330 Remote Sensing and GZS

Magnified image
3535 37 37 3838
Original image 38 38
35 35 37 37
3$ 37 3842
3232 34 32 32
323432 31
32 32 34 34 32 32
37 38 35 36
37 37 38 38 35 35

3737 38 38 35 3,5

Fig. 10.1| 2x image magnification

aninteger factor nsquared, cach pixel inthe original imageis usually replaced by an n
nblock of pixels all of which have the same DN values as the original input pixels. x
An
example ofthelogicofa 2x magnificationisshownin Fig. 10.11. This form of
doubles the size of each of the original pixelvalues. magnification
In many sophisticated digital image processing systems, the analyst can specify afloating
point magnification rate such as 2.75x. This requires that the original remote sensor d
be resampled in near real time using one of the image interpolation algorithms (i.e. nearest
neighbour., bilinear interpolation, or cubic convolution).
I0.10.3 Colour Compositing
A colour image can be generated by compositing three selected bands of multi-band
images with the use of three primary colours. Different colour images may be obtained
depending on the selection of three-band images and the assignment of the three primary
colours.
There are two methods of colour composite; an additive colour composite and a
subtractive colour composite. Additive colour composite uses three light sources of three
primary colours (RGB), for instance, colour graphics display. The subtractive colour
composite, uses three pigments of three primary colours (cyan, magenta, and yellow), for
instance, in colour printing.
Our display device has three colour guns (RGB), by which it throws colours to the
viewer. Multi-spectral images generally contain more than three spectral bands. But we
can view a maximum of three bands of any multi-spectral image at a time, as the
conmputer
display device throws colours using three colour guns (RGB)and the human eye cannot
detect any region of 'colour'(or electromagnetic spectrum) other than RGB. This implies
that several bands of a multi-spectral image are invisible to the human being (other
than
the visible range, 0.4-0.7 um). But we can view those commonly invisible bands
(like
NIR band) using the colour combination technique. True colour (also called
naturat
colour) combination is a combination when image captured in blue band is passed
throug
the blue colour gun of a display device, green band through the green colour gun, andred
band through the red colour gun (Fig. 10.12(a).
 Digital Image Processing 331

In some cases, especially where at least three bands in the visible region are not
available,some bands, which are out of visible region, are mathematically combined in
such a way that the appearance of theimage resembles a visible normal colour appearance.
Forrinstance, in casc of IRS LISS-III/LISS-IV or SPOT HRV, there is no blue band and
it is not possible to generate true colour composite normally for thOse images. But a
reasonably good natural colour composite can be produced bythe following combination
bands:
ofthe spectral
REDcolour gun = Red
GREEN colour gun =(3 x Green +NIRY4 =0.75 x Green +0.25 x NIR
BLUE colour gun = (3 x Green - NIR)4 =0.75 x Green - 0.25 x NIR
Colour combination generated in such manner often called false natural colour composite
(Fig. 10.12(b).
Note The concept of generating false natural colour composites from green, red, and NIR can also
be applied in the cases where we already have the original blue band. This artificially generated
natural colour composite is visually more realistic than the composite generated by using the
original blue band. This is due to higher atmospheric interference in the original blue band.
In practice, we use various colour combinations to facilitate the visual interpretation of an
image. Combinations other than true colour combination are known as false colour
combinations (Figs 10.12(c) and (d). With this type of combination, we see apleasing and
satisfying image. For instance, we may pass blue band of the image through blue display
gun, green band through green display gun, and infrared band throuh red display gun. Thus,
nomally invisible infrared band is visible by obtaining redcolour tones and the image may
highlight some features, which were not clearly identifiable in true colour composition. This
takes place because reflectance value of an object differs from band to band.
Particularly, the colour composite with the assignment ofblue colour gun to the green band,
green gun to the red band, and red gun to the NIR band is very popular, and is called an infrared
colour composite, (Fig. 10.12(c) which is the same as that found in colour infrared film.

(a) (b (c) (d)

Different colour composites: (a) natural, (b) false natural, (c) infrared, and (d) other
Fig, 10.12
false colour (RGB = 3,2,4) (see also Plate 8)
332 Amode SensiNg and GS
corresponding to R, G, and B
In the case of digital data, three values
colour combinations. For instance, if we
have a multi-spectral image of six makes various
threbae ndscan, thebe
be obtained in a combination of
number of colour composites that can
determined by the following formula:
6!(6-3)! = 120
multi-spectral
Colour composites are not only created from asingle change image
from different dates (multitemporal) are also used
donc by passing single bands of imagery
from
for
different dates
detection. It is
through the
set.generImagaelsy
guns of acolour monitor for displaying the image (Banner and ILynham 1981). To RGB colour
the changes betweenimages fromtwo different dates,
guns,
a single
while
band (fromnold identify
the same spectral date) can
t
be passed through both the blue and green colour band (
(for
(Bhatta 2010b).
the second or new date) is passed through the red colour gun
changethat is responsible for the change in reflectance in the
bright red or cyan, while areas with little or no-change will
selected bands
lappear in grey
.Land-cappearover
will
shades.
will appear in bright red if the pixel reflectance value increases due to change. In case of Changes
decrease in reflectance value, changes will appear in cyan colour instead of red. For
detecting the changes among imageriessoof three dates, it is required to pass
corresponding
bands through all three RGB colour guns. Resulting colours/shades are then cyan, yellow,
magenta, and grey; where, grey indicates little or no-change, and others indicate
in land-cover. echange
I0.10.4 Transect Extraction
Users of remotely sensed imagery frequently extract brightness values between points on
an image. A transect is a straight line between any two user-specified points within an
image. Transects are used in many digital image processing applications. The pixels that
lie on the transect can be used to measure DNS and
displayed to compare spectral or spatial differences.
Consider the example in Fig. 10.13.Here, we can
measure variations in pixel values that occurs between
points A and B. The graphs below the image contan
the associated values that stretch between points Aand
B for bands 1, 2. 3. and 4. The x-axis contains the
distance in pixels and the y-axis, the raw pixel values.
Transect at band Transect at band 2 As we can see, each graph contains a unique
distribution. Consider transect at band 3and notice tha
the spectral differences that occur along the transe!
Astudy
inthis band is much; greater than in other bands.optimal
Transect at band 3 Transect at band 4 such as this may be important in determining
bands for further analysis.
Fig. 10.13 Transects of multi-spectral
image
 Digilal Image Processing 333

10.10.5 Contrast Enhancement

problenin remote sensing is that the range of reflectance values collected by
A common
sensor may not match the capabilities of the colour
display monitor. In raw imagery, the
digital values (i.e.
ul data often populates only asmallportion of theavailable range of
a
the earth's
256 levels) that can be displayed by a computer screen. Materials on
8 bits or amounts of energy that are recorded by the sensor as
surfacereflect and emit different
on an image is referred to as contrast.
image data, The range of brightness values present
features stand out more clearly
Contrast enhancement is a process that makes the image
display or output device.
hy making optimal use of the colour intensities available on theenable conversion from an
Look-up table (LUT) is a pre-set function in a computer to
while changing the contrast,
innut signal to an output signal in real time. This implies that stores the changed
the original brightness values do not change; instead, the computer image.
brichtness values inthe LUT and utilizes these values to display the
Contrast manipulations involve changing the range of values in an image in order to
of brightness values between
inerease contrast. For instance, an image might have a range differences between features
-255. the
40and 90. Whenthis is stretched to a range of
similar amounts of energy
are enhanced. Unfortunately, different features often reflect
low contrast image. In
throughout the electromagnetic spectrum, resulting in a relatively
biophysical materials, there
addition, besides the obvious low contrast characteristics of
developingg countries often use natural
are cultural factors at work. For instance, people in
construction inthe rural areas. This results in a
building materials (e.g.,wood and soil) for opposed to urban areas where
remotely sensed imagery with a much lower contrast as
more prevalent. Thus, it is
concrete, asphalt, and fertilized green vegetation may be
components while enhancing an
important to consider both the biophysical and human
image for maximum contrast.
comprehend the concept of an
The key to understand contrast enhancements is to
histogram. A histogram is a graphical representation of the brightness values that
image In
comprise an image (refer Appendix B, Section B.8).
1000 other words, a histogram is a statistical graphic
Number
of
pixcls representation of the range of tones from dark to light and
associated number of pixels for each tone for an image.
For a single band of data, the horizontal axis ofa histogram
Histrogram
is therange of all possible brightness values (Fig. 10.14).
The vertical axis is the number of pixels associated with

300 each brightness value. Interpreting image histograms is

become
probably the single most important concept to
familiar with digital images while working with them. A
has been
histogram canconvey whether or not our image
X
0
255
100
oF flat, and
Brightness values properly exposed, whether the lighting is harsh describe
can also
Fig 10.14 what adjustments works best. A histogram
Histogram of an image
 and GIS
334 Remote NezLsing

(a)
(b)

Fig. 10.15 Narrow histogram refers lowcontrast (a) andI wide histogram referst
high contrst ()
a measure of thedifference in
the amount of contrast. Contrast is
light and dark arcas in a scene. Broad
histograms reflect a scene
with brightness between
contrast, whereas narrow histograms reflect less
(Fig.10.15).
contrast and
may appear flat or dull significant
10.10.5.I Linear Contrast Enhancement

Linear contrast enhancement linearly expands the origina Idigital values of the
sensed data into anew distribution. By expanding the original input values offthe remotely
the total range of colour resolution of the display device can be utilized. There are image,four
main methods of linear contrast enhancement.
Minimum-maximum stretch
" Saturation stretch
Average and standard deviation stretch
" Piecewise stretch
Minimum-Maximum Stretch While using the minimum-maximum linear contrast
stretch, the original minimum and maximum values of the data are assigned to anewly
specified set of values that utilize the full range of available brightness values of the
display unit. Consider an image with a minimum brightness value of 45 and a maximum
value of 205. When such an image is viewed without enhancements, the values of 044
and 206-255 are not displayed. Important spectral differences can be detected by stretching
the minimum value of 45 to 0and the maximum value of 205 to 255 (Fig. 10.16).

Max

o Min Max 255 0 Min 255

DNs DNs
(a) (b)
Fig. 10.1l6 Histogram before stretch (a) and after min-max stretch
(b)
 Frequency

Fig. minimum
determining deviation
Averageapplied to higher at range trimmed This Occupy are pixelsmethodSaturationstretch 'output'input' i., where image-processing Many digitalvalues value,
using minimum
10. minimum-maximum and 0 If An
5% is output and
I7 Min
are the (Fig. and
arealgorithm
to 255 than a uses or is is Output theimage
from and each reasonable ta il a the inmin following and scaled the
5%
mapped main
brightness
values
Input the and definedis and 10.17). Stretch is
value
saturation Standardthe ta il lost, specified trim) obtained value are process1ng old
minimum advantage th en =[(input maximum proportionately can
maximum mean
of
however, to is between of the maximum
the maximum either Generally,
amount relationship: be
the minimumsimilar The
in t
maximumhe
linear Deviation is remainder - values used
and oftenhistogram linear saturation real image in,min(inmax software systemshave
the (for
0 of
of to 0
contrast Max value). percent very th e form, and value that
maximumvalues. used stretch.
remainder DNs brightness and pixel and to
5% minimum-maximum 255, perform between matches
Stretch
distribution. part few it minimum optimize to
stretch Max 255 to
Frequency The push to It
The lesslinear maximum stretch stored is th e
values is
of pixels
values. rounded be in,min)) - built-in
information
part th an the the new the
following the not contrast (also converted a the
Min in linear new old
for ist
tails necessary ofdefinedhistogram residevalues referred a values, X capabilities
full maximum
similar histogram Sometimes to
look-up 255
such of at nearest mapping range minimum minimum
content
minimum stretch. two thatlinear respectively,
brightness
Output
values formula the
stretch is to table
histogram to that is enhanced ends lie
percent as
integer. ofthat value.
and Digital
Pixels these contrast
in
percentage brightness
available
between values. value
(ERDAS the more of automatically
can the
value) of a maximum All Image
same tails certain in the to
beyondstandard
stretch. Acomparedenhancedpixels outside morehistogram, a
be of stretch the input the
percentage or ol d
1999).
applied saturate that 255
prominently. the
percentage linear input newProcessing
the the values.
Many
intermediate
(for histogram except and expand
original defined but contrast image minimum
Max 255
DNs output,
for be they this 335
of the
 336

150 200 255 Remote

Input
Fig.10.19 a
piecewise
identification Piecewise
in selected Stretch
an
multiplied Eig
brightness : the analyst In Sensig
100 this 10
modes
region MaximumMinimum g
200
255 linear case,
by
value |
U3 may
any trim,
Different GISnd
Piecewise
linear of of
(DN) the contrast of standard
number a stretch number (d)
valuevalue
histogram. brightness When 20%
contrast
certain deviation
Pieces
Min stretch.
of (e.g., = = tail
the mean mean
contrast
stretch linear trim,
values.
distribution l enhancements:
200 100
Figure values
Input
(DN) A orhas
enhancement piecewise 3 (2+ (2 (e)
or
been (e)
255 200 100 Max 10.19 This of × × lx
4) SD standard
SD
the and
multiplied
method of of of (a)
200 150 0 Min Ouput
(DN') shows histogram
linear different
a image
image no
steps histogram deviation,
stretch,
255 200 Max the contrast of
150 results by DNDN
thatcontrast
logic 2.values)
|Decreased
NoIncreased expands for in values)
The () (b)
change enhancement increased an can 2x
min-max
used standard standard
enhancement image be
contrast
contrast in the achieved
in stretch, (D
contrast a
brightness is
piecewise enhancement bi-deviation deviation
in in
the the
involves or(Fig. (c)
is
tri-modal,
output output called 5%
linearranges 10.18), can tail
the be
a in
 number givencertainreduction tool.
radiometric
values output parts redistributed,
darkbrightness
of histogram non-linear
ofHistogram values. the have through When
of stretch10.10.5.2 givenstretch, nthehistogram
The To Histogram the Reference
" stretch " " " output original In
enhancement withcontin raast
DN. Thresholding normalization
DensityHistogram
slicing equalization
Histogram one the
total new perform together values histogram Major is the of yalues
rangeseveral a
of number When major the notimage normal single
stretch.
pixels theofvalues. of
contrast image, image use and
numbervalues, resolution Equalization
any appropriate. Non-linear of
over
equalization so non-linear values 0procedure, the
histogram for associated
image, there an disadvantage. of
histogram breakpoints, andlinear histogram. In
per based of (or
with Contrast shape so to an features the
bins, better wide a image' s
enhancement. the another
ofpixels
bin, 'peaks'). are algorithm. are contrast 255
in to objects piecewise
upon
as equalization, which a approximately a contrast Non-linear at image
manageability. 8-bitrange can is histogram
histogram isHistogram Contrast is defined. a they Because
shown in increased histogram. asymmetric that
constant
the the are also
greyscale (forautomatically
It in
EachHistogram stretch,
represent
in bin s The enhancement th e increase analysts stretch,
image simply instance,separate contrast
the that task originalvalue Enhancement piecewise
to the the at equalization Most level the
which
is The equalized,
the an has (unequal
mode). in
following numbered
divided pixel tails equal of conversion minimum
displaying
bin pixels most the thenhancement
e of
or of ofmust series a
the
reduces histogram techniques scene decrease
intensity. linear
they values function of same non-linear e real be
equation: by Bins into a
populated number input thdistribution), of
are sets normally all is lose very
andworld. contrast small
the of are distincta the frequency
pixel one image is th e
assigned. of an Quickbird
number equalization
of their the
involves maximum In familiar Digital
provided
isused contrast of are: the
pixels.image pixels conversion
can contrast min-max
distributed range values the correct methods contrast
stretch piecewise
togroups, the stretch
group along most have histogram wi th Image
of The are image in of to simple values
bins, to brightness
very each of is relative of t he stretches
is
pixels
reassigned serveranges if the useful several of th e
equalling havinghistogram. the to the linear aProcessing
there light whole
of transform linear are
modes very
as thimage
e conversion image
are a of are forms brightnessvalueshistogram
stretched powerful are
l1-bit values
oroutput
range contrast contrast
the then to data data few very are for of set 337
a a of in th e up
a
 338
Frequency
Remote
10.20Fig. one
equalization where
where
To this Ibyas
isThe
0
can Figure assign Sensing
5
to
histogramclosepixels
1 5 see k
H, B,int H, A 240 A=T=N= T/A=
N
2 10
10.20 B,i=nt pixels and
Histogram that
=aparticular isnumber =a =equalized
= = = ofthe the the
3 15 performed. number
bin =output integer the pixels/10 A
cach
the shows particular to input
as G2S
Input
DNs 4 number k=0
to possible.
10equalized total
number
60 i-1
(a) enhanced bin output bins,
before 5 of function number
two pixels number number bins
6 40 of
When brightness H,
+ bins, there value of
equalization 7 30 histograms. pixels number
= thereareIf bins
image per (truncating 2 H, 24 of
8 10
(which e would
thpixels are
this of
9 A=
24 bin withpixels following assigned pixels
of
histogram
gains value
Frequency per be240 pixels in
(a) The considered is the per
and contrast
assigned realvalue bin pixels to the
first numbers bin equation =A bins, perimage
after 20 is bin
0
compared
histogram equalization) after (input) i represented so
equalization 1 3 in
15 that
2 4 60 the
as to is
input integer) used the
3 most
OutputDNs to (ERDAS number
4 shows
the
(b) (b) populated
60 histogram. a
equalized
(ERDAS 6 the of
7 6 40 1900). pixels
values .To
1999) 89 30 areas in
histogram,
8 15 equalize each
A=24 of before
the bin
 Frequency
10.Fi2g1.
Brightness
Sometimes Although (lumping
uSng Tange.ImageSingle Density theuseful arspecification)
e the oneReference distribution extreme mostHistogram ofa orvery
necessary and Input )8.to
histogramNormala of there Imagebright original
two slightly dataimage histogram
values few
density The intovalue. input for values255 valHowever,
s0 I
together) Slicing mosaicking of are greyscales.
are analysts through ueshistogram.
contoured some the to
Stretch rarer,
information.
clouds characteristics
densityseries a
Density different resemble Normalization are
slicing. images is is the
details same the actualy at
by or in must
slices of Another the process the near Ifone data In.
into As densityslicing DNs should because or scene Reference the our This
of change tails be allvalues 2 this
sections a the are straightforward histogram image.theon would
objects to normal becausehistogram.the original for the bell-shaped the a data, is awareusually instance,
is of have or of formHowever, at technique
trying have l
result intervals
image th en of histogram densitya middle, Normal
different
the detection. adjacent determiningstretch be human the histogram
similar the having a or th at
associated the the at
of displayed
of process of reallocated
normalized Gaussian it image ends to
similar is sun involves This in by bring while output tails the
density another. (also in of distribution
distribution groups input
lost, orvalues anglescenes
To similar this observation. an
may
slices, characteristics. method
th e
of the equalization
of
grey the
either
of form achieve
known a conversion, image the peak histogram
out withvalues the
slicing, convertingwithin Histogram
orthat lookup histogram.
to the pixel be curve. pixels range
effect each of histogram.
brightness other information
level. atmospheric fittingvaluessometimes of would of (also the original
as enhancement good were table as contrast In th e (Fig. of of
corresponding
anuniform a histogram different Generally, th at ta il 0.
of specified pixels this may equalization This 3
th e results captured of create bell
10.21). known are pixels.histogram to7
image noise matching that and Digital
continuous values As the convertedsense,
enhancement curve. not about very results is
grey wieffects.
th converts brightnessa withobservedcalled is as be stretched
may can rangeresults matching result, an a In Image
also tones to histogram on is may same th eimagenormalValues distribution
Gaussian
appropriate.
a ) data dark may inare
be separate a distribution, normal the
agrey orfrom This useful the
appear same
brightness Gaussian to
histogram in orhide grouped Processing
to
segmented, bespecific
or interval values, histogram the distribution terrain very loss
reduced as tonescombining is histogram
for or is that that th e
especially
matching, histograrm objects based
normalized bright much
colours. matching
days, different is are shadows together.
th e
of range
digital ofinto in stretch natural
ordervalue upon of more 339
or by or of to th e in to
of dark
of
an a or a of t he 1
 Fig. 340
0 Frequency
10.22 Remole
10.Fi2g3. components
frequencydomain
are tone overneighbours within anfeatures Filtering
inconcept thFiltering
e I0.10.6 boundaries.Thresholding 10.22.Fidistinct
g. Thresholding
cach dividesBrightness values
Filtering an appearance Sensing
a
over smallimage. mapping
(a) or of in
encompasses OutputOutput into
two it
in several values Threshold and
frequency
spatial (a)of techniques
the area,Rough image an Functions
Thresholding
spatial pixels, have
image of DNDN (black This the on GIS
Examples texture, an spatial distinct a 255
textured image. based
image. value value
another used type histogram
domain can have high 0=
represented bespatial the on 1
0 produces distribution values
zero It Spatial = =for whiteand = of
implemented by low areasrefers manner their set input
inputthreshold image
spatial (b) convolution. frequencies, of
spatial of to spatialfiltersdigital DNs DNs binary values,features, the narrow
severalsimplemap a can assigned to when works This
are anthe in enhancement of computer
frequency. which its be a
given frequencies image, processing <
mapping > then assigned
another on
abestdensity
through frequencyfrequency. are DNDN separated
(bitonal) )
outputcorresponding surface
setgiven
inThe while greyscaledes1gned of of several some of
low(b) terms where threshold
threshold are the of monitor single-band
dimages.
DN
slice
i
coordinates the segments each
(Fig.smooth ofSpatialfunctions, as by grey-level
grey
ofFourier th e to follows: a distribution values.feature (also
spatial frequency the values feature. have rlevel, (or level
(c) variations highlight ini
10.23). areaschanges threshold
wi th called
transformof
frequency which th e (intensity) The has
frequency I change slices usually
printout).
the with sharply image different
of neWunique a level It
(called 2D or are S is
in in suppress combined
may black.
ne
little tone torelative is used DN (separable)
DNs single
espec1alyslice')
and , as
space thein relatedtheto defined as be All for and
(c) trequemy frequency variation are that to shown intoproduced, This other
display generally
value
in spatial DNs.yieldsD method
high which abrupt appear theirto specificenhance DNs usefu
1n in DN If on is