In the Name God

Digital Image Processing

Chapter 3:Image Enhancement

in the Spatial Domain
[email protected] Abbas Koochari
Principle Objective of Enhancement

 Process an image so that the result will be more suitable

than the original image for a specific application.
 The suitableness is up to each application.
 A method which is quite useful for enhancing an image may
not necessarily be the best approach for enhancing another
images

2
2 domains
 Spatial Domain : (image plane)
 Techniques are based on direct manipulation of pixels in an
image
 Frequency Domain :
 Techniques are based on modifying the Fourier transform of
an image
 There are some enhancement techniques based on
various combinations of methods from these two
categories.

3
Good images
 For human visual
 The visual evaluation of image quality is a highly subjective
process.
 It is hard to standardize the definition of a good image.
 For machine perception
 The evaluation task is easier.
 A good image is one which gives the best machine
recognition results.
 A certain amount of trial and error usually is required
before a particular image enhancement approach is
selected.

4
Spatial Domain

 Procedures that operate

directly on pixels.
g(x,y) = T[f(x,y)]
where
 f(x,y) is the input image
 g(x,y) is the processed image
 T is an operator on f defined
over some neighborhood of
(x,y)

5
Mask/Filter

 Neighborhood of a point (x,y) can

be defined by using a
square/rectangular (common
(x,y
used) or circular subimage area
)
centered at (x,y)
•  The center of the subimage is
moved from pixel to pixel starting
at the top of the corner

6
Point Processing
 Neighborhood = 1x1 pixel
 g depends on only the value of f at (x,y)
 T = gray level (or intensity or mapping)
transformation function
s = T(r)
 Where
 r = gray level of f(x,y)
 s = gray level of g(x,y)

7
Point Operations

8
 Point Processing

Nonlinear Lower
Original Darken Lower Contrast Contrast

Nonlinear Raise
Invert Lighten Raise Contrast Contrast

9
 Point Processing
Nonlinear Lower
Original Darken Lower Contrast Contrast

x x - 128 x / 2 ((x / 255.0) ^ 0.33) * 255.0

Nonlinear Raise
Invert Lighten Raise Contrast Contrast

255 - x x + 128 x * 2 ((x / 255.0) ^2) * 255.0

10
Contrast Stretching
 Produce higher contrast
than the original by
 darkening the levels below
m in the original image
 Brightening the levels
above m in the original
image

11
Thresholding
 Produce a two-level
(binary) image

12
Mask Processing or Filter
 Neighborhood is bigger than 1x1 pixel
 Use a function of the values of f in a predefined
neighborhood of (x,y) to determine the value of g at
(x,y)
 The value of the mask coefficients determine the
nature of the process
 Used in techniques
 Image Sharpening
 Image Smoothing

13
Neighborhood Operations

14
3 basic gray-level transformation
functions
 Linear function
Negative
nth root
 Negative and identity
transformations
Log
nth power  Logarithm function
 Log and inverse-log
transformation
 Power-law function
Identity Inverse Log
 nth power and nth root
transformations
Input gray level, r

15
Identity function
 Output intensities are
identical to input
Negative
intensities.
nth root
 Is included in the graph
Log
nth power
only for completeness.

Identity Inverse Log

Input gray level, r

16
Image Negatives

 An image with gray level in the

Negative range [0, L-1]
nth root where L = 2n ; n = 1, 2…
 Negative transformation :
Log
nth power s = L – 1 –r
 Reversing the intensity levels of
an image.
 Suitable for enhancing white or
Inverse Log
gray detail embedded in dark
Identity
regions of an image, especially
when the black area dominant in
Input gray level, r
size.
17
Example of Negative Image

Original mammogram Negative Image : gives a

showing a small lesion of better vision to analyze
18
a breast the image
Image Negatives

L
y=L-x
0 L x

19
Log Transformations
s = c log (1+r)
Negative
 c is a constant
nth root
and r  0
Log  Log curve maps a narrow
nth power
range of low gray-level
values in the input image
into a wider range of output
levels.
Identity Inverse Log  Used to expand the values
of dark pixels in an image
while compressing the
Input gray level, r higher-level values.

20
Log Transformations
 It compresses the dynamic range of images with large
variations in pixel values
 Example of image with dynamic range: Fourier
spectrum image
 It can have intensity range from 0 to 106 or higher.
 We can’t see the significant degree of detail as it will
be lost in the display.

21
Example of Logarithm Image

Fourier Spectrum with Result after apply the log

range = 0 to 1.5 x 106 transformation with c = 1,
22
range = 0 to 6.2
Range Compression

Y=clog10dd

y  c log10 (1  x)
x
0 L

c=100
23
Inverse Logarithm Transformations
 Do opposite to the Log Transformations
 Used to expand the values of high pixels in an image
while compressing the darker-level values.

24
Power-Law Transformations
s = cr
 c and  are positive
constants
 Power-law curves with
fractional values of  map a
narrow range of dark input
values into a wider range of
output values
 .
 c =  = 1  Identity function
Input gray level, r
Plots of s = cr for various values
of 
(c = 1 in all cases)
25
Gamma correction

 Cathode ray tube (CRT)

devices have an intensity-
Monitor
to-voltage response that is
a power function, with 
varying from 1.8 to 2.5
 The picture will become
 = 2.5
Gamma
correction
darker.
 Gamma correction is done
by preprocessing the image
Monitor
before inputting it to the
monitor with s = cr1/

26
 =1/2.5 = 0.4
Another example : MRI a b
c d
(a) a magnetic resonance image of an
upper thoracic human spine with a
fracture dislocation and spinal cord
impingement
 The picture is predominately dark
 An expansion of gray levels are desirable
 needs  < 1
(b) result after power-law
transformation with  = 0.6, c=1
(c) transformation with  = 0.4
(best result)
(d) transformation with  = 0.3
(under acceptable level)
27
Effect of decreasing gamma
 When the  is reduced too much, the image begins to
reduce contrast to the point where the image started to
have very slight “wash-out” look, especially in the
background

28
Another example a b
c d
(a) image has a washed-out
appearance, it needs a
compression of gray levels 
needs  > 1
(b) result after power-law
transformation with  = 3.0
(suitable)
(c) transformation with  = 4.0
(suitable)
(d) transformation with  = 5.0
(high contrast, the image has
areas that are too dark, some
detail is lost)

29
Piecewise-Linear Transformation Functions
 Advantage:
 The form of piecewise functions can be arbitrarily complex
 Disadvantage:
 Their specification requires considerably more user input

30
Contrast Stretching

 x 0 xa

y    ( x  a )  ya a xb
yb
  ( x  b)  y bxL ya
 b
x
0 a b L

31 a  50, b  150,  0.2,   2,   1, ya  30, yb  200

Contrast Stretching

32
Gray-Scale Modification

33
Gray-level Slicing
Brighten in the
range

Highlighting a specific
range of gray levels in
an image
Display a high value
of all gray levels
in the range of
interest and a low
value for all
other gray levels

(a) transformation
highlights range [A,B]
of gray level and
reduces all others to a
constant level

(b) transformation
highlights range [A,B]
34 but preserves all other
levels
Contrast Stretching

 increase the dynamic range of the gray

levels in the image
 (b) a low-contrast image : result from
poor illumination, lack of dynamic
range in the imaging sensor, or even
wrong setting of a lens aperture of
image acquisition
 (c) result of contrast stretching: (r1,s1)
= (rmin,0) and (r2,s2) = (rmax,L-1)
 (d) result of thresholding

35
 Bit-plane slicing

 Highlighting the contribution

made to total image appearance
by specific bits
One 8-bit byte Bit-plane 7
(most significant)  Suppose each pixel is
represented by 8 bits
 Higher-order bits contain the
majority of the visually
significant data
 Useful for analyzing the relative
Bit-plane 0 importance played by each bit
(least significant)
of the image

36
Example

 The (binary) image for bit-

plane 7 can be obtained by
processing the input image
with a thresholding gray-
level transformation.
 Map all levels between 0 and
127 to 0
 Map all levels between 129 and
255 to 255

An 8-bit fractal image

37
8 bit planes

Bit-plane 7 Bit-plane 6

Bit- Bit- Bit-

plane 5 plane 4 plane 3

Bit- Bit- Bit-

plane 2 plane 1 plane 0

38
Bit-plane Slicing

39
Histogram Processing
 Histogram of a digital image with gray levels in the
range [0,L-1] is a discrete function
h(rk) = nk
 Where
 rk : the kth gray level
 nk : the number of pixels in the image having gray level rk
 h(rk) : histogram of a digital image with gray levels rk

40
Normalized Histogram
 dividing each of histogram at gray level rk by the total
number of pixels in the image, n
p(rk) = nk / n
 For k = 0,1,…,L-1
 p(rk) gives an estimate of the probability of
occurrence of gray level rk
 The sum of all components of a normalized
histogram is equal to 1

41
Histogram Processing
 Basic for numerous spatial domain processing techniques
 Used effectively for image enhancement
 Information inherent in histograms also is useful in image
compression and segmentation

42
 h(rk) or p(rk)
Example
rk

Dark image
Components of
histogram are
concentrated on the
low side of the gray
scale.
Bright image
Components of
histogram are
concentrated on the
high side of the gray
43
scale.
Example

Low-contrast image
histogram is narrow
and centered toward
the middle of the
gray scale
High-contrast image
histogram covers broad
range of the gray scale
and the distribution of
pixels is not too far from
uniform, with very few
vertical lines being much
44 higher than the others
 Histogram Modification

 Histogram Stretching
 Histogram Shrink
 Histogram Sliding
Histogram Stretching

46
Histogram Stretching

47
Histogram Stretching

48
Histogram Shrinking

49
Histogram Shrinking

50
Histogram Sliding

51
Histogram Equalization
 As the low-contrast image’s histogram is narrow and
centered toward the middle of the gray scale, if we
distribute the histogram to a wider range the quality
of the image will be improved.
 We can do it by adjusting the probability density
function of the original histogram of the image so
that the probability spread equally

52
 Histogram transformation

s = T(r)
s  Where 0  r  1
 T(r) satisfies
 (a). T(r) is single-valued
and monotonically
sk= T(rk) increasingly in the
interval 0  r  1
T(r)  (b). 0  T(r)  1 for
0r1

0 rk 1 r
53
 2 Conditions of T(r)

 Single-valued (one-to-one relationship) guarantees that

the inverse transformation will exist
 Monotonicity condition preserves the increasing order
from black to white in the output image thus it won’t
cause a negative image
 0  T(r)  1 for 0  r  1 guarantees that the output gray
levels will be in the same range as the input levels.
 The inverse transformation from s back to r is
r = T -1(s) ; 0  s  1

54
Probability Density Function
 The gray levels in an image may be viewed as random
variables in the interval [0,1]
 PDF is one of the fundamental descriptors of a random
variable

55
Random Variables
 instead of talking about the probability of a specific
value, we talk about the probability that the value of
the random variable lies in a specified range.
 In particular, we are interested in the probability that
the random variable is less than or equal to (or,
similarly, greater than or equal to) a specified
constant a.
 We write this as
F(a) = P(x  a)

56
Random Variables
 If this function is given for all values of a (i.e.,   < a
< ), then the values of random variable x have been
defined.
 Function F is called the cumulative probability
distribution function or simply the cumulative
distribution function (cdf).
 The shortened term distribution function also is
used.

57
Random Variables

 Observe that the notation we have used makes no

distinction between a random variable and the values it
assumes.
 If confusion is likely to arise, we can use more formal
notation in which we let capital letters denote the
random variable and lowercase letters denote its
values.
 For example, the cdf using this notation is written as
FX(x) = P(X  x)
58
Random Variables
 When confusion is not likely, the cdf often is written
simply as F(x).
 This notation will be used in the following discussion
when speaking generally about the cdf of a random
variable.

59
Random Variables
 Due to the fact that it is a probability,
the cdf has the following properties:
1. F(-) = 0
2. F() = 1
3. 0  F(x)  1
4. F(x1)  F(x2) if x1 < x2
5. P(x1 < x  x2) = F(x2) – F(x1)
6. F(x+) = F(x),
where x+ = x + , with  being a positive,
infinitesimally small number.

60
Random Variables

The probability density function

(pdf or shortly called density function)
of random variable x is defined as the
derivative of the cdf:

dF ( x )
p( x ) 
dx
61
Random Variables
The pdf satisfies the following properties:

62
Random Variables
 The preceding concepts are applicable to discrete
random variables.
 In this case, there is a finite no. of events and we talk
about probabilities, rather than probability density
functions.
 Integrals are replaced by summations and, sometimes,
the random variables are subscripted.
 For example, in the case of a discrete variable with N
possible values we would denote the probabilities by
P(xi), i=1, 2,…, N.

63
 Random Variables
 If a random variable x is transformed by a monotonic
transformation function T(x) to produce a new random
variable y,
 the probability density function of y can be obtained
from knowledge of T(x) and the probability density
function of x, as follows:

dx
p y ( y )  px ( x )
dy
where the vertical bars signify the absolute value.
64
Random Variables

 A function T(x) is monotonically increasing if T(x1) <

T(x2) for x1 < x2, and
 A function T(x) is monotonically decreasing if T(x1) >
T(x2) for x1 < x2.
 The preceding equation is valid if T(x) is an increasing or
decreasing monotonic function.

65
 Applied to Image
 Let
 pr(r) denote the PDF of random variable r
 ps (s) denote the PDF of random variable s
 If pr(r) and T(r) are known and T-1(s) satisfies condition (a)
then ps(s) can be obtained using a formula :

dr
p s(s)  p r (r)
ds

66
Applied to Image

The PDF of the transformed variable s

is determined by
the gray-level PDF of the input image
and by
the chosen transformation function

67
Transformation function

 A transformation function is a cumulative distribution

function (CDF) of random variable r :

r
s  T (r )  ( L  1)  pr ( w)dw
0

where w is a dummy variable of integration

Note: T(r) depends on pr(r)

68
Cumulative
Distribution function
 CDF is an integral of a probability function (always
positive) is the area under the function
 Thus, CDF is always single valued and monotonically
increasing
 Thus, CDF satisfies the condition (a)
 We can use CDF as a transformation function

69
Finding ps(s) from given T(r)

Probability density functions (PDF):

dr
p s ( s )  pr ( r )
ds
r
s  T (r )  ( L  1)  pr ( w)dw
0

d  r
pr ( w)dw  ( L  1) pr (r )
ds dT (r )
dr

dr
 ( L  1) 
dr  0 

1
ps ( s ) 
L 1

70
ps(s)
 As ps(s) is a probability function, it must be zero outside
the interval [0,1] in this case because its integral over all
values of s must equal 1.
 Called ps(s) as a uniform probability density
function
 ps(s) is always a uniform, independent of the form of pr(r)

71
Histogram Equalization

72
Discrete
transformation function

 The probability of occurrence of gray level in an image is

approximated by
nk
pr (rk )  where k  0, 1, ..., L-1
n
 The discrete version of transformation
k
sk  T ( rk )   pr ( r j )
j 0
k nj
 where k  0 , 1, ..., L- 1
j 0 n
73
Histogram Equalization
 Thus, an output image is obtained by mapping each
pixel with level rk in the input image into a
corresponding pixel with level sk in the output image
 In discrete space, it cannot be proved in general that
this discrete transformation will produce the discrete
equivalent of a uniform probability density function,
which would be a uniform histogram

74
Example

before after Histogram

equalization

75
Example

before after Histogram

equalization

The quality is
not improved
much because
the original
image already
has a broaden
gray-level scale
76
Histogram Equalization

77
Histogram Equalization

78
Histogram Equalization

Histogram

Cumulative
Normalized Histogram

79
Histogram Equalization

transformed intensity
gk = (L-1) * T(k).
To encompass the
whole dynamic range.

80
Example 2

No. of pixels
6
2 3 3 2 5

4 2 4 3 4

3 2 3 5 3

2
2 4 2 4
1
Gray level
4x4 image
0 1 2 3 4 5 6 7 8 9
Gray scale = [0,9]
histogram
81
Gray
0 1 2 3 4 5 6 7 8 9
Level(j)
No. of
0 0 6 5 4 1 0 0 0 0
pixels
k

n
j 0
j 0 0 6 11 15 16 16 16 16 16

k nj 6 11 15 16 16 16 16 16
s 0 0 / / / / / / / /
j 0 n
16 16 16 16 16 16 16 16
3.3 6.1 8.4
s*9 0 0 9 9 9 9 9
3 6 8
82
Example

No. of pixels
6
3 6 6 3 5

8 3 8 6 4

6 3 6 9 3

2
3 8 3 8
1

Output image
0 1 2 3 4 5 6 7 8 9
Gray scale = [0,9] Gray level
83
Histogram equalization
Note

 It is clearly seen that

 Histogram equalization distributes the gray level to reach the
maximum gray level (white) because the cumulative
distribution function equals 1 when
0  r  L-1
 If the cumulative numbers of gray levels are slightly different,
they will be mapped to little different or same gray levels as
we may have to approximate the processed gray level of the
output image to integer number
 Thus the discrete transformation function can’t guarantee the
one to one mapping relationship

84
 Histogram Matching (Specification)

 Histogram equalization has a disadvantage which is that it

can generate only one type of output image.
 With Histogram Specification, we can specify the shape of
the histogram that we wish the output image to have.
 It doesn’t have to be a uniform histogram

85
Histogram Matching (Specification)

Given a desired histogram, the goal is to transform the image intensity such that the
transformed image has a histogram matches the desired histogram.

Let the initial image histogram be pr, the desired image histogram be pz.

Let T be the function that equalizes the original image and G be the function that
equalizes the desired image.
– r: The initial image intensity.
– s: The image intensity after equalization
– z: The image intensity of the desired image.
– v: The image intensity after equalization of the desired image.

86
Histogram Matching (Specification)

Histogram equalization
r
s  T ( r )  ( L  1)  pr ( w) dw
0

z
G ( z )  ( L  1)  p z (t ) dt  s
0 Histogram equalization

z  G 1 ( s )  G 1[T (r )] pz (z ) is the desired PDF

Assume G-1 exists and satisfies the condition (a) and (b)
We can map an input gray level r to output gray level z

87
Histogram Matching (Specification)
k k nj
sk  T (rk )  ( L  1) pr (rj )  ( L  1) , k  0,1,2,..., L  1
j 0 j 0 n

k
vk  G ( z k )  ( L  1) p z ( zi ) sk , k  0,1,2,..., L  1
i 0

zk  G 1[T (rk )], k  0,1,2,..., L 1

88
Procedure Conclusion (continous)
1. Obtain the transformation function T(r) by
calculating the histogram equalization of the input
image
r
s  T (r )  ( L  1)  pr ( w)dw
0
2. Obtain the transformation function G(z) by
calculating histogram equalization of the desired
density function

z
G ( z )  ( L  1)  p z (t )dt  s
0
89
Procedure Conclusion
3. Obtain the inversed transformation function G-1

z = G-1(s) = G-1[T(r)]
4. Obtain the output image by applying the processed
gray-level from the inversed transformation function to
all the pixels in the input image

90
Histogram Matching (Specification)

91
Example

92
Example

93
94
95
96
97
Example

Assume an image has a gray level probability density

function pr(r) as shown.

Pr(r)   2r  2 ;0  r  1
pr ( r )  
2  0 ; elsewhere

1 r

 p ( w )dw  1
0
r

98 0 1 2 r
 Example

We would like to apply the histogram specification with

the desired probability density function pz(z) as shown.

Pz(z)
 2z ;0  z  1
2 pz ( z )  
 0 ; elsewhere
1 z

 p ( w )dw  1
z
z 0
99 0 1 2
Step 1:

Obtain the transformation function T(r)

r
s=T(r)
s  T ( r )   pr ( w )dw
0
1 r
  ( 2w  2 )dw
One to one 0
mapping r
function   w  2w
2
0
r
0 1   r 2  2r
100
Step 2:

Obtain the transformation function G(z)

z
G ( z )   ( 2w )dw
z
z 2
z 2
0
0

101
Step 3:

Obtain the inversed transformation function G-1

G( z )  T ( r )
z   r  2r
2 2

z  2r  r 2

We can guarantee that 0  z 1 when 0  r 1

102
Discrete formulation

k
sk  T ( rk )   pr ( r j )
j 0
k nj
 k  0 ,1,2 ,..., L  1
j 0 n
k
G ( z k )   pz ( z i )  sk k  0 ,1,2 ,..., L  1
i 0

z k  G 1 T ( rk )
103 G 1
sk  k  0 ,1,2 ,..., L  1
Example

Image is dominated by large,

dark areas, resulting in a
histogram characterized by a
large concentration of pixels in
pixels in the dark end of the gray
scale

Image
104 of Mars moon
 Image Equalization

Result image
after histogram
equalization
Transformation function
Histogram of the result image
for histogram equalization
The histogram equalization doesn’t make the result image look better than
the original image. Consider the histogram of the result image, the net
effect of this method is to map a very narrow interval of dark pixels into
the upper end of the gray scale of the output image. As a consequence, the
output
105image is light and has a washed-out appearance.
Solve the problem
Histogram Equalization

Since the problem with the

transformation function of the
histogram equalization was caused
by a large concentration of pixels in
the original image with levels near 0
Histogram Specification

a reasonable approach is to modify

the histogram of that image so that
it does not have this property

106
Histogram Specification

 (1) the transformation

function G(z) obtained
from
k
G ( z k )   pz ( z i )  sk
i 0

k  0 ,1,2 ,..., L  1
 (2) the inverse
transformation G-1(s)

107
 Result image and its histogram

Original image After applied the The output image’s

histogram equalization histogram

Notice that the output histogram’s

low end has shifted right toward
the lighter region of the gray scale
as desired.

108
Note
 Histogram specification is a trial-and-error process
 There are no rules for specifying histograms, and one
must resort to analysis on a case-by-case basis for any
given enhancement task.

109
Note
 Histogram processing methods are global processing, in
the sense that pixels are modified by a transformation
function based on the gray-level content of an entire
image.
 Sometimes, we may need to enhance details over small
areas in an image, which is called a local enhancement.

110
 a) Original image
(slightly blurred to
reduce noise)
Local Enhancement b) global histogram
equalization (enhance
noise & slightly
increase contrast but
the construction is
not changed)
c) local histogram
equalization using
7x7 neighborhood
(reveals the small
squares inside larger
ones of the original
image.
(a) (b) (c)
 define a square or rectangular neighborhood and move the center of this
area from pixel to pixel.
 at each location, the histogram of the points in the neighborhood is
computed and either histogram equalization or histogram specification
transformation function is obtained.
 another approach used to reduce computation is to utilize nonoverlapping
regions, but it usually produces an undesirable checkerboard effect.
111
 Histogram using a local 3*3 neighborhood

112
Explain the result in c)

 Basically, the original image consists of many small

squares inside the larger dark ones.
 However, the small squares were too close in gray level
to the larger ones, and their sizes were too small to
influence global histogram equalization significantly.
 So, when we use the local enhancement technique, it
reveals the small areas.
 Note also the finer noise texture is resulted by the
local processing using relatively small neighborhoods.

113
Using Histogram Statistics

Use of histogram statistics for image enhancement:

r denotes a discrete random variable
P(ri) denotes the normalized histogram component
corresponding to the ith value of r
L 1
Mean: m   ri p ( ri )
i 0

L 1

The nth moment:

 n (r )   (ri  m) n p (ri )
i 0

L 1

The second moment:  2 (r )   (ri  m) 2 p (ri )

i 0
 3.4 Fundamentals of Spatial
Filtering

115
Spatial Filtering
 Use of spatial masks for image processing (spatial filters)

 Linear and nonlinear filters

 Low-pass filters eliminate or attenuate high frequency

components in the frequency domain (sharp image
details), and result in image blurring.

116
Spatial Filtering
 High-pass filters attenuate or eliminate low-frequency
components (resulting in sharpening edges and other
sharp details).

 Band-pass filters remove selected frequency regions

between low and high frequencies (for image restoration,
not enhancement).

117
Spatial Filtering
a b
g(x, y)    w(s,t) f (x  s, y  t)
sa tb

a=(m-1)/2 and b=(n-1)/2,

m x n (odd numbers)

 For x=0,1,…,M-1 and y=0,1,…,N-1

 Also called convolution (primarily in the frequency

domain)

118
Spatial Filtering

 The basic approach is to sum products between the mask

coefficients and the intensities of the pixels under the
mask at a specific location in the image:

R  w1 z1  w2 z2  ...  w9 z9

(for a 3 x 3 filter)

119
 Image Enhancement in the
Spatial Domain

120
121
Spatial Correlation and Convolution
Spatial Filtering
 Non-linear filters also use pixel neighborhoods but do
not explicitly use coefficients

 e.g. noise reduction by median gray-level value computation in

the neighborhood of the filter

123
Smoothing Spatial Filters
 used for blurring and for noise reduction
 blurring is used in preprocessing steps, such as
 removal of small details from an image prior to object extraction
 bridging of small gaps in lines or curves
 noise reduction can be accomplished by blurring with a
linear filter and also by a nonlinear filter

124
 Smoothing Linear Filters
 output is simply the average of the pixels contained in the
neighborhood of the filter mask.
 called averaging filters or lowpass filters.

125
Smoothing Linear Filters
 replacing the value of every pixel in an image by the
average of the gray levels in the neighborhood will
reduce the “sharp” transitions in gray levels.
 sharp transitions
 random noise in the image
 edges of objects in the image
 thus, smoothing can reduce noises (desirable) and
blur edges (undesirable)

126
3x3 Smoothing Linear Filters

box filter weighted average

the center is the most important and other
pixels are inversely weighted as a function of
127 their distance from the center of the mask
Weighted average filter
 the basic strategy behind weighting the center point the
highest and then reducing the value of the coefficients as
a function of increasing distance from the origin is simply
an attempt to reduce blurring in the
smoothing process.

128
General form : smoothing mask
 filter of size mxn (m and n odd)

a b

  w(s, t ) f ( x  s, y  t )
g ( x, y )  s   at   b
a b

  w(s, t )
s   at   b

summation of all coefficient of the mask

129
 a b
c d
Example
e f

 a). original image 500x500 pixel

 b). - f). results of smoothing with
square averaging filter masks of size
n = 3, 5, 9, 15 and 35, respectively.
 Note:
 big mask is used to eliminate small objects
from an image.
 the size of the mask establishes the
relative size of the objects that will be
blended with the background.

130
Example

original image result after smoothing result of thresholding

with 15x15 averaging mask

we can see that the result after smoothing and thresholding,

the remains are the largest and brightest objects in the image.
131
Order-Statistics Filters (Nonlinear Filters)
 the response is based on ordering (ranking) the pixels
contained in the image area encompassed by the filter
 example
 median filter : R = median{zk |k = 1,2,…,n x n}
 max filter : R = max{zk |k = 1,2,…,n x n}
 min filter : R = min{zk |k = 1,2,…,n x n}
 note: n x nis the size of the mask

132
Median Filters
 replaces the value of a pixel by the median of the gray
levels in the neighborhood of that pixel (the original
value of the pixel is included in the computation of
the median)
 quite popular because for certain types of random
noise (impulse noise  salt and pepper noise) , they
provide excellent noise-reduction capabilities, with
considering less blurring than linear smoothing filters
of similar size.

133
Median Filters

 forces the points with distinct gray levels to be more

like their neighbors.
 isolated clusters of pixels that are light or dark with
respect to their neighbors, and whose area is less than
n2/2 (one-half the filter area), are eliminated by an n x n
median filter.
 eliminated = forced to have the value equal the median
intensity of the neighbors.
 larger clusters are affected considerably less

134
Example : Median Filters

135
Sharpening Spatial Filters
 to highlight fine detail in an image
 or to enhance detail that has been blurred, either in error
or as a natural effect of a particular method of image
acquisition.

136
 Blurring vs. Sharpening

 as we know that blurring can be done in spatial domain by

pixel averaging in a neighbors
 since averaging is analogous to integration
 thus, we can guess that the sharpening must be accomplished
by spatial differentiation.

137
Derivative operator

 the strength of the response of a derivative operator is

proportional to the degree of discontinuity of the
image at the point at which the operator is applied.
 thus, image differentiation
 enhances edges and other discontinuities (noise)
 deemphasizes area with slowly varying gray-level values.

138
First-order derivative
 a basic definition of the first-order derivative of a one-
dimensional function f(x) is the difference

f
 f ( x  1)  f ( x)
x

139
Second-order derivative
 similarly, we define the second-order derivative of a one-
dimensional function f(x) is the difference

 f
2
 f ( x  1)  f ( x  1)  2 f ( x)
x 2

140
Foundation
Foundation
 First and Second-order derivative of f(x,y)
 when we consider an image function of two variables,
f(x,y), at which time we will dealing with partial
derivatives along the two spatial axes.

f ( x, y ) f ( x, y ) f ( x, y )
Gradient operator f   
xy x y
Laplacian operator
 f ( x , y )  f ( x, y )
2 2

(linear operator)  f 
2

143 x 2
y 2
 Discrete Form of Laplacian

from  f 2
 f ( x  1, y )  f ( x  1, y )  2 f ( x, y )
x 2

 f
2
 f ( x, y  1)  f ( x, y  1)  2 f ( x, y )
y 2

yield,

 f  [ f ( x  1, y )  f ( x  1, y )
2

 f ( x, y  1)  f ( x, y  1)  4 f ( x, y )]
144
Result Laplacian mask

145
Laplacian mask implemented an
extension of diagonal neighbors

146
Other implementation of Laplacian
masks

give the same result, but we have to keep in mind that

when combining (add / subtract) a Laplacian-filtered
147image with another image.
Effect of Laplacian Operator
 as it is a derivative operator,
 it highlights gray-level discontinuities in an image
 it deemphasizes regions with slowly varying gray levels
 tends to produce images that have
 grayish edge lines and other discontinuities, all superimposed
on a dark,
 featureless background.

148
Correct the effect of featureless
background

 easily by adding the original and Laplacian image.

 be careful with the Laplacian filter used

if the center coefficient

of the Laplacian mask is
 f ( x, y )   2 f ( x, y ) negative
g ( x, y )  
 f ( x , y )   2
f ( x, y )
if the center coefficient
of the Laplacian mask is
positive
149
Example

 a). image of the North pole

of the moon
 b). Laplacian-filtered image
with
1 1 1
1 -8 1
1 1 1

 c). Laplacian image scaled for

display purposes
 d). image enhanced by
addition with original image
150
151
Mask of Laplacian + addition
 to simply the computation, we can create a mask which
do both operations, Laplacian Filter and Addition the
original image.

152
Mask of Laplacian + addition

g ( x, y )  f ( x, y )  [ f ( x  1, y )  f ( x  1, y )
 f ( x, y  1)  f ( x, y  1)  4 f ( x, y )]
 5 f ( x, y )  [ f ( x  1, y )  f ( x  1, y )
 f ( x, y  1)  f ( x, y  1)]

0 -1 0
-1 5 -1
0 -1 0
153
Example

154
Note  f ( x, y )   f ( x, y )
2
g ( x, y )  
 f ( x, y )   f ( x, y )
2

0 -1 0 0 0 0 0 -1 0
-1 5 -1 = 0 1 0 + -1 4 -1
0 -1 0 0 0 0 0 -1 0

-1 -1 -1 0 0 0 -1 -1 1
-1 9 -1 = 0 1 0 + -1 8 -1
-1 -1 -1 0 0 0 -1 -1 -1

155
Unsharp Masking and Highboost
filtering
Unsharp Masking and Highboost
Filtering
Unsharp masking
Subtract a blurred version of an image from the image itself

gmask ( x, y)  f ( x, y)  f ( x, y)

f(x,y) : The image, f̄(x,y): The blurred image

g ( x, y )  f ( x, y )  k * g mask ( x, y ) ,k 1

High boost Filtering:

g ( x, y)  f ( x, y)  k * g mask ( x, y) ,k 1
Unsharp Masking and Highboost
Filtering
Unsharp Masking and Highboost
Filtering
High-boost Masks (how?)

• A1
• if A = 1, it becomes “standard” Laplacian
sharpening
160
Example

161
  f 
Gradient Operator Gx   x 
f      f 
G y   
 y 
 first derivatives are implemented using the
magnitude of the gradient.

1
f  mag (f )  [G  G ]
2
x
2
y
2

commonly approx.
1
 f  2  f  2
 2

      
 x   y  
f  Gx  Gy
the magnitude becomes nonlinear
162
 z1 z2 z3
Gradient Mask
z4 z5 z6
 simplest approximation, 2x2 z7 z8 z9

Gx  ( z8  z5 ) and Gy  ( z6  z5 )
1 1
f  [G  G ]
2
x
2
y
2
 [( z8  z5 )  ( z6  z5 ) ]
2 2 2

f  z8  z5  z6  z5

163
 z1 z2 z3
Gradient Mask z4 z5 z6
z7 z8 z9
 Roberts cross-gradient operators, 2x2

Gx  ( z9  z5 ) and Gy  ( z8  z6 )
1 1
f  [G  G ]
2
x
2
y
2
 [( z9  z5 )  ( z8  z6 ) ]
2 2 2

f  z9  z5  z8  z6

164
 Gradient Mask z1 z2 z3
z4 z5 z6
 Sobel operators, 3x3 z7 z8 z9

Gx  ( z7  2 z8  z9 )  ( z1  2 z 2  z3 )
G y  ( z3  2 z6  z9 )  ( z1  2 z 4  z7 )

f  Gx  Gy
the weight value 2 is to
achieve smoothing by
giving more important
to the
165 center point
Note
 the summation of coefficients in all masks equals 0,
indicating that they would give a response of 0 in an area
of constant gray level.

166
Example

167
3.7 Combining Spatial
Enhancement Tools
Example of Combining Spatial
Enhancement Methods
 want to sharpen the
original image and bring
out more skeletal detail.
 problems: narrow
dynamic range of gray
level and high noise
content makes the
image difficult to
enhance

169
Example of Combining Spatial
Enhancement Methods
 solve :
1. Laplacian to highlight fine detail
2. gradient to enhance prominent edges
3. gray-level transformation to increase
the dynamic range of gray levels

170
171
172