21 x 21

Module-1

Digital Image Fundamentals

 • Wh t is n im ge?

• Wh t is digit l im ge?

• Wh t is digit l im ge processing?

• Pixel?

• Im ges from di erent sources - (Electrom gnetic Spectrum)

• Im ge Processing, Where it st rts nd where it goes?

• Three types of computerised process in im ge processing

a
a
a
a
a
a
a
a
a
ff
a
a
a
a
a
a
What is an image?

• Its 2 dimension l function, f(x, y)

where, x nd y re sp ti l (pl ne ) coordin tes

• Amplitude of f t ny p ir of coordin tes (x, y) is c lled intensity

or grey level of the im ge t th t point.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
What is digital image?
When f re ll Finite, Discrete qu ntities, we c ll the im ge Digit l Im ge.
a
a
a
a
a
a
a
a
What is digital image processing?

The ield of digit l im ge processing refers to Processing Digit l Im ges

by me ns of Digit l Computers.
f
a
a
a
a
a
a
a
 Pixel?
• A digit l im ge consist of inite set of elements, nd it h s p rticul r
loc tion nd v lue. These elements re c lled picture elements, im ge
elements, pels, nd pixels.

• Pixel denote the elements of digit l im ge.

a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
 Applications of Digital Image
processing
• Hum ns re limited to visible b nd of electrom gnetic spectrum.

• But, im ging m chines covers entire EM spectrum which includes Ultr

sound, Electron Microscopy, X-r y, IR im ges etc…

• Thus, wide nd v ried ield of pplic tions re there in digit l im ge

processing.
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
 Boundary for various IP Task

Im ge processing Im ge An lysis Computer Vision

a
a
a
Three level process involved in Digit l im ge processing

a
a
Origins of Digital Image Processing
• First pplic tion - in newsp per industry

• Sent by subm rine c ble between London nd NewY rk.

• Took more th n week.

a
a
a
a
a
a
a
a
a
 Continued…

• Introduced B rtl ne C ble Picture Tr nsmission system in 1920’s.

• Took less th n 3 hours.

• These im ges were coded using 5 distinct levels of gr y.

a
a
a
a
a
a
a
 Continued…

• In 1929, incre sed to 15 levels of gr y , its im ge is s shown below

a
a
a
a
 Continued…

• These were not considered s Digit l Im ge Processing bec use

computers were not involved.

• There fore, History of Digit l Im ge Processing is Tied to Development

of Digit l Computers
a
a
a
a
a
a
a
 Continued…

• The irst computer powerful enough to c rry out me ningful im ge

processing t sks ppe red in e rly 1960’s.
f
a
a
a
a
a
a
a
 Continued…

• In 1964, the irst im ge of the moon t ken by US sp cecr ft nd

processed by computer to correct v rious types of im ge distortion.
f
a
a
a
a
a
a
a
 Continued…
From e rly 1970’s,

St rted using Im ge Processing Techniques in-

• Medic l

• Remote E rth Resource Observ tion

• Astronomy, Geogr phy, Biology

• Nucle r Medicine

• L w Enforcement

• Defence nd Industry.
a
a
a
a
a
a
a
a
a
a
 Ap rt from these,

• Autom tic Ch r cter Recognition

• Industri l M chine Vision for product ssembly nd inspection

• Milit ry

• Processing of Fingerprints

• Screening of X-R ys nd Blood s mples

• Processing of Aeri l nd S tellite im gery for we ther Applic tions

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
Examples of ields that use DIP

f
 Gamma-Ray Imaging

M jor uses include Nucle r Medicine nd Astronomic l Observ tion

a
a
a
a
a
 Im ges of this sort re
used to loc te sites of
Bone p thology, such s
Infections or Tumors.
a
a
a
a
a
 This im ge shows Tumor in
the br in nd one in the
lung, e sily visible s sm ll
white m sses.
a
a
a
a
a
a
a
a
 A st r in the constell tion of Cygnus exploded bout 15,000 ye rs go,
gener ting superhe ted st tion ry g s cloud (known s the Cygnus
Loop) th t glows in spect cul r rr y of colors.
•
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge of the Cygnus Loop
in the g mm -r y b nd.
a
a
a
a
a
 Im ge of g mm r di tion from
v lve in nucle r re ctor. An re of
strong r di tion is seen in the lower
left side of the im ge.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 X-Ray Imaging

• Us ge in Medic l di gnostics, Industry, stronomy

a
a
a
a
 Ex mple im ge of Chest X-r y

- im ge of bone
a
a
a
a
a
 Ex mple im ge to show ngiogr phy

- im ges of blood vessels

a
a
a
a
a
 Ex mple im ge of CAT

- Typic l he d CAT slice im ge

a
a
a
a
a
a
 Ex mple x-r y im ge of n electronic circuit bo rd.

-To recognise lows in m nuf cturing (missing

components or broken tr cks)
a
a
f
a
a
a
a
a
a
 Im ge of Cygnus Loop t ken in X-r y b nd
a
a
a
a
 Imaging in the Ultraviolet Band

Includes lithogr phy, industri l inspection, microscopy, l sers, biologic l im ging

nd stronomic l observ tions.

Here, we sh ll look into microscopy nd stronomy

a
a
a
a
a
a
a
a
a
a
a
a
 Fluorescence microscope im ge of norm l corn

a
a
 Corn infected by “smut”, dise se of
cere ls, corn, gr sses, onions, nd sorghum
c used by p r sitic fungi.
a
a
a
a
a
a
a
a
 Cygnus Loop im ged in the high-energy
region of the UV b nd.
a
a
Im ging in the Visible nd Infr red B nd

Here, we sh ll look into light microscopy, stronomy, remote sensing,

industry nd l w enforcement.
a
a
a
a
a
a
a
a
 T xol ( ntic ncer gent) - 240x

Cholesterol - 40x

Microprocessor - 60x

Nickel oxide thin ilm - 600x

Surf ce of udio CD - 1750x

Org nic superconductor - 450x

a
a
a
a
a
a
f
a
 T ble below shows them tic b nds in NASA’s LANDSAT s tellite. The prim ry
function of LANDSAT is to obt in nd tr nsmit im ges of the E rth from sp ce
for purposes of monitoring environment l conditions on the pl net.
a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge of W shington D C, which includes fe tures such s buildings, ro ds,
veget tion nd m jor river(the Potom c) going through the city
a
a
a
a
a
a
a
a
a
a
 One of the dev st ting storms c lled
Hurric ne in Western Hemisphere.

T ken by N tion l Oce nogr phic

nd Atmospheric Administr tion
(NOAA) s tellite using sensors in visible
nd infr red b nds.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Few more im ges in visu l spectrum is in utom ted visu l inspection of
m nuf ctured goods.

Controller bo rd for CD-ROM drive

- To inspect for missing p rts ( bl ck

squ re on the top right)
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge of pill cont iner

-To look for missing pills

a
a
a
a
 To look for bottles th t re not illed
up to n ccept ble level
a
a
a
a
a
f
 Cle r-pl stic p rt with n un ccept ble
number of ir pockets in it
a
a
a
a
a
a
a
 To detect the presence of nom lies
such s burned l kes.
a
f
a
a
a
Imaging in the Microwave Band

• Applic tion of im ging in this b nd is R d r.

• Inste d of c mer lens, r d r uses n ntenn nd digit l computer

processing to record its im ges.

• In r d r im ge, one c n see only the microw ve energy th t w s

re lected b ck tow rd the r d r ntenn .
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 R d r im ge covering rugged
mount inous re of southe st
Tibet. In the lower right corner,
you c n ind wide v lley of the
Lh s River, which is popul ted
by Tibet n f rmers.
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
Imaging in the Radio Band

• M jor pplic tion in this b nd - Medicine (MRI) nd stronomy.

a
a
a
a
a
a
 MRI im ge of Hum n knee
nd spine
a
a
a
Examples in which other Imaging
Modalities are used

• Acoustic im ging

• Electron microscope

• Synthetic im ging.
a
a
Fundamental Steps in DIP
Im ge Acquisition -

• It could be s simple s being given n im ge th t is lre dy in digit l form.

• Gener lly, this st ge involves preprocessing, such s sc ling

a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge enh ncement -

• process of m nipul ting n im ge so th t the result is more suit ble

th n the origin l for speci ic pplic tion.

• Enh ncement technique is problem oriented, i.e, method used for X-r y
im ges m y not be suit ble for s tellite im ges.

• Its completely subjective

a
a
a
a
a
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
 Im ge restor tion-

• De ls with improving the ppe r nce of n im ge.

• Unlike im ge enh ncement, its objective b sed- involves m them tic l

or prob bilistic models of im ges..
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Color Im ge processing -

• De ls with colour models nd b sic colour processing in digit l

dom in.
a
a
a
a
a
a
a
W vlet nd Multi-Resolution Tr nsform
Representing im ges in v rious degrees of resolution.
a
a
a
a
a
Compression -

• De ls with techniques for reducing stor ge required to s ve n im ge

(.jpg)
a
a
a
a
a
Morphologic l processing -

It focuses on the sh pe nd structure of objects within n im ge.

• Removes sm ll noise nd det ches connected objects.

• Helps ill g ps nd m ke objects more visible.
• Removes sm ll noise while keeping the m in sh pe int ct.
• Fills g ps inside objects.
• Detects speci ic sh pes or p tterns in n im ge.
a
f
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Segment tion-

• de ls with p rtition of im ge into constituent p rts or objects.

• Used for object identi ic tion.

• More ccur te the segment tion, more likely recognition is to succeed.

a
a
a
a
a
f
a
a
a
a
Represent tion nd description-

• Extr cts, either bound ry of region or ll the points in the region itself.

• Representing regions involves

1. Extern l Ch r cteristics (bound ry —>to extr ct sh pe )

2. Intern l Ch r cteristics (pixels —> to extr ct color/texture )

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Recognition -

• Assigns l bel to n object b sed on its description in the previous

st ge.
a
a
a
a
a
Components of an Image Processing
System
Problem dom in -

• Subject of interest (Depends on Applic tions)

a
a
 Im ge Sensors

H s two components

1. Sensor (converts light intensity to electric l output)

2. Digitiser (converts electric l output to digit l d t )

a
a
a
a
a
a
a
 Im ge processing h rdw re

• Cont ins digitiser + ALU th t perform Arithmetic nd Logic oper tions

in p r llel on n entire im ge

Eg - Aver ging im ges s quickly s they digitised(used for noise

reduction)

• Also c lled s front end subsystem

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Computer

• Its gener l-purpose computer from PC to Super computers.

a
a
 Im ge Processing Softw re

• Consist of speci lised modules th t perform speci ic t sks on im ges

a
a
a
a
f
a
a
 M ss Stor ge

For 1024*1024 pixel size im ge with intensity of e ch pixel 8 bit, requires

lmost 1Mb of stor ge without compression.

F lls into 3 princip l c tegories

1. Short-term stor ge - required during processing (computer memory)

2. On-line stor ge - required for f st rec ll (M gnetic disk or optic l medi )

3. Archiv l stor ge - required for infrequent ccess (m gnetic t pes)

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge displ ys

• Color TV monitors re used with ddition l gr phic c rds.

a
a
a
a
a
a
a
 H rdcopy

• Used for recording im ges which includes l ser printers, ilm c mer s,
inkjet units etc…
a
a
a
f
a
a
Network

• For quick communic tion of im ges vi remote sites through internet.

a
a
a
Elements of Visual Perception
•Cili ry body
•Cili ry muscle
•Cili ry ibers
•Corne
•Iris
•Anterior ch mber
•Lens
•Retin
•Scler
•Choroid
•Vitreous Humor
•Fove
a
a
a
a
a
a
a
f
a
Elements of Visual Perception
Structure of the Human Eye
 Cones nd Rods

Cones Rods

Between 6 nd 7 million 75 to 150 million

Loc ted prim rily in centr l portion of the

Distributed over the retin l surf ce
retin (fove )

Highly sensitive to color Sensitive to low level of illumin tion

Cone vision is c lled photopic / bright- Rods vision is c lled scotopic / dim-light
light vision vision
a
a
a
a
a
a
a
a
a
a
a
a
Distribution of rods nd cones in the retin
a
a
Im ge Form tion in the Eye

How to obt in dimensions of n im ge formed on the retin ?

a
a
a
a
a
a
 • A person is looking t tree 15m high t dist nce of 100m.

• The r nge of foc l length is pproxim tely 14 mm to 17 mm(eye is

rel xed).

• h denote height of th t object in the retin l im ge

• From the geometry of igure, 15/100 = h/17 or h=2.55mm.

a
a
a
a
a
a
f
a
a
a
a
a
a
a
Brightness Ad pt tion nd Discrimin tion

Brightness Ad pt tion

Few terminologies

• Scotopic vision -vision of the eye under low-light conditions.

• Photopic vision - vision of the eye under well-lit conditions

• Subjective brightness - intensity s perceived by hum n visu l system

a
a
a
a
a
a
a
a
a
 M x brightness under which
the object is just visible

Intensity perceived by
HVS is c lled subjective
brightness

Min brightness under which

the object is just visible
a
a
 • Even though the r nge of light intensity levels to which the hum n visu l
system c n d pt is enormous, it c nnot oper te over such r nge
simult neously. It ccomplishes this l rge v ri tion by ch nging its
over ll sensitivity, this phenomenon is known s “Brightness
Ad pt tion.”
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 • The long solid curve represents the r nge of intensities to which the
visu l system c n d pt. But, the visu l system c nnot oper te over
such r nge simult neously. R ther, it ccomplishes this l rge v ri tion
by ch nging its over ll sensitivity, phenomenon known s Brightness
Ad pt tion.

• For ny given set of condition, the current sensitivity level of the visu l
system is c lled brightness d pt tion level, corresponds to brightness
Ba s shown in igure nd object re not observ ble below Bb nd bove
the limit.
a
a
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Brightness Discrimin tion

Brightness discrimin tion refers to the eye’s bility to di erenti te

between di erent brightness levels within n d pted r nge.
ff
a
a
a
a
a
a
a
ff
a
 • A Cl ssic Experiment -

H ving subject look t l t, uniformly illumin ted re (di user-device th t

distribute light evenly) l rge enough to occupy the entire ield of view.

It is illumin ted from behind by light source whose intensity, I , c n be v ried.

To this ield, n increment l illumin tion ΔI is dded in the form of short-

dur tion l sh th t ppe rs s circle in the centre of the uniformly illumin ted
ield, s shown in igure.
f
a
a
a
a
f
a
f
a
a
a
f
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
f
a
ff
a
a
a
a
a
 • If ΔI is not bright enough, subject s ys “no”, indic ting no perceiv ble
ch nges

• If ΔI gets stronger, subject s ys “yes”, indic ting perceiv ble ch nge.

ΔIc
• The Qu ntity , where ΔIc is the increment of illumin tion
I
discrimin ble 50% of the time with b ckground illumin tion I, is c lled
Weber r tio.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 • A sm ll v lue of weber r tio me ns th t sm ll percent ge ch nge in
intensity is required. This represents “good” brightness discrimin tion.

• A L rge v lue of weber r tio me ns th t l rge percent ge ch nge

in intensity is required. This represents “poor” brightness
discrimin tion.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 ΔIc
A Plot of log s function of logI is s shown in next slide
I
a
a
a
Low level of
illumin tion (rod cells)

High level of
illumin tion (cones)
a
a
 Two Phenomen cle rly demonstr te th t Perceived brightness is not
simple function of intensity.

1- M ch b nds e ect

2- Simult neous contr st.

a
a
a
a
ff
a
a
a
a
a
 M ch B nds E ect

Visu l system tends to undershoot or

overshoot round bound ry
a
a
a
a
ff
a
 Simult neous Contr st

Perceived brightness does not depend simply on its intensity

a
a
Other ex mple of hum n perception phenomen re optic l illusions -
a
a
a
a
a
 Im ge sensing nd Acquisition

• Im ge cquisition using single sensor

• Im ge cquisition using sensor strip

• Im ge cquisition using sensor rr y

a
a
a
a
a
a
a
a
a
a
1. Im ge cquisition using
a
single sensor
a
a
• Sensing m teri l is Silicon

• Sensor type is photo diode

• Output w veform is proportion l to light

• Incoming energy is tr nsformed into volt ge.

• Filter is to improve selectivity

a
a
a
a
a
a
 2D im ging using single sensor
a
 • A ilm neg tive is mounted onto drum whose mech nic l rot tion
provides displ cement in one direction

• Single sensor mounted on le d screw th t provide motion in

perpendicul r direction
f
a
a
a
a
a
a
a
a
a
a
 2. Im ge Acquisition Using Sensor Strips (Line sensor)
a
• Inline rr ngement of sensors in the form of sensor strip

• Strip provides im ging elements in one direction

• Line r motion of the strip provides im ging in other direction

• Upto 4000 or more in-line sensors re possible in strip

• Sensor strip mounted in ring con igur tion re used in medic l nd

industri l im ging to obt in cross section l (slice) im ges of 3D objects
a
a
a
a
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
3. Im ge Acquisition Using Sensor Arr ys

a
 • Individu l sensor rr nged in 2D rr y

• A typic l sensor is CCD rr y

• P ck ged in rr ys of 4000 x 4000 elements or more

• Collect incoming energy nd focus onto im ge pl ne

• Digit l nd n log circuitry sweep these output nd convert them into

digitised from.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 A Simple Im ge Form tion Model

Im ges re two-dimension l functions of the form f(x, y) nd its v lue

must be nonzero nd inite, i.e

0 < f(x, y) < ∞

a
a
a
a
f
a
a
a
a
 The function f(x, y) is ch r cterised by two components:

1. Amount of source illumin tion incident on the scene - illumin tion

component, i(x, y)

2. Amount of illumin tion re lected by the objects in the scene - re lect nce
component, r(x, y)

i.e, f(x, y) = i(x, y)r(x, y)

Where, 0 < i(x, y) < ∞

nd, 0 < r(x, y) < 1

Tot l Tot l
bsorption re lect nce
a
a
f
a
a
a
a
a
a
a
f
a
f
a
 Im ge S mpling nd Qu ntis tion

To cre te digit l im ge, need to convert continuous sensed d t into

digit l form which involves two process:

S mpling nd Qu ntiz tion

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 • An im ge is continuous in both - coordin tes (x, y) nd in mplitude(intensity).

• To convert into digit l form, we need to s mple both, in coordin tes nd in

mplitude.

• Digitizing the coordin te v lues is c lled S mpling

• Digitizing mplitude v lue is c lled Qu ntiz tion.

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge gener ted by single sensing element
a
a
a
 Im ge gener ted by sensor rr y

Continuous im ge projected Result of im ge s mpling

onto sensor rr y nd qu ntis tion
a
a
a
a
a
a
a
a
a
a
a
a
a
 Representing Digit l Im ges

1. Gr phic l represent tion.

2. M them tic l represent tion.

a
a
a
a
a
a
a
a
a
 Gr phic lly

Three b sic w ys to represent f(x, y)

a
a
a
a
 Im ge plotted s surf ce

two xis determining sp ti l loc tion nd

third xis being v lues of f s function of
the two sp ti l v ri bles x nd y.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge displ yed s visu l intensity
rr y

Intensity of e ch point is
proportion l to the v lue
of f t th t point
a
a
a
a
a
a
a
a
a
a
a
a
 Im ge shown s 2-D numeric l
rr y (0,0.5, 1) representing
d rk grey nd white

Displ y the numeric l v lues of

f(x, y) s n rr y (m trix)
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 M them tic lly

f here is of size 600 x 600

elements, or 360,000 numbers

E ch element of this m trix

is c lled n element, picture
element, pel or pixel

Where, M = Number of Rows

aij = f(x = i, y = j) = f(i, j)
N = Number of Columns
a
a
a
a
a
a
a
No restrictions pl ced on M, N , other th n they h ve to be positive integers.

But, The number of intensity levels L needs to be n integer power of 2

k
i.e, L =2
This is due to stor ge nd qu ntising h rdw re consider tions
a
a
a
a
a
a
a
a
a
a
 Dyn mic r nge-

• R tio of the m ximum me sur ble intensity

to the minimum detect ble intensity level in
the system.

• As rule, the upper limit is determined by

s tur tion nd the lower limit by noise.

• It est blishes the lowest nd highest intensity

levels th t system c n represent.

• It is closely ssoci ted with im ge contr st.

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
high dyn mic r nge, me n to h ve high contr st.

low dyn mic r nge, typic lly h s dull, w shed-out grey look.
a
a
a
a
a
a
a
a
a
a
a
Number of Bits, b, required to store digitized im ge is

b=M×N×k

When M = N,
2
b=N k

a
a
 Sp ti l nd Intensity Resolution
a
a
a
 SPATIAL RESOLUTION > due to s mpling

• Sp ti l resolution is me sure of sm llest discernible det il in n im ge.

St ted in two me sures -

Line p irs per unit dist nce &

Dots(pixels) per unit dist nce

a
a
a
a
a
a
a
a
a
a
a
a
a
a
Line p ir per unit dist nce-

• Suppose we construct ch rt with ltern ting bl ck nd white vertic l

lines, e ch of width W units(W c n be less th n 1).

• The width of line p ir is thus 2W nd there re 1/2W line p irs per unit
dist nce.

• A widely used de inition of im ge resolution is the l rgest number of

discernible line p irs per unit dist nce (eg - 100 line p irs per mm)
a
a
a
a
f
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Dots per unit dist nce

• Me sure of im ge resolution used commonly in printing nd publishing

industry.

• Expressed s dots per inch (dpi).

a
a
a
a
a
 Illustr ting sp ti l resolution

1250dpi 300dpi 150dpi 72dpi

Checker bo rds e ect

a
a
a
a
ff
INTENSITY RESOLUTION > due to qu ntis tion

• is me sure of sm llest discernible ch nge in intensity level.

• Number of intensity level is n integer power of two (Eg - 8bits, 16bits, 32bits…)
a
a
a
a
a
a
a
 Intensity
Resolution

256 level, k=? 128 level 64 level 32 level

16 level-f lse contouring 8 level 4 level 2 level

a
 • F lse contouring e ect - use of insu icient number of intensity levels
in smooth re s of digit l im ge

• Checkerbo rd e ect - bec use of under s mpling of im ge(insu icient

number of pixels)
a
a
a
a
ff
ff
a
a
a
a
ff
a
a
ff
Rough rule of thumb -

Im ge size of 256 x 256 pixels with 64 intensity levels nd printed on size on

the order of 5 x 5 cm re bout the lowest sp ti l nd intensity resolution
im ges th t c n be expected to be re son bly free from checkerbo rds nd
f lse contouring.
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Image Interpolation

Interpol tion - is the process of using known d t to estim te v lues t

unknown loc tion.

• Ne rest Neighbour interpol tion

• Biline r interpol tion

• Bicubic interpol tion

a
a
a
a
a
a
a
a
a
a
a
a
Nearest Neighbour interpolation
Adv nt ge

• Simple to implement

Dis dv nt ge

• Tendency to produce undesir ble rtef cts - distortion of str ight edges
a
a
a
a
a
a
a
a
a
 Bilinear interpolation
• Use four ne rest neighbours to estim te the intensity t given loc tion.

• Let v(x, y) denote intensity v lue to be ssigned, its v lue is obt ined using
the equ tion

v(x, y) = ax + by + cxy + d

Where a, b, c & d re the four coe icients determined from four equ tions in
four unknown using four ne rest neighbours of point (x, y)

• Gives much better result th n ne rest neighbour interpol tion.

a
a
a
a
a
a
a
ff
a
a
a
a
a
a
a
a
a
 Bicubic interpolation
• Involves sixteen ne rest neighbours of point.

• Let v(x, y) denote intensity v lue to be ssigned, its v lue is obt ined
using the equ tion
3 3
i j
∑∑
v(x, y) = aij x y
i=0 j=0

• Where sixteen coe icients re determined from the sixteen equ tions
in sixteen unknowns th t c n be written using 16 ne rest neighbours
of point (x, y).
a
ff
a
a
a
a
a
a
a
a
a
a
a
 Some B sic Rel tionship between Pixels

Neighbours of Pixel

• A pixel p t coordin tes (x, y) h s two horizont l nd two vertic l neighbours

whose coordin tes re given by

(x + 1,y), (x − 1,y), (x, y + 1), (x, y − 1)

• This set of pixels, c lled 4-neighbors of p, is denoted by N4(p).

a
a
a
a
a
a
a
a
a
a
a
a
• The four di gon l neighbours of p h ve coordin tes

(x + 1,y + 1), (x + 1,y − 1), (x − 1,y + 1), (x − 1,y − 1)

• Denoted by ND(p).

• These points, together with 4-neighbours, re c lled the 8-neighbours of

p, denoted by N8(p).
a
a
a
a
a
a
 Adj cency, Connectivity, Regions, nd Bound ries

Adj cent?

Two pixels th t re neighbours nd h ve s me grey levels re c lled dj cent .

Types -

. 4- dj cency

b. 8- dj cency

c. m- dj cency (mixed dj cency)

a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 . 4- dj cency - N4(p)

• Two pixels p nd q with v lues from V re 4- dj cent if q is in the set

N4(p).

b. 8- dj cency - N8(p)

• Two pixels p nd q with v lues from V re 8- dj cent if q is in the set

N8(p).
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 c. m- dj cency (mixed dj cency) -

It is introduced to elimin te the mbiguities th t often rise when 8- dj cency

is used.

Note - Connected pixels - two pixels p nd q re connected if there exists

p th between p nd q
a
a
a
a
a
a
a
a
a
a
a
a
a
a
a
 Two pixels p nd q with v lues from V re m- dj cent if

i. q is in N4(p), or

ii. q is in ND(p) nd the set N4(p) ∩ N4(q) h s no pixels whose v lues

re from V.
a
a
a
a
a
a
a
a
a
 Connectivity
Let S represent subset of pixels in n im ge. Two pixels p nd q re
s id to be connected in S if there exists p th between them
consisting entirely of pixels in S.
a
a
a
a
a
a
a
a
Region
Let R be subset of pixels in n im ge, we c ll R region of the im ge if R is
connected set.
a
a
a
a
a
a
a
 Two regions, Ri nd Rj re s id to be dj cent if their union forms
connected set.

Regions th t re not dj cent re s id to be disjoint.

a
a
a
a
a
a
a
a
a
a
a
a
 Bound ry (border or contour)
The bound ry of region R is the set of points th t re dj cent to points is the
complement of R.
a
a
a
a
a
a
a
 Dist nce Me sures

For the pixels p, q nd z, with coordin tes (x, y), (s, t) nd (v, w), respectively,
D is dist nce function or metric if

. D(p, q) ≥0

b. D(p, q) = D(q, p), nd

c. D(p, z) ≤ D(p, q) + D(q, z) .

a
a
a
a
a
a
a
a
a
Three types of dist nce me sure

1. The Euclide n dist nce between p nd q is de ined s

2 2 1/2
De(p, q) = [(x − s) + (y − t) ]
a
a
a
a
a
f
a
2. The D4 dist nce (city-block dist nce) between p nd q is de ined s

D4(p, q) = | x − s | + | y − t |
a
a
a
f
a
3. The D8 dist nce (chessbo rd dist nce) between p nd q is de ined s

D8(p, q) = max( | x − s | , | y − t | )
a
a
a
a
f
a