8/19/2019 ch03-enhancementspdom
1/74
H.R. Pourreza
Digital ImageProcessing
H.R.Pourreza
Image Enhancement in theSpatial Domain
(Chapter 3)
8/19/2019 ch03-enhancementspdom
2/74
H.R. Pourreza
The principal objective of enhancement is toprocess an images so that the result is moresuitable than the original image for aSPECIFIC application
Category of image enhancement• Spatial domain• Frequency domain
Objective
8/19/2019 ch03-enhancementspdom
3/74
H.R. Pourreza
Background
g ( x , y) = T [ f ( x , y) ]
8/19/2019 ch03-enhancementspdom
4/74
H.R. Pourreza
Point Processing,Gray-Level rans!or"ation Function
s = T (r )
Contrast
stretching
Thresholding
8/19/2019 ch03-enhancementspdom
5/74
H.R. Pourreza
g ( x , y) = T [ f ( x , y) ] 1
0
0
-1
0
0
0
0
0
T
f ( x, y) g ( x, y)
#ask Processing Filtering
8/19/2019 ch03-enhancementspdom
6/74
8/19/2019 ch03-enhancementspdom
7/74
H.R. Pourreza
I"age $egative
s = L - 1 - r
8/19/2019 ch03-enhancementspdom
8/74
H.R. Pourreza
Log rans!or"ations
s = c log(1+r )
Pixel values dynamic range=[ ! "#$%"&'
8/19/2019 ch03-enhancementspdom
9/74
H.R. Pourreza
Po%er-La% rans!or"ations
s = cr
s = r 2.5
r = [1 10 20 30 40 210 220 230 240 250 255]
s(γ = 2.5) = [0 0 0 1 2 157 176 197 219 243 255]
s(γ =.4) = [28 70 92 108 122 236 240 245 249 253 255]
8/19/2019 ch03-enhancementspdom
10/74
H.R. Pourreza
Po%er-La% rans!or"ations
&uto"atic
Selection
8/19/2019 ch03-enhancementspdom
11/74
H.R. Pourreza
Po%er-La% rans!or"ations - Ga""aCorrection
8/19/2019 ch03-enhancementspdom
12/74
H.R. Pourreza
Po%er-La% rans!or"ations
c=1
γ = 0.4c=1
γ = 0.3
c=1γ = 0.6
8/19/2019 ch03-enhancementspdom
13/74
H.R. Pourreza
Po%er-La% rans!or"ations
c=1
γ = 3
c=1
γ = 5
c=1
γ = 4
8/19/2019 ch03-enhancementspdom
14/74
H.R. Pourreza
• A!antage• Ar"itraril# comple$
• Disa!antage• %ore user input
• pe o' &rans'ormations•
Contrast stretching• ra#le!el slicing
• *itplane slicing
Piecewise-Linear Transformation
Functions
8/19/2019 ch03-enhancementspdom
15/74
H.R. Pourreza
Contrast Stretc'ing
(bjective+ Increase the #namic range o' the gra#le!els in the image
Causes for poorimagePoor illumination,rong lens aperture,rong shutter spee-ac o' #namicrangein the imaging sensor
8/19/2019 ch03-enhancementspdom
16/74
H.R. Pourreza
)xample "For image *ith intensity range [$ ! "$'+hat should ,r"-s". and ,r/-s/. be to increase
the dynamic range of the image to [ !/$$'0
r
s
Contrast Stretc'ing
8/19/2019 ch03-enhancementspdom
17/74
H.R. Pourreza
(bjective1 2ighlighting a speci3c range of gray levels inan image#
Gray-Level Slicing
8/19/2019 ch03-enhancementspdom
18/74
H.R. Pourreza
255 138 30
65 12 201
180 111 85
255
138
30
65
12
201
183
111
85
1 1 1 1 1 1 1 1
0 1 0 1 0 0 0 1
1 1 1 1 1 0 0 0
1 0 0 0 0 0 1 0
0 0 1 1 0 0 0 0
1 0 0 1 0 0 1 1
1 1 1 0 1 1 0 1
1 1 1 1 0 1 1 0
1 0 1 0 1 0 1 0
MSBLSB
1 1 0
0 0 1
1 0 0
MSB plane
1 0 1
1 0 1
1 1 1
LSB plane
Bit-Plane Slicing
8/19/2019 ch03-enhancementspdom
19/74
H.R. Pourreza
Bit-Plane Slicing
8/19/2019 ch03-enhancementspdom
20/74
H.R. Pourreza
Bit-Plane Slicing
8/19/2019 ch03-enhancementspdom
21/74
H.R. Pourreza
2istogram &he histogram o' a igital image /ith gra# le!els inthe range 012 L4 is a iscrete 'unction
h(r k
) 5 nk
2
/here r k is the k th gra# le!el an nk is the num"er o'
pi$els in the image ha!ing gra# le!el r k .
4ormali5ed 2istogram
Di!iing each !alue o' the histogram "# the total numo' pi$els in the image2 enote "# n.
p(r k ) 5 nk 6n.
7ormalize histogram pro!ie use'ul image statistics.
(istogra" Processing
8/19/2019 ch03-enhancementspdom
22/74
H.R. Pourreza
!n"#$on %o&'al$e*$# = 'g*$#($'g)
[,/]=$e($'g)
$#=e&o(256,1)
o& & = 1o& "=1/
$#($'g(&,")+1,1)=$#($'g(&,")+1,1)+1 8
en
en
%o&'al$e*$# = $#(/) plo#(%o&'al$e*$#)
(istogra" E)traction using #atlab
8/19/2019 ch03-enhancementspdom
23/74
H.R. Pourreza
(istogra" Processing
8/19/2019 ch03-enhancementspdom
24/74
H.R. Pourreza
2istogram equali5ation is used to enhance imagecontrast and gray!level detail by spreading thehistogram of the original image# s = , r . r "-*here r and s are normali5ed pixel intensities
Conditions for the transformation*a+ , r . is single!valued and monotonically increasing
in the interval r "
,b. , r . " for r "
1
(istogra" Eualiation
8/19/2019 ch03-enhancementspdom
25/74
H.R. Pourreza
9":ecti!e o' histogram e;ualization &rans'orm the histogram 'unction o' theoriginal image pr (r ) to a uni'orm histogram
'unction. ps(s) 5 1≤ s ≤
!al$e&
&ano&'a#$on pr (r ) p s( s)
∫ ==r
r dww pr T s
0
)()( ∑=
==k
j
jr k k r pr T s0
)()(
Continuous case Discrete case
(istogra" Eualiation
8/19/2019 ch03-enhancementspdom
26/74
H.R. Pourreza
0
1
2
3
128
129
130
253
254
255
0
0
0
0.004
0.15
0.05
0.005
0.006
0.004
pr (r k )k
T (r k )
00
0
0
00.004
0.154
0.204
0.989
0.994
1
sk
0
0
0
0
10
39
52
252
253
255
0
0.004
0.008
0.011
0.5
0.505
0.51
0.992
0.996
1
r k
255 255
np!# $'age !#p!# $'age
8/19/2019 ch03-enhancementspdom
27/74
H.R. Pourreza
(istogra" Eualiation
8/19/2019 ch03-enhancementspdom
28/74
H.R. Pourreza
255
(istogra" Eualiation
(i E li il$
8/19/2019 ch03-enhancementspdom
29/74
H.R. Pourreza
(istogra" Eualiation
15
1
8
0.5
4
0.25
12
0.75
0.4
0.2
0.3
0.1
0
1
23
4
5
67
8
9
10
11
12
13
14
15
0
0
00
0.09
0.06
0.10.15
0.3
0.18
0.05
0.07
0
0
0
0
0
0
00
0.09
0.15
0.250.4
0.7
0.88
0.93
1
1
1
1
1
r k pr (r k ) %o&'al$e sk
0
0
00
1
2
46
11
13
14
15
15
15
15
15
sk
plo#( sk )
np!# $'age $e 100 × 100
8/19/2019 ch03-enhancementspdom
30/74
H.R. Pourreza
6s histogram equali5ation a good approachto enhance the belo* image0
(istogra" #atc'ing
8/19/2019 ch03-enhancementspdom
31/74
H.R. Pourreza
2istogram 7atching method generates aprocessed image that has a speci3edhistogram#
np!# $'age
pr (r )
T (r )
∫ r
r dt t p
0
)( s
e$&e!#p!# $'age
p z ( z )
G ( z )
∫ z
z dt t p
0
)( v
s ≈ v, $n"e :o#; $'a#el? !n$o&' pdf
z = G-1[T ( s )] s = G[T ( r )]
(istogra" #atc'ing
8/19/2019 ch03-enhancementspdom
32/74
H.R. Pourreza
(istogra" #atc'ing
8/19/2019 ch03-enhancementspdom
33/74
H.R. Pourreza
(istogra" #atc'ing
8/19/2019 ch03-enhancementspdom
34/74
H.R. Pourreza
0.6
0.12
0.08
0.0
0.01
0.005
0
0.6
0.72
0.8
0.8
0.9
0.95
1
153
184
204
204
230
242
255
(istogra" Eualiation
8/19/2019 ch03-enhancementspdom
35/74
H.R. Pourreza
) %oi'# the histogram o' the image to o"tain p z ( z ).
e histogram in
step .3) =in the in!erse G ( z )?) Enhance image is
o"taine "# appl#ing G to thepi$els o' the histograme;ualizeimage sho/n in =ig. 3.
8/19/2019 ch03-enhancementspdom
36/74
H.R. Pourreza
125128
0.98
0.002
250255
0.70.3
178
255
Lo"al $#og&a'
@lo:al $#og&a'
125 128
Local (istogra" En'ance"ent
8/19/2019 ch03-enhancementspdom
37/74
H.R. Pourreza
Local (istogra" En'ance"ent
8/19/2019 ch03-enhancementspdom
38/74
H.R. Pourreza
Contrast manipulation using local statistics2 such as the mean an!ariance2 is use'ul 'or images /here part o' the image is accepta"ut other parts ma# contain hien 'eatures o' interest.
(istogra" Statistics !or I"age En'anc.
8/19/2019 ch03-enhancementspdom
39/74
H.R. Pourreza
-et ( x 2 y ) "e the coorinates o' a pi$el in an image2 an let S xy enote a neigh"orhoo (su"image) o' speci>e size2 centere at( x 2 y ).
.)(][
)(
),(
,2
,2
),(
,,
∑
∑
∈
∈
−=
=
xy
xy xy
xy xy
S t s
t sS t sS
S t s
t st sS
r pmr
r pr m
σ
&he local mean an !ariance are the ecision 'actors to /hetherlocal enhancement or not.
(istogra" Statistics !or I"age En'anc.
8/19/2019 ch03-enhancementspdom
40/74
H.R. Pourreza
≤≤≤=
o#;e&e
parameters
E = 4,
k 0 = 0.4,
k 1= 0.02,
k = 0.4
S$e o lo"al a&ea = 3 × 3
(istogra" Statistics !or I"age En'anc.
8/19/2019 ch03-enhancementspdom
41/74
H.R. Pourreza
125 135 145 135 ..
168 175 158 149 ..
210 231 215 129 ..
187 192 145 200 ..
174
'age Mean
164
187
µ ( x, y)
σ( x, y)
E
(istogra" Statistics !or I"age En'anc.
8/19/2019 ch03-enhancementspdom
42/74
H.R. Pourreza
Enhance Image 9riginal Image
(istogra" Statistics !or I"age En'anc.
8/19/2019 ch03-enhancementspdom
43/74
H.R. Pourreza
Arithmetic6logic operations in!ol!ing images areper'orme on a pi$el"#pi$el "asis "et/een t/o ormore images.
8rithmetic (perations Aition2 Su"traction2 %ultiplication2 an Di!ision
9ogic (perations
A7D2 9R2 79&
En'ance"ent /sing &rit'"etic0LogicO1.
8/19/2019 ch03-enhancementspdom
44/74
H.R. Pourreza
A%
-ogic operations are per'orme on the "inar#
representation o' the pi$el intensities
En'ance"ent /sing 84: and (; LogicO1.
8/19/2019 ch03-enhancementspdom
45/74
H.R. Pourreza
En'ance"ent /sing &rit'"etic O1. 2S
8/19/2019 ch03-enhancementspdom
46/74
H.R. Pourreza
Pro"lem+ &he pi$el intensities in the i@erence image can range 'rom
8/19/2019 ch03-enhancementspdom
47/74
H.R. Pourreza
[ ]2
),(
2
),(
1
1),(),(
),(1
),(
),(),(),(
y x y x g
k
k
y x f y x g E
y x g k
y x g
y x y x f y x g
η σ σ
η
==
=
+=
∑=
9riginal image 7oise /ith zero mean
En'ance"ent /sing &rit'"etic O1.8veraging
8/19/2019 ch03-enhancementspdom
48/74
H.R. Pourreza
@a!$an %o$e
'ean = 0
a&$an"e = 64
! = 8
! = 32
! = 16
! = 128
f ( x, y) g ( x, y)
µ µ+σµ-σ x
En'ance"ent /sing &rit'"etic O1.8veraging
8/19/2019 ch03-enhancementspdom
49/74
H.R. Pourreza
Di@erenceimages"et/eenoriginal imagean imageso"taine 'roma!eraging.
! = 8
! = 16
! = 32
! = 128
7otice themean an
!arianceo' thei@erenceimagesecrease as
K increases.
En'ance"ent /sing &rit'"etic O1.8veraging
8/19/2019 ch03-enhancementspdom
50/74
H.R. Pourreza
Spatial >ltering are>ltering operationsper'orme on the pi$elintensities o' an imagean not on the're;uenc# componentso' the image.
∑ ∑−= −=
++="
" s
#
#t
t y s x f t sw y x g ),(),(),(
" = (m - 1) 2 # = ($ - 1) 2
Basics o! S1atial Filtering - 9inear
8/19/2019 ch03-enhancementspdom
51/74
H.R. Pourreza
Response2 R2 o' an m × n mas at an# point ( x 2 y )
∑=
=m$
z w %
1
Special consieration is gi!en /hen the center o' theapproach the "oarer o' the image.
Basics o! S1atial Filtering
8/19/2019 ch03-enhancementspdom
52/74
H.R. Pourreza
7onlinear spatial >lters operate on neigh"orhoos2 anthe mechanics o' sliing a mas past an image are thesame as /as :ust outline. In general ho/e!er2 the>ltering operation is "ase conitionall# on the !alues o'
the pi$el in the neigh"orhoo uner consieration2 anthe# o not e$plicitl# use coeBcients in the sumo'proucts manner escri"e pre!iousl#.
)xampleComputation 'or the meian is a nonlinear operation.
$onlinear o! S1atial Filtering
8/19/2019 ch03-enhancementspdom
53/74
H.R. Pourreza
Smoothing >lters are use 7oise reuction Smoothing o' 'alse contours Reuction o' irrele!ant etail
nesira"le sie e@ect o' smoothing >lters *lur eges
Ce$g;#e
ae&age
Bo>
$l#e&
,eighte a!erage >lterreuces "lurring in the
smoothing process.
S"oot'ing S1atial Filtering - Linear &veraging *lo%-1ass+ Filters
8/19/2019 ch03-enhancementspdom
54/74
H.R. Pourreza
$ = $l#e& $e
$ = 3
$ = 5 $ = 9
$ = 15
$ = 35
S"oot'ing S1atial Filtering 2 Linear &veraging *lo%-1ass+ Filters
8/19/2019 ch03-enhancementspdom
55/74
H.R. Pourreza
$l#e& $e
$ = 15
;&; = 25D o
;$g;e# $n#en$#?
S"oot'ing S1atial Filtering &veraging 4 'res'old
8/19/2019 ch03-enhancementspdom
56/74
H.R. Pourreza
9rerstatistics >lters are nonlinear spatial >lters/hoseresponse is "ase on orering (raning) the pi$elscontaine in the image area encompasse "# the
>lter2 an then replacing the !alue o' the centerpi$el /ith the !alue etermine "# the raningresult.3 × 3 Me$an $l#e& [10 125 125 135 141 141 144 230 240] = 141
3 × 3 Ma> $l#e& [10 125 125 135 141 141 144 230 240] = 2403 × 3 M$n $l#e& [10 125 125 135 141 141 144 230 240] = 10
%eian >lter eliminates isolate clusters o' pi$elsthat are light or ar /ith respect to their neigh"ors2an /hose area is less than n
8/19/2019 ch03-enhancementspdom
57/74
H.R. Pourreza
$ = 3
Ae&age
$l#e&
$ = 3
Me$an
$l#e&
Order Statistic Filters
8/19/2019 ch03-enhancementspdom
58/74
H.R. Pourreza
The principal objective of sharpening is tohighlight 3ne detail in an image or toenhance detail that has been blurred#
∫
∑9
1
9
1
9
1
91
z
z
'age Bl!&&e 'age
The derivatives of a digital function are de3ned interms of di>erences#
S'ar1ening S1atial Filters
8/19/2019 ch03-enhancementspdom
59/74
H.R. Pourreza
Re;uirements 'or igital eri!ati!e =irst eri!ati!e ) %ust "e zero in at segment
8/19/2019 ch03-enhancementspdom
60/74
H.R. Pourreza
S'ar1ening S1atial Filters
8/19/2019 ch03-enhancementspdom
61/74
H.R. Pourreza
Comparing the response bet*een 3rst! and second!orderedderivatives1". First!order derivative produce thic?er edge/. Second!order derivative have a stronger responseto 3ne detail- such as thin lines and isolated points#@. First!order derivatives generally have a strongerresponse to a gray!level step A/ B "$B. Second!order derivatives produce a doubleresponse at step changes in gray level#
In general the secon eri!ati!e is "etter than the >rsteri!ati!e 'or image enhancement. &he principle use o' >rsteri!ati!e is 'or ege e$traction.
S'ar1ening S1atial Filters
/se o! First 5erivative !or Edge
8/19/2019 ch03-enhancementspdom
62/74
H.R. Pourreza
First derivatives in image processing are implementedusing the magnitude of the gradient#
y x
t
GG y
f
x
f m"g f
y
f
x
f
+≈
∂∂
+
∂∂
=∇=∇
∂∂
∂∂
=∇
5.022
)( f
f z 1 z 2
z 6 z 8
z 4 z 7
z 3
z 9 z 5
;oberts operator
G> = ( z 9- z 5) an G? = ( z 8 - z 6)Sobel operator
G> = ( z 3+2 z 6 +9) - ( z 1+2 z 4+7) an
G?
= ( z 7
+2 z 8
+9
) - ( z 1
+2 z 2
+3
)
/se o! First 5erivative !or EdgeE)tractionGradient
8/19/2019 ch03-enhancementspdom
63/74
/se o! First 5erivative !or Edge
8/19/2019 ch03-enhancementspdom
64/74
H.R. Pourreza
f ( x, y) = [40, 140]
E. 0 0 0 0 0 0
E. 0 200 400 400 400 400 EE
E. 0 400 600 400 400 400 E...
E. 0 400 400 0 0 0
E. 0 400 400 0 0 0
E. 0 400 400 0 0 0
-1 -2
0
2
0
1
-1
1
0
-1 0
2
0
-2
-1
1
1
0
E 0 0 0 0 0
E 0 100 200 200 200
E 0 200 0 0 0
E 0 200 0 0 0
-11
-11
/se o! First 5erivative !or EdgeE)tractionGradient
/se o! First 5erivative !or Edge
8/19/2019 ch03-enhancementspdom
65/74
H.R. Pourreza
/se o! First 5erivative !or EdgeE)tractionGradient
8/19/2019 ch03-enhancementspdom
66/74
H.R. Pourreza
2
2
2
22
y
f
x
f f
∂
∂+
∂
∂=∇
)],(4)1,()1,(),1(),1([
),(2)1,()1,(
),(2),1(),1(
2
2
2
2
2
y y f y x f y x f y x f y x f f
y x f y x f y x f y
f
y x f y x f y x f x
f
−−+++−++=∇
−−++=∂∂
−−++=∂∂
6nd 5erivative 2 La1lacian
8/19/2019 ch03-enhancementspdom
67/74
H.R. Pourreza
2
2
2
22
y
f
x
f f
∂
∂+
∂
∂=∇
6sotropic 3lter response is independent of thedirection of the discontinuities in the image to*hich the 3lter is applied#
/se o! 6nd 5erivative !or En'ance"ent La1lacian
8/19/2019 ch03-enhancementspdom
68/74
8/19/2019 ch03-enhancementspdom
69/74
8/19/2019 ch03-enhancementspdom
70/74
H.R. Pourreza
g ( x, y) = f ( x, y) -
0 1
1
1
1
0
0
0
-4
),(2 y x f ∇
=
)],(4)1,()1,(),1(),1([2 y x f y x f y x f y x f y x f f −−+++−++=∇
/se o! 6nd 5erivative !or En'ance"ent La1lacian
8/19/2019 ch03-enhancementspdom
71/74
H.R. Pourreza
High"oost >ltering is use /hen the original image is"lurre an ar.
),(),( 2 y x f y x &f f '# ∇−= & 1
/n-s'ar1 #asking and (ig'-boostFiltering
8/19/2019 ch03-enhancementspdom
72/74
H.R. Pourreza
( :)
/n-s'ar1 #asking and (ig'-boostFiltering
Co"bining S1atial
8/19/2019 ch03-enhancementspdom
73/74
H.R. Pourreza
Co"bining S1atialEn'ance"ent#et'ods
Co"bining S1atial
8/19/2019 ch03-enhancementspdom
74/74
Co"bining S1atialEn'ance"ent#et'ods