rotated by 180 9
1 1 1
degree
1 1 1
Input Output
1 2 0 1 3
2 1 4 2 2
12
9
1 0 1 0 1
1 2 1 0 2
2 5 3 1 2
2/22/2009 217
Correlation on images
1 1 1 1
9
1 1 1
1 1 1
Input Output
1 2 0 1 3
2 1 4 2 2
12 11
9 9
1 0 1 0 1
1 2 1 0 2
2 5 3 1 2
2/22/2009 218
Applications of
Convolution/correlation
– Blurring
– Edge detection
–Template matching
2/22/2009 219
Blurring (smoothing)
Also know as: Smoothing kernel, Mean

filter, Low pass filter
The simplest filter: 1 1 1
1
– Spatial low pass filter 1 1 1
9
1 1 1 1
1 1 1
3
1 2 1
Another mask: 1
– Gaussian filter: 16
2 4 2
1 1 2 1
1 2 1
2/22/2009 220
4
Applications of smoothing
Blurring to remove
identity or other details
Degree of blurring = kernel
size
Show: camera, mean, convolution

2/22/2009 221
Applications of smoothing
Preprocessing: enhance objects
Smooth + Thresholding
2/22/2009 222
Uneven illumination
Improve segmentation
Uneven illumination
– Within an image
– Between images
Solution
– “Remove background”
– Algorithm: g(x,y) = f(x,y) – f(x,y), f(x,y)=mean
– Use a big kernel for f(x,y), e.g., 10-50
(IJ: mean=50, sub,TH)

2/22/2009 223
Uneven illumination
Input f(x,y) Mean f(x,y) f(x,y) – f(x,y) Edges
2/22/2009 224
Application of smoothing
Remove noise
2/22/2009 225
Correlation application:
Template Matching
2/22/2009 226
Template Matching
The filter is called a template or a mask
Input image Output Output as 3D
Template
The brighter the value in the output, the better the match
Implemented as the correlation coefficient
2/22/2009 227
Template Matching
Output
2/22/2009 228
Correlation application:
Edge detection
2/22/2009 229
Edge detection
Edge
detection
2/22/2009 230
Edge detection
g x ( x, y ) ≈ f ( x + 1, y ) − f ( x − 1, y )
g y ( x, y ) ≈ f ( x, y + 1) − f ( x, y − 1)
2/22/2009 231
Edge detection
g x ( x, y ) ≈ f ( x + 1, y ) − f ( x − 1, y )
g y ( x, y ) ≈ f ( x, y + 1) − f ( x, y − 1)
2/22/2009 232
Properties of convolution
commutative: f ⊗g=g⊗ f
associative: f ⊗ (g ⊗ h) = ( f ⊗ g) ⊗ h
multiple convolutions can be carried out in any order.

distributive:
2/22/2009 f ⊗ (g + h) = f ⊗ g + f ⊗ h 233
Convolution Theorem
ℑ{ f ⊗ g} = F(k)⋅ G(k)
In other words, convolution in real space is equivalent to

multiplication in Frequency space.
2/22/2009 234
Proof of convolution
Theorem
So we can rewrite the convolution integral,
∞
f ⊗g= ∫ f (x)g(x'−x)dx
as, −∞
1 ∞ ∞ ∞
f ⊗g= ∫ dx ∫ F(k)e dk ∫ G(k' )eik'( x'− x )dk'
ikx
4π 2 −∞ −∞ −∞
change the order of integration and extract a delta function,
1 ∞ ∞ 1 ∞
f ⊗g= ∫ dkF(k) ∫ dk'G(k')eik'x' ∫ eix(k −k')dx
2π −∞ −∞ 2π −∞
1 4 42 4 43
2/22/2009
δ (k−k') 235
theorem
1 ∞ ∞
ik'x' 1
∞
f ⊗g= ∫ dkF(k) ∫ dk'G(k')e ∫ eix(k −k')dx
2π −∞ −∞ 2π −∞
1 4 42 4 43
δ (k−k')
or,
1 ∞ ∞
f ⊗g= ∫ dkF(k) ∫ dk'G(k')eik'x'δ (k − k')
2π −∞ −∞
Integration over the delta function selects out the k’=k value.
1 ∞
2/22/2009
f ⊗g= ∫ dkF(k)G(k)eikx' 236
2π −∞
theorem
1 ∞
f ⊗g= ∫ dkF(k)G(k)eikx'
2π −∞
This is written as an inverse Fourier transformation. A Fourier

transform of both sides yields the desired result.
ℑ{ f ⊗ g} = F(k)⋅ G(k)
2/22/2009 237
Convolution in 2-D
For such a system the output h(x,y)
is the convolution of f(x,y) with
the impulse response g(x,y)
2/22/2009 238
Convolution in 2-D
2/22/2009 239
Example of 3x3
convolution mask
2/22/2009 240
In Plain Words
Convolution is essentially equivalent to

computing a weighted sum of image pixels
where filter is rotated 180 degree.
Convolution is Linear operation
2/22/2009 241
Why Mathematical
transformations?
Why
– To obtain a further information from the signal
that is not readily available in the raw signal.
Raw Signal
– Normally the time-domain signal
Processed Signal
– A signal that has been "transformed" by any of
the available mathematical transformations
Fourier Transformation
– The most popular transformation
2/22/2009 242
What is a Transform and
Why do we need one ?
Transform: A mathematical operation that takes a
function or sequence and maps it into another one
Transforms are good things because…
– The transform of a function may give additional /hidden
information about the original function, which may not be
available /obvious otherwise
– The transform of an equation may be easier to solve than
the original equation (recall your fond memories of Laplace
transforms in DFQs)
– The transform of a function/sequence may require less
storage, hence provide data compression / reduction
– An operation may be easier to apply on the transformed
function, rather than the original function (recall other 243
2/22/2009
fond memories on convolution).
Why transform ?
2/22/2009 244
Introduction to Fourier
Transform
f(x): continuous function of a real variable x
Fourier transform of f(x):

∞
ℑ{ f ( x)} = F (u ) = ∫ f ( x) exp[− j 2πux]dx Eq. 1
−∞
where j = −1
2/22/2009 245
transform
(u) is the frequency variable.
The integral of Eq. 1 shows that F(u) is

composed of an infinite sum of sine and
cosine terms and…
Each value of u determines the frequency of

its corresponding sine-cosine pair.
2/22/2009 246
transform
Given F(u), f(x) can be obtained by the
inverse Fourier transform:
−1
ℑ {F (u )} = f ( x)
∞
= ∫ F (u ) exp[ j 2πux]du
−∞
• The above two equations are the Fourier

transform pair.
2/22/2009 247
transform
jθ
e = cosθ + j sinθ
cos(−θ ) = cos(θ )
1 M −1
F(u) = ∑ f (x)[cos2πux/ M − j sin2πux/ M ]
M x=0
Each term of the FT (F(u) for every u)
is composed of the sum of all values of
f(x)
2/22/2009 248
Introduction to
Fourier transform
The Fourier transform of a real
function
function is
is generally
generally complex
complex and
and we
we
use
use polar
polar coordinates:
coordinates:
F ( u ) = R ( u ) + jI ( u )
jφ ( u )
F (u ) = F (u ) e
F ( u ) = [ R 2 ( u ) + I 2 ( u )] 1 / 2
⎡ I (u ) ⎤
φ ( u ) = tan −1
⎢ R (u ) ⎥
2/22/2009
⎣ ⎦ 249
Introduction to
Fourier transform
|F(u)| (magnitude function) is the Fourier
spectrum of f(x) and φ(u) its phase angle.
The square of the spectrum
P(u ) = F (u ) = R (u ) + I (u )
2 2 2
is referred to as the power spectrum of f(x)

(spectral density).
2/22/2009 250
Introduction to
Fourier transform
Fourier spectrum: [
F(u, v) = R (u, v) + I (u, v)
2 2
]
1/ 2
⎡ I (u , v ) ⎤
• Phase: φ (u , v ) = tan ⎢ −1
⎥
⎣ R (u , v ) ⎦
P(u,v) = F(u,v) = R (u,v) +I (u,v)
2 2 2
• Power spectrum:
2/22/2009 251
Spatial Frequency
decomposition
0.25µm myelin
• Any image can be
decomposed into a series of
sines and cosines added
together to give the image
I(x) = ∑ ai (cos k i x) + ibi (sin k i x)

i
Amplitudes Phase
Fourier Transform
-50 0 50 100 150 200 250 300
Pixel
-50 0 50 100 150 200 250 300
2/22/2009 252
Pixel
FT
Fourier Transform
of the Myelin Image
Low frequency
High frequency
2/22/2009 253
FT reversible
Fourier transform of myelin
F - 1
=
2/22/2009 254
2-D Image Transform
General Transform
N −1 N −1
F(u, v) = ∑∑T (x, y, u, v) f (x, y)
x=0 y=0
N −1 N −1
f ( x, y ) = ∑∑ I ( x, y, u, v) F (u, v)
u =0 v =0
2/22/2009 255
Discrete Fourier
Transform
2/22/2009 256
Discrete Fourier
Transform
A continuous function f(x) is discretized
into a sequence:
{f (x0), f (x0 +Δx), f (x0 +2Δx),...,f (x0 +[N −1]Δx)}
by taking N or M samples Δx units apart.
2/22/2009 257
Discrete Fourier
Transform
Where x assumes the discrete values

(0,1,2,3,…,M-1) then
f ( x) = f ( x0 + xΔx)
• The sequence {f(0),f(1),f(2),…f(M-1)}
denotes any M uniformly spaced samples
from a corresponding continuous function.
2/22/2009 258
Discrete Fourier
Transform
The discrete Fourier transform pair that
applies to sampled functions is given by:
M −1
F(u) = ∑ f (x)exp[− j2πux / M]
1
For u=0,1,2,…,M-1
M x= 0
and
M −1
f ( x) = ∑ f (u) exp[ j2πux / M ] For x=0,1,2,…,M-1
u= 0
2/22/2009 259
Discrete Fourier
Transform
To compute F(u) we substitute u=0 in the
exponential term and sum for all values of x
We repeat for all M values of u
It takes M*M summations and multiplications
M −1
F(u) = ∑ f (x)exp[−j2πux/ M]
1
For u=0,1,2,…,M-1
M x= 0
The Fourier transform and its inverse always

exist!
2/22/2009 260
Discrete Fourier
Transform
The values u = 0, 1, 2, …, M-1 correspond to

samples of the continuous transform at
values 0, Δu, 2Δu, …, (M-1)Δu.
i.e. F(u) represents F(uΔu), where:

1
Δu =
MΔx
2/22/2009 261
Discrete Fourier
Transform
In a 2-variable case, the discrete FT pair is:
1 M −1 N −1
F (u, v) = ∑∑
MN x=0 y =0
f ( x, y) exp[− j 2π (ux / M + vy / N )]
For u=0,1,2,…,M-1 and v=0,1,2,…,N-1
M −1 N −1
AND:
f (x, y) = ∑∑ F (u, v) exp[j2π (ux / M + vy / N)]
u=0 v=0
For x=0,1,2,…,M-1 and y=0,1,2,…,N-1

2/22/2009 262
Discrete Fourier
Transform
Sampling of a continuous function is now in a

2-D grid (Δx, Δy divisions).
The discrete function f(x,y) represents

samples of the function f(x0+xΔx,y0+yΔy)
for x=0,1,2,…,M-1 and y=0,1,2,…,N-1.
1 1
Δu = , Δv =
MΔx NΔy
2/22/2009 263
Discrete Fourier
Transform
When images are sampled in a square
array, M=N and the FT pair becomes:
N −1 N −1
1
F (u , v) =
N
∑∑ f ( x, y) exp[− j 2π (ux + vy) / N ]
x =0 y =0
For u,v=0,1,2,…,N-1
N −1 N −1
1
AND: f ( x, y ) =
N
∑∑ F (u, v) exp[ j 2π (ux + vy) / N ]
u =0 v =0
For x,y=0,1,2,…,N-1
2/22/2009 264
Properties of 2-D Fourier
transform
Translation
Distributivity and Scaling
Rotation
Periodicity and Conjugate Symmetry
Separability
Convolution and Correlation

2/22/2009 265
Translation
f (x,y)exp[j2π (u0 x /M + v0 y /N)]⇔F(u − u0,v −v0 )
and
f (x − x0,y − y0 ) ⇔F(u,v)exp[−j2π (ux0 /M + vy0 /N)]
2/22/2009 266
Translation
The previous equations mean:
– Multiplying f(x,y) by the indicated exponential

term and taking the transform of the product
results in a shift of the origin of the frequency
plane to the point (u0,v0).
– Multiplying F(u,v) by the exponential term shown

and taking the inverse transform moves the origin
of the spatial plane to (x0,y0).
– A shift in f(x,y) doesn’t affect the magnitude of its

Fourier transform
2/22/2009 267
Distributivity & Scaling
ℑ{ f1 ( x, y ) + f 2 ( x, y )} = ℑ{ f1 ( x, y )} + ℑ{ f 2 ( x, y )}
ℑ{ f1 ( x, y ) ⋅ f 2 ( x, y )} ≠ ℑ{ f1 ( x, y )} ⋅ ℑ{ f 2 ( x, y )}
Distributive over addition but not over

multiplication.
2/22/2009 268
Distributivity and Scaling
For two scalars a and b,
af ( x, y ) ⇔ aF (u, v)
1
f (ax, by ) ⇔ F (u / a, v / b)
ab
2/22/2009 269
Rotation
Polar coordinates:
x = r cosθ , y = r sin θ , u = ω cos ϕ , v = ω cos ϕ
Which means that:
f ( x, y ), F (u , v) become f (r ,θ ), F (ω , ϕ )
2/22/2009 270
Rotation
f (r ,θ + θ 0 ) ⇔ F (ω , ϕ + θ 0 )
Which means that rotating f(x,y) by an

angle θ0 rotates F(u,v) by the same
angle (and vice versa).
2/22/2009 271
Periodicity & Conjugate
Symmetry
The discrete FT and its inverse are

periodic with period N:
F(u,v)=F(u+M,v)=F(u,v+N)=F(u+M,v+N)
2/22/2009 272
Periodicity & conjugate
symmetry
Although F(u,v) repeats itself for infinitely many
values of u and v, only the M,N values of each
variable in any one period are required to obtain
f(x,y) from F(u,v).
This means that only one period of the transform is

necessary to specify F(u,v) completely in the
frequency domain (and similarly f(x,y) in the spatial
domain).
2/22/2009 273
Symmetry
(shifted spectrum)
move the origin of the
transform to u=N/2.
2/22/2009 274
Symmetry
For real f(x,y), FT also exhibits conjugate
symmetry:
F (u , v) = F * (−u ,−v)
F (u , v) = F (−u ,−v)
F (u ) = F (u + N )
or
F (u ) = F (−u )
• i.e. F(u) has a period of length N and the

magnitude of the transform is centered on
the origin.
2/22/2009 275
Separability
The discrete FT pair can be expressed

in separable forms which (after some
manipulations) can be expressed as:
1 M −1
F(u,v) = ∑
M x= 0
F(x,v)exp[− j2πux / M]
⎡ 1 N −1 ⎤
Where: F(x,v) = ⎢ ∑ f (x, y)exp[− j2πvy /N]⎥
⎢⎣ N y= 0 ⎥⎦
2/22/2009 276
Separability in Specific
forms
Separable
T ( x, y , u , v ) = T1 ( x, u )T2 ( y , v )
Symmetric
T ( x, y , u , v ) = T1 ( x, u )T2 ( y , v )
2/22/2009 277
Separability
For each value of x, the expression

inside the brackets is a 1-D transform,
with frequency values v=0,1,…,N-1.
Thus, the 2-D function F(x,v) is

obtained by taking a transform along
each row of f(x,y) and multiplying the
result by N.
2/22/2009 278
Separability
The desired result F(u,v) is then

obtained by making a transform along
each column of F(x,v).
2/22/2009 279
Energy preservation
g = f
2 2
N −1N −1 N −1N −1
∑ ∑ f ( x, y ) = ∑ ∑ g (u , v)
2 2
x =0 y =0 u =0 v =0
2/22/2009 280
Energy Compaction !
2/22/2009 281
An Atom
Both functions have circular symmetry.
The atom is a sharp feature, whereas its
transform is a broad smooth function. This
illustrates the reciprocal relationship between a
function and its Fourier transform.
2/22/2009 282
Original Image – FA-FP
2/22/2009 283
A Molecule
2/22/2009 284
Fourier duck
2/22/2009 285
Reconstruction from
Phase of cat & Amplitude
of duck
2/22/2009 286
Reconstruction from
Phase of duck &
Amplitude of cat
2/22/2009 287
Original Image-Fourier Amplitude
Keep Part of the Amplitude Around the Origin and Reconstruct
Original Image (LOW PASS filtering)
2/22/2009 288
Keep Part of the Amplitude Far from the Origin and
Reconstruct Original Image (HIGH PASS filtering)
2/22/2009 289
Example
2/22/2009 290
Reconstruction from
phase of one image and amplitude of the other
2/22/2009 291
Example
2/22/2009 292
Reconstruction from
phase of one image and amplitude of the other
2/22/2009 293
Reconstruction
Example
Cheetah Image
Fourier Magnitude (above)
Fourier Phase (below)
2/22/2009 294
Reconstruction
example
Zebra Image
Fourier Magnitude (above)
Fourier Phase (below)
2/22/2009 295
Reconstruction
Reconstruction with
Zebra phase,
Cheetah Magnitude
2/22/2009 296
Reconstruction
Reconstruction with
Cheetah phase,
Zebra Magnitude
2/22/2009 297
2/22/2009 298
Optical illusion
2/22/2009 299
Optical illusion
2/22/2009 300
Optical illusion
2/22/2009 301
Optical illusion
2/22/2009 302
Optical illusion
2/22/2009 303
Optical illusion
2/22/2009 304
Optical illusion
2/22/2009 305
Optical illusion
2/22/2009 306
Optical illusion
2/22/2009 307
Optical illusion
2/22/2009 308
Optical illusion
2/22/2009 309
2/22/2009 310
Optical illusion
2/22/2009 311
Optical illusion
2/22/2009 312
2/22/2009 313
Optical illusion
2/22/2009 314
Optical illusion
2/22/2009 315
Discrete Cosine
Transform 1-D
N −1
⎡ (2 x + 1)uπ ⎤
C (u ) = a (u ) ∑ f ( x) cos
x =0 ⎣⎢ 2 N ⎥⎦
u = 0,1,Κ , N − 1
⎧ 1
⎪ u=0
⎪ N
a (u ) = ⎨
⎪ 2
2/22/2009
⎪⎩ u = 1,Κ , N − 1 316
N
IDCT – 1D
N −1
⎡ (2 x + 1)uπ ⎤
f ( x) = ∑ a (u )C (u ) cos
u =0 ⎢⎣ 2 N ⎥⎦
2/22/2009 317
1D Basis functions N=8
u=0 u=1 u=2 u=3
1.0 1.0 1.0 1.0
0.5 0.5 0.5 0.5
0 0 0 0
-0.5 -0.5 -0.5 -0.5
-1.0 -1.0 -1.0 -1.0
u=4 u=5 u=6 u=7

1.0 1.0 1.0 1.0
0.5 0.5 0.5 0.5
0 0 0 0
-0.5 -0.5 -0.5 -0.5
-1.0 -1.0 -1.0 -1.0
2/22/2009 318
1D Basis functions N=16
2/22/2009 319
Example : 1D signal
2/22/2009 320
DCT
2/22/2009 321
2-D DCT
N −1N −1
⎡ ( 2 x + 1)uπ ⎤ ⎡ ( 2 y + 1)vπ ⎤
C (u , v ) = a (u ) a (v ) ∑ ∑ f ( x, y ) cos cos
x =0 y =0 ⎣⎢ 2N ⎥⎦ ⎣⎢ 2N ⎥⎦
N −1N −1
⎡ ( 2 x + 1)uπ ⎤ ⎡ (2 y + 1)vπ ⎤
f ( x, y ) = ∑ ∑ a (u )a (v )C (u , v) cos cos
u =0 v =0 ⎢⎣ 2 N ⎥⎦ ⎢⎣ 2 N ⎥⎦
u, v = 0,1,Κ , N − 1
2/22/2009 322
Advantages
Notice that the DCT is a real transform.
The DCT has excellent energy compaction

properties.
There are fast algorithms to compute the

DCT similar to the FFT.
2/22/2009 323
2-D Basis functions N=4
v
0 1 2 3
3
2/22/2009 324
2-D Basis functions N=8
2/22/2009 325
Separable
2/22/2009 326
Example: Energy
Compaction
2/22/2009 327
Relation between DCT &
DFT
g ( x ) = f ( x ) + f (2 N − 1 − x )
⎧ f ( x ), 0 ≤ x ≤ N −1
=⎨
⎩ f ( 2 N − 1 − x ), N ≤ x ≤ 2 N − 1
N − point 2 N − point DFT 2 N − point N − point

f ( x) → g( x) → G (u ) → C f (u )
2/22/2009 328
DFT to DCT
F ro m D F T to D C T (C o n t.)
D C T has a higher com pression ration than D FT

- D C T avoids the generation of spurious spectral
com ponents
2/22/2009 329
December 21 1807
illusion
“An arbitrary function, continuous or with

discontinuities, defined in a finite interval by an
arbitrarily capricious graph can always be
expressed as a sum of sinusoids”
J.B.J. Fourier
Jean B. Joseph Fourier

(1768-1830)
2/22/2009 330
Frequency analysis
Frequency Spectrum
– Be basically the frequency components (spectral
components) of that signal
– Show what frequencies exists in the signal
Fourier Transform (FT)
– One way to find the frequency content
– Tells how much of each frequency exists in a
signal
N −1 ∞
X (k + 1 ) = ∑ x (n + 1 ) ⋅ W
n=0
N
kn
X(f )= ∫ x(t ) ⋅ e
− 2 jπft
dt
N −1 −∞
1 ∞
x (n + 1) = ∑ X (k + 1) ⋅W N− kn
N k =0 x(t ) = ∫ X ( f ) ⋅ e 2 jπft
df
2π ⎞
− j ⎛⎜ ⎟ −∞
2/22/2009 w N = e ⎝ N ⎠ 331
Complex Function = ∑ (weight )i • (Simple Function )i
i
Complex function representation through simple
building blocks
– Basis functions
Using only a few blocks Î Compressed

representation
Using sinusoids as building blocks Î Fourier

transform
– Frequency domain representation of the function
− jωt 1 jωt
F(ω) = ∫ f (t)e
2/22/2009 dt f (t) = ∫ F(ω)e dω 332
2π
How does it work
Anyway?
Recall that FT uses complex exponentials (sinusoids)
as building blocks.
e j ω t = cos (ω t ) + j sin (ω t )
For each frequency of complex exponential, the
sinusoid at that frequency is compared to the signal.
If the signal consists of that frequency, the

correlation is high Æ large FT coefficients.
− jω t 1 j ωt
F (ω ) = ∫ f (t )e dt f (t ) = ∫ F (ω ) e dω
2π
If the signal does not have any spectral component
at a frequency, the correlation at that frequency is
2/22/2009 333
low / zero, Æ small / zero FT coefficient.

FT At work
x1 (t ) = cos(2π ⋅ 5 ⋅ t )
x2 (t ) = cos(2π ⋅ 25 ⋅ t )
x3 (t ) = cos(2π ⋅ 50 ⋅ t )
2/22/2009 334
FT At work
x1 (t ) F X 1 (ω )
x2 (t ) F X 2 (ω )
x3 (t ) F X 3 (ω )
2/22/2009 335
FT At work
x4 (t ) = cos(2π ⋅ 5 ⋅ t )
+ cos(2π ⋅ 25 ⋅ t )
+ cos(2π ⋅ 50 ⋅ t )
x4 (t ) F X 4 (ω )
2/22/2009 336
FT At work
Complex exponentials
(sinusoids) as basis
functions:
∞
∫ f (t ) ⋅ e
− jωt
F (ω ) = dt
−∞
∞
1
∫ F (ω ) ⋅ e
jωt
f (t ) = dt
2π −∞
2/22/2009 337
Stationarity of Signal
Stationary Signal
– Signals with frequency content unchanged
in time
– All frequency components exist at all times
Non-stationary Signal
– Frequency changes in time
– One example: the “Chirp Signal”
2/22/2009 338
Stationary & Non
Stationary Signals
FT identifies all spectral components present in the signal, however it does
not provide any information regarding the temporal (time) localization of
these components. Why?
Stationary signals consist of spectral components that do not change in time
– all spectral components exist at all times
– no need to know any time information
– FT works well for stationary signals
However, non-stationary signals consists of time varying spectral
components
– How do we find out which spectral component appears when?
– FT only provides what spectral components exist , not where in time
they are located.
– Need some other ways to determine time localization of spectral
components
2/22/2009 339
Stationary & Non Stationary Signals
Stationary signals’ spectral characteristics do not
change with time
x4 (t ) = cos(2π ⋅ 5 ⋅ t )
+ cos(2π ⋅ 25 ⋅ t )
+ cos(2π ⋅ 50 ⋅ t )
Non-stationary signals have time varying spectra
x5 (t ) = [ x1 ⊕ x2 ⊕ x3 ]
⊕ Concatenation
2/22/2009 340
Stationary &
Nonstationary signals
2 Hz + 10 Hz + 20Hz 3 6 0 0
2 5 0 0
Magnitud
Magnitud
1 4 0 0
Stationary
e
0 3 0 0
e
-1 2 0 0
-2 1 0 0
-3 0
0 0 .2 0 .4 0 .6 0 .8 1 0 5 1 0 1 5 2 0 2 5
Time Frequency (Hz)
0.0-0.4: 2 Hz + 1 2 5 0
0 .8
0.4-0.7: 10 Hz +
Magnitud
0 .6 2 0 0
Magnitud
0.7-1.0: 20Hz 0 .4
0 .2 1 5 0
Non-
e
-0 .2 e 1 0 0
Stationary -0 .4
-0 .6 5 0
2/22/2009 -0 .8
341
-1 0
0 0 .5 1 0 5 1 0 1 5 2 0 2 5
Time Frequency (Hz)
Chirp signals
Frequency: 2 Hz to 20 Hz Different in Time DomainFrequency: 20 Hz to 2 Hz
1 150 1 150
0.8 0.8
0.6 0.6
0.4 0.4
Magnitud
Magnitud
100 100
Magnitud
Magnitud
0.2 0.2
0 0
e
e
e
e
-0.2 -0.2
50 50
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 0 -1 0
0 0.5 1 0 5 10 15 20 25 0 0.5 1 0 5 10 15 20 25
Time Frequency (Hz) Time Frequency (Hz)
Same in Frequency Domain

2/22/2009 342
Atwhat
At whattime
timethe
thefrequency
frequencycomponents
componentsoccur?
occur? FT
FTcan
cannot
nottell!
tell!
Stationary & Non Stationary Signals
Perfect knowledge of what

frequencies exist, but no
information about where
these frequencies are
located in time
2/22/2009 343
FFT Vs Wavelet
FFT, basis functions: sinusoids

Wavelet transforms: small waves, called
wavelet
FFT can only offer frequency
information
Wavelet: frequency + temporal
information
Fourier analysis doesn’t work well on
2/22/2009 344
discontinuous, “bursty” data

Fourier Vs. Wavelet
Fourier
– Loses time (location) coordinate completely
– Analyses the whole signal
– Short pieces lose “frequency” meaning
Wavelets
– Localized time-frequency analysis
– Short signal pieces also have significance
– Scale = Frequency band
2/22/2009 345
Shortcomings of FT
y Sinusoids and exponentials
– Stretch into infinity in time, no time localization
– Instantaneous in frequency, perfect spectral
localization
– Global analysis does not allow analysis of non-stationary
signals
y Need a local analysis scheme for a time-frequency
representation (TFR) of nonstationary signals
– Windowed F.T. or Short Time F.T. (STFT) : Segmenting
the signal into narrow time intervals, narrow enough to
be considered stationary, and then take the Fourier
transform of each segment, Gabor 1946.
– 2/22/2009
Followed by other TFRs, which differed from each other
346
by the selection of the windowing function
Nothing More Nothing
less
FT Only Gives what Frequency Components Exist in
the Signal
The Time and Frequency Information can not be
Seen at the Same Time
Time-frequency Representation of the Signal is
Needed
Most of Transportation Signals are Non-stationary.

(We need to know whether and also when an incident was happened.)
ONE EARLIER
2/22/2009 SOLUTION: SHORT-TIME FOURIER TRANSFORM (STFT)347
Short Time Fourier
Transform -STFT
1. Choose a window function of finite length
2. Place the window on top of the signal at t=0
3. Truncate the signal using this window
4. Compute the FT of the truncated signal,
save.
5. Incrementally slide the window to the right
6. Go to step 3, until window reaches the end
of the signal
For each time location where the window is
centered, we obtain a different FT
– Hence, each FT provides the spectral information
of a separate time-slice of the signal, providing 348
2/22/2009
simultaneous time and frequency information

STFT
Time Frequency Signal to FT Kernel

parameter parameter be analyzed (basis function)
∫ [x(t ) ⋅W (t − t )]⋅ e dt
ω ′ − j ωt
STFTx (t , ω ) = ′
t
STFT of signal x(t): Windowing Windowing function

Computed for each function centered at t=t’
window centered at t=t’
2/22/2009 349
STFT
t’=-8 t’=-2
t’=4 t’=8
2/22/2009 350
STFT
STFT provides the time information by computing a
different FTs for consecutive time intervals, and then
putting them together
– Time-Frequency Representation (TFR)
– Maps 1-D time domain signals to 2-D time-frequency signals
Consecutive time intervals of the signal are obtained
by truncating the signal using a sliding windowing
function
How to choose the windowing function?
– What shape? Rectangular, Gaussian, Elliptic…?
– How wide?
Wider window require less time steps Æ low time resolution
Also, window should be narrow enough to make sure that the portion of the
signal falling within the window is stationary
Can we choose an arbitrarily narrow window…?
2/22/2009 351
Selection of STFT Window
STFTxω (t ′, ω ) = ∫ [x(t ) ⋅ W (t − t ′)]⋅ e − jωt dt
t
Two extreme cases:
W(t) infinitely long: W (t ) = 1 Î
STFT turns into FT, providing excellent
frequency information (good frequency
resolution), but no time information
W(t) infinitely short: W (t ) = δ (t )
STFTxω (t ′, ω ) = ∫ [x(t ) ⋅ δ (t − t ′)]⋅ e − jωt dt = x(t ′) ⋅ e − jωt ′

t
Î STFT then gives the time signal back, with
a phase factor. Excellent time information
(good time resolution), but no frequency
2/22/2009 352
information
Drawbacks of STFT
2/22/2009 353
Drawbacks of STFT
Unchanged Window
Dilemma of Resolution
– Narrow window -> poor frequency
resolution
– Wide window -> poor time resolution
Heisenberg Uncertainty Principle
– Cannot know what frequency exists
at what time intervals
2/22/2009 354
Heisenberg principle
1
Δt ⋅ Δf ≥
4π
Time resolution: How well Frequency resolution: How

two spikes in time can be well two spectral components
separated from each other in can be separated from each
the transform domain other in the transform domain
Both time and frequency resolutions cannot be arbitrarily high!!!

Î ÎWe cannot precisely know at what time instance a frequency component is
located. We can only know what interval of frequencies are present in which time
intervals
2/22/2009 355
Drawbacks of STFT
2/22/2009 T 356
Multiresolution analysis
Wavelet Transform
– An alternative approach to the short time Fourier transform
to overcome the resolution problem
– Similar to STFT: signal is multiplied with a function
Multiresolution Analysis
– Analyze the signal at different frequencies with different
resolutions
– Good time resolution and poor frequency resolution at high
frequencies
– Good frequency resolution and poor time resolution at low
frequencies
– More suitable for short duration of higher frequency; and
longer duration of lower frequency components
2/22/2009 357
Wavelet Definition
“The wavelet transform is a tool that

cuts up data, functions or operators into
different frequency components, and
then studies each component with a
resolution matched to its scale”
2/22/2009 358
Principles of wavelet
transform
Split Up the Signal into a Bunch of
Signals
Representing the Same Signal, but all
Corresponding to Different Frequency
Bands
Only Providing What Frequency Bands
Exists at What Time Intervals
2/22/2009 359
The wavelet transform
Overcomes the preset resolution problem of the STFT
by using a variable length window
Analysis windows of different lengths are used for
different frequencies:
– Analysis of high frequenciesÎ Use narrower
windows for better time resolution
– Analysis of low frequencies Î Use wider windows
for better frequency resolution
This works well, if the signal to be analyzed mainly
consists of slowly varying characteristics with
occasional short high frequency bursts.
Heisenberg principle still holds!!!
The function used to window the signal is called the
wavelet
2/22/2009 360
Wavelet transform
Scale and shift original waveform

Compare to a wavelet
Assign a coefficient of similarity
2/22/2009 361
Definition of continuous wavelet
transform
Translation parameter, Scale parameter, A normalization
(location of window) measure of frequency constant Signal to be
analyzed
ψ ψ ∗⎛ t −τ ⎞
∫ x(t )ψ
1
CWTx (τ , s ) = Ψx (τ , s ) = ⎜ ⎟dt
s t ⎝ s ⎠
Continuous wavelet transform The mother wavelet. All kernels are

of the signal x(t) using the obtained by translating (shifting) and/or
analysis wavelet ψ(.) scaling the mother wavelet
Wavelet
– Small wave
Scale = 1/frequency
– Means the window function is of finite length
Mother Wavelet
– A prototype for generating the other window functions
– 2/22/2009
All the used windows are its dilated or compressed and 362
shifted versions
CWT
for each Scale

for each Position

Wavelet (S,P)
∫
Coefficient (S,P) = Signal x
all time
end
end
Coefficient Scale
2/22/2009 363
Scaling-- value of “stretch”
Scaling a wavelet simply means
stretching (or compressing) it.
f(t) = sin(t) f(t) = sin(2t) f(t) = sin(3t)
scale factor 2 scale factor 3
scale factor1
2/22/2009 364
Scale
•It lets you either narrow down the frequency band of interest, or
•determine the frequency content in a narrower time interval
•Scaling = frequency band
•Good for non-stationary data
Scale
– S>1: dilate the signal
– S<1: compress the signal
High Scale -> a Stretched wavelet -> Non-detailed
Global View of Signal -> Span Entire Signal –> Low
Frequency -> Slowly changing, coarse features
Low Scale -> a Compressed Wavelet -> Rapidly
Changing details -> High Frequency -> Detailed View
Last in Short Time
Only Limited Interval of Scales is Necessary
2/22/2009 365
Scale is (sort of) like
frequency
Small scale
-Rapidly changing details,
-Like high frequency
Large scale
-Slowly changing
details
-Like low frequency
2/22/2009 366
Scale is (sort of ) like
frequency
The scale factor works exactly the same with

wavelets. The smaller the scale factor, the
more "compressed" the wavelet.
2/22/2009 367
Shifting
Shifting a wavelet simply means delaying (or hastening) its
onset. Mathematically, delaying a function f(t) by k is
represented by f(t-k)
2/22/2009 368
Shifting
C = 0.0004
C = 0.0034
2/22/2009 369
Computation of CWT
ψ ψ ∗⎛ t −τ ⎞
∫ x(t )ψ
1
CWTx (τ , s ) = Ψx (τ , s ) = ⎜ ⎟dt
s t ⎝ s ⎠
Step 1: The wavelet is placed at the beginning of the signal, and set s=1 (the most
compressed wavelet);
Step 2: The wavelet function at scale “1” is multiplied by the signal, and integrated
over all times; then multiplied by ;
Step 3: Shift the wavelet to t= , and get the transform value at t= and s=1;
Step 4: Repeat the procedure until the wavelet reaches the end of the signal;
Step 5: Scale s is increased by a sufficiently small value, the above procedure is
repeated for all s;
Step 6: Each computation for a given s fills the single row of the time-scale plane;
Step 7: CWT is obtained if all s are calculated.
2/22/2009 370
Simple steps for CWT
1. Take a wavelet and compare it to a section at the start of
the original signal.
2. Calculate a correlation coefficient c
2/22/2009 371
Simple steps to CWT
3. Shift the wavelet to the right and repeat steps 1 and 2 until
you've covered the whole signal.
4. Scale (stretch) the wavelet and repeat steps 1 through 3.
5. Repeat steps 1 through 4 for all scales.
2/22/2009 372
CWTxψ (τ , s ) = Ψψ
x (τ , s ) =
1
∫
s t
x (t )
ψ ∗⎛ t −τ ⎞
⎜
⎝ s
⎟dt
⎠ WT At work
Lowfrequency
Low frequency(large
(large scale)
scale)
2/22/2009 373
WT At work
2/22/2009 374
WT At work
2/22/2009 375
WT At work
2/22/2009 376
Resolution of Time &
Frequency
Better time
resolution;
Poor
frequency
resolution
Frequenc
y
Better
frequency
resolution;
Poor time Time
resolution • Each box represents a equal portion
2/22/2009 • Resolution in STFT is selected once for entire 377
analysis
Comparison of
Transformations
2/22/2009 378
From http://www.cerm.unifi.it/EUcourse2001/Gunther_lecturenotes.pdf, p.10
Discretization of CWT
It is Necessary to Sample the Time-Frequency (scale) Plane.
At High Scale s (Lower Frequency f ), the Sampling Rate N can be
Decreased.
The Scale Parameter s is Normally Discretized on a Logarithmic Grid.
The most Common Value is 2.
The Discretized CWT is not a True Discrete Transform
Discrete Wavelet Transform (DWT)

– Provides sufficient information both for analysis and synthesis
– Reduce the computation time sufficiently
– Easier to implement
– Analyze the signal at different frequency bands with different resolutions
– Decompose the signal into a coarse approximation and detail information
2/22/2009 379
Discrete Wavelet
transforms
CWT computed by computers is really not CWT, it is a
discretized version of the CWT.
The resolution of the time-frequency grid can be
controlled (within Heisenberg’s inequality), can be
controlled by time and scale step sizes.
Often this results in a very redundant representation
How to discretize the continuous time-frequency plane,
so that the representation is non-redundant?
– Sample the time-frequency plane on a dyadic (octave)
grid ψ ∗⎛ t −τ ⎞
∫ x(t )ψ
ψ 1
CWTx (τ , s ) = Ψx (τ , s ) = ⎜ ⎟dt
s t ⎝ s ⎠
2/22/2009 (
ψ kn (t ) = 2 − k ψ 2 − k − n ) ∈Z
k , n380
Multiresolution analysis
Analyzing a signal both in time domain and
frequency domain is needed many a times
– But resolutions in both domains is limited by
Heisenberg uncertainty principle
Analysis (MRA) overcomes this , how?
– Gives good time resolution and poor frequency
resolution at high frequencies and good frequency
resolution and poor time resolution at low
frequencies
– This helps as most natural signals have low
frequency content spread over long duration and
high frequency content for short durations
2/22/2009 381
Discrete wavelet
transform
signal
lowpass highpass
filters
Approximation Details
(a) (d)
2/22/2009 382
Discrete wavelet
transform
Dyadic sampling of the time –frequency plane results

in a very efficient algorithm for computing DWT:
– Subband coding using multiresolution analysis
– Dyadic sampling and multiresolution is achieved
through a series of filtering and up/down sampling
operations
x[n] H y[n]
y[n ] = x[n ] * h[n ] = h[n ] * x[n ]

N
= ∑ x[k ] ⋅ h[n − k ]
k =1
N
2/22/2009 = ∑ h[k ] ⋅ x[n − k ] 383
k =1
DWT implementation
~
x[n]
yhigh [k ] = ∑ x[n] g[− n + 2k ] ∑ yhigh [k ] ⋅ g[−n + 2k ] x[n]
n k
~
G G +
2 2
~ 2 ~ 2
H G 2 2 G + H
~ 2 2 H
∑ yhigh [k ] ⋅ g[−n + 2k ]
~ H
ylow[k ] = ∑ x[n] h[− n + 2k ]
n k
Decomposition Reconstruction
G Half band high pass filter 2 Down-sampling

H Half band low pass filter 2 Up-sampling
2-2/22/2009
level DWT decomposition. The decomposition can be continues as long
384
as there are enough samples for down-sampling.
DWT - Demystified
g[n] h[n]
Length: 256
Length: 256 2 2 B: 0 ~ π/2 Hz
B: π/2 ~ π Hz
a1 |G(jw)|
d1: Level 1
DWT
Coeff. g[n] h[n]
Length: 128
w
Length: 128 2 2 B: 0 ~ π /4 Hz -π -π/2 π/2 π
B: π/4 ~ π/2 Hz a2
d2: Level 2
DWT g[n] h[n]
Coeff.
2 2 Length: 64
Length: 64
B: 0 ~ π/8 Hz
B: π/8 ~ π/4 Hz
d3: Level 3 …a3….Level 3 approximation
2/22/2009 DWT 385
Coefficients
Coeff.
2D DWT
Generalization of concept to 2D
2D functions ÍÎ images f(x,y) ÍÎ
I[m,n] intensity function
Why would we want to take 2D-DWT
of an image anyway?
– Compression
– Denoising
– Feature extraction
Mathematical ∞form ∞
f o ( x , y ) = ∑ ∑ a o ( i , j ) ⋅ sφφ ( x − i , y − j)
i = −∞ j = −∞
a o ( i , j ) =< f ( x , y ), sφφ ( x − i , y − j ) >
2/22/2009 386
sφφ ( x, y ) = φ ( x) ⋅ φ ( y ) sψψ ( x, y ) = ψ ( x) ⋅ψ ( y )
Implementation of 2d
DWT
~
LL Ak +1
COLUMNS
H 1 2
ROWS ~ 2 1
H
COLUMNS
…… ~ ( h)
G 1 2 D
LH k +1
ROWS
……
INPUT COLUMNS
IMAGE ~
H 1 2 HL Dk(v+)1
~ 2 1
G
ROWS
~ (d )
G 1 2 D
HH k +1
COLUMNS
LLL LLH LLH

LL LH LH LH
INPUT LHL LHH LL
LHL LHH
IMAGE
HL HH HL HH HL HH
2/22/2009 387
Up and down …. Up and
down
Downsample columns along the rows: For each row, keep the
2 1 even indexed columns, discard the odd indexed columns
Downsample columns along the rows: For each column, keep the
1 2 even indexed rows, discard the odd indexed rows
Upsample columns along the rows: For each row, insert zeros at
2 1 between every other sample (column)
Upsample rows along the columns: For each column, insert zeros
1 2 at between every other sample (row)
2/22/2009 388
Reconstruction
LL Ak +1 1 2 H
2 1 H
LH D ( h) 1 2 G
k +1 ORIGINAL
IMAGE
HL D (v )
k +1 1 2 H
2 1 G
HH Dk( d+)1 1 2 G
2/22/2009 389
Subband coding
algorithm
Halves the Time Resolution
– Only half number of samples resulted
Doubles the Frequency Resolution
– The spanned frequency band halved
0-1000 Hz
256
X[n] Filter 1 D1: 500-1000 Hz
512
256
S
S Filter 2 D2: 250-500 Hz
A1 128
A1 D1 128
Filter 3 D3: 125-250 Hz
A2 64
A2 D2
A3 2/22/2009
D3 A3: 0-125 Hz390
64
Application of wavelets
Compression
De-noising
Feature Extraction
Discontinuity Detection
Distribution Estimation
Data analysis
– Biological data
– NDE data
– Financial data
2/22/2009 391
Fingerprint compression
Wavelet:
Haar
Level:3
2/22/2009 392
Image denoising using
wavelet transform
Image de-noising using wavelet transform:
Utilize the same principles as for signal decomposition and de-noising.
Each column of an image matrix is convolved with high-pass and low-pass filter
followed by downsampling.
The same process is applied to image matrix rows.
The choice of threshold limits δ for each decomposition level and modification of
its coefficients for k=0, 1, … N-1
Backward image
| c(k ) |> δ out of

if reconstruction
) |> δ
| c(kwavelet
ifmodified
transform coefficients
2/22/2009 393
Image enhancement
using wavelet transform
2/22/2009 394

Introduction To DIP

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Introduction To DIP

Transféré par

Droits d'auteur :

Formats disponibles

Image Processing and the

 Image processing refers to a set of

Image Analysis/ Image Coding/

Each component in the image ⎡10 10 16 28⎤

Gray scale values

1 0.1 0.5 0.3

we’ll have a look at the human eye

•IIs able to perceive electromagnetic waves in a

•IIs able to distinguish between wavelengths in this

•HHas a higher density of receptors in the center

•Maps our 3D reality to a 2 dimensional

Each cell contains a numerical value (e.g.

•TThe numerical value of the cell represents the

•VValue = 0: BLACK (no illumination / energy)

What to do to get info on 3D world?

– The geometry of image formation,

– The physics of light, which determines

 Digital image generation is first step in any

– The first task is related to the camera

– The second task is related to surface

From the two similar triangles (A’B’P’) and

From the last two equations, perspective

Long focal length -> narrow field of view

The scaling factor m is usually set to 1 or –1 for

 Foreshortening and Solid angle

 Special cases and simplifications : Lambertain,

Two sources that look the same to a receiver must

Measured in steradians (sr)

 Irradiance (E): energy arriving at a surface

Total power = integrate irradiance

 Radiance leaving a surface in a particular direction:

= ∫ ρ (x, θ i ' , φi ' ,θe' , φe ' ) L( Rα , β (θ i ' , φi ' )) cos(θ i )dωi

Reflection behaves like

Image irradiance = 1 * scene

Surface normal ( n ) – It is directional vector with magnitude

If the surface is smooth, simply neglect higher order terms

Cross product of OA & OB vectors gives the surface normal

 Dull, matte surfaces like chalk or latex paint

Irradiance Unit incoming

Total Energy leaving ⎡ ⎤

 Let’s consider the case of reflection in detail

rays originating from a particular point on the

When a camera captures the image of an

Where integration is over solid angle occupied

By definition, the image irradiance is the

- Let Ѳr’ be the angle between the surface

From the previous equations

The solid angle occupied by the lens as seen

by the distance Cos f ( α )

That is, the image irradiance is proportional to

 Analog images are continuous

object observe digitize store process Refresh

 assume the image is bounded by a 200

rectangle [0,a]×[0,b] 150

 The analog signal representing a

 To create a digital image, we need to

The original analog Measurements are Discrete samples

 The assignment is made depending on the

Fig. 3.2 (a) Continuous image projected onto a sensor array.

• SSampling means measuring the value of an

Fig. 3.3. Coordinate convention used to represent

Fig. 3.4. A digital image of size M x N

Image processing refers to a set of

Digital image generation is first step in any

Foreshortening and Solid angle

Special cases and simplifications : Lambertain,

Measured in steradians (sr)

Irradiance (E): energy arriving at a surface

Radiance leaving a surface in a particular direction:

Dull, matte surfaces like chalk or latex paint

Let’s consider the case of reflection in detail

Analog images are continuous

assume the image is bounded by a 200

The analog signal representing a

To create a digital image, we need to

The assignment is made depending on the

Monitor pixels are actually circular light

110- and 12- bit pixel can display 1024 and

IIncreasing pixel bit depth improves image

Number of bits to represent pixel color

If the light is achromatic its only attribute is its

(Left) Image of food originating from a digital camera.

Also know as: Smoothing kernel, Mean

f(x): continuous function of a real variable x

Fourier transform of f(x):

(u) is the frequency variable.

The integral of Eq. 1 shows that F(u) is

Each value of u determines the frequency of

The square of the spectrum

Where x assumes the discrete values

The Fourier transform and its inverse always

The values u = 0, 1, 2, …, M-1 correspond to

i.e. F(u) represents F(uΔu), where:

Sampling of a continuous function is now in a

The discrete function f(x,y) represents

Distributive over addition but not over