Image Filtering

Imran N. Junejo
Associate Professor
Faculty of Computer Science
Institute of Business Administration - Karachi

1
 Lecture outline
• point-operations
• Learn about image filtering

2
Point Operations
• Point transformation: changes a pixel’s value without changing its
location.

In this case f is referred to as a grey level transformation function or a point

processing operation
3
Point Operations
• Point transformation: changes a pixel’s value without changing its
location.
• Can be applied by: Arithmetic Operations
Arithmetic Operations

• A useful class of gray level transformations is the set of arithmetic

operations, depicted in graphical form in the following figure:

5
 Arithmetic Operations

Overflow issue?

6
 Arithmetic Operations
What if b is negative?

If b is negative, the image will darken

7
Analytic Transformations
• Graylevel transformations can be specified using analytic functions such
as the logarithm, exponential, or power functions:

8
Logarithmic Transformations can be used to brighten the intensities of an
image (like the Gamma Transformation, where gamma < 1). More often, it is
used to increase the detail (or contrast) of lower intensity values.

9
Point Operations
• Point transformation: changes a pixel’s value without changing its
location.
• Can be applied by: Thresholding
Intensity Transformations
Point Operations
• Point transformation: changes a pixel’s value without changing its
location.
• Can be applied by: Histogram Modifications
 What is a Filter
• Point operations are limited (why)
• They cannot accomplish tasks like sharpening
or smoothing

13
 Noise Filtering or Image Filtering
• Given a corrupted image I by noise n,
attenuate, if possible eliminate n, without
altering I significantly.

14
 Image filtering
• Filtering refers to linear transforms that change the
frequency contents of signals.
• Transformations where each output pixel is a
function of the input pixel and its neighbors.
• Really important!
– Enhance images
• Denoise, resize, increase contrast, etc.
– Extract information from images
• Texture, edges, distinctive points, etc.
– Detect patterns
• Template matching
15
 Image filtering
• Filtering refers to linear transforms that change the
frequency contents of signals (def: signal processing).
• Transformations where each output pixel is a
function of the input pixel and its neighbors.
• Really important!
– Enhance images
• Denoise, resize, increase contrast, etc.
– Extract information from images
• Texture, edges, distinctive points, etc.
– Detect patterns
• Template matching
16
 Smoothing an image by
averaging
• Replace each pixel by the average of its neighboring pixels
• Assume a 3x3 neighborhood:

P0 P1 P2
P3 P4 P5
P6 P7 P8

17
P0 P1 P2
P3 P4 P5
P6 P7 P8

18
 Averaging filter

19
 G

• In general a filter applies a function over the values of a

small neighborhood of pixels to compute the result
• The size of the filter = the size of the neighborhood: 3x3,
5x5, 7x7, …, 21x21,..
• The shape of the filter region is not necessarily square,
can be a rectangle, a circle…
• Filters can be linear or nonlinear
20
 Linear Filtering
• The output is the linear combination of the
neighborhood pixels
h[ m, n]   g[ k , l ] f [ m  k , n  l ]
k ,l

1 3 0 1 0 -1
2 10 2 1 0.1 -1 = 5
4 1 1 1 0 -1
Image filter Filter Output
21
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0
0 0
0 0
0 90
90 0
0 90
90 90
90 90
90 0
0 0
0

0
0 0
0 0
0 90
90 90
90 90
90 90
90 90
90 0
0 0
0

0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0

0
0 0
0 90
90 0
0 0
0 0
0 0
0 0
0 0
0 0
0

0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
22
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
23
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
24
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20 30

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
25
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20 30 30

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
26
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20 30 30

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0
?
0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
27
k ,l Credit: S. Seitz
Image filtering 1 1 1

g[ , ] 1 1 1

1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20 30 30

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0
?
0 0 0 90 0 90 90 90 0 0

0 0 0 90 90 90 90 90 0 0 50

0 0 0 0 0 0 0 0 0 0

0 0 90 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
28
k ,l Credit: S. Seitz
Image filtering g[ , ] 1 1 1
1 1 1
1 1 1

f [.,.] h[.,.]
0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 10 20 30 30 30 20 10

0 0 0 90 90 90 90 90 0 0 0 20 40 60 60 60 40 20

0 0 0 90 90 90 90 90 0 0 0 30 60 90 90 90 60 30

0 0 0 90 90 90 90 90 0 0 0 30 50 80 80 90 60 30

0 0 0 90 0 90 90 90 0 0 0 30 50 80 80 90 60 30

0 0 0 90 90 90 90 90 0 0 0 20 30 50 50 60 40 20

0 0 0 0 0 0 0 0 0 0 10 20 30 30 30 30 20 10

0 0 90 0 0 0 0 0 0 0 10 10 10 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0

h[ m, n]   g[ k , l ] f [ m  k , n  l ]
29
k ,l Credit: S. Seitz
 g

f'

30
Box Filter

g[ , ]
What does it do?
• Replaces each pixel with 1 1 1
an average of its
neighborhood 1 1 1

1 1 1
• Achieve smoothing effect
(remove sharp features)

31
Slide credit: David Lowe (UBC)
Smoothing with box filter

32
 Practice with linear filters

?
0 0 0
0 1 0
0 0 0

Original

33
Source: D. Lowe
 Practice with linear filters

0 0 0
0 1 0
0 0 0

Original Filtered
(no change)

34
Source: D. Lowe
 Practice with linear filters

?
0 0 0
0 0 1
0 0 0

Original

35
Source: D. Lowe
 Practice with linear filters

0 0 0
0 0 1
0 0 0

Original Shifted left

By 1 pixel

36
Source: D. Lowe
 Practice with linear filters

-
0 0 0 1 1 1
0
0
2
0
0
0
1
1
1
1
1
1
?
(Note that filter sums to 1)
Original

37
Source: D. Lowe
 Practice with linear filters

-
0 0 0 1 1 1
0 2 0 1 1 1
0 0 0 1 1 1

Original Sharpening filter

- Accentuates differences with local
average

38
Source: D. Lowe
Sharpening

39
Source: D. Lowe
Other filters

1 0 -1
2 0 -2
1 0 -1
Sobel

Vertical Edge
(absolute value) 40
Other filters

1 2 1
0 0 0
-1 -2 -1
Sobel

Horizontal Edge
(absolute value) 41
 Convolution
[filtering in frequency domain]

f i, j   I * H   I k , l H i  k , j  l 
k l
H
I  Image
X  flip
H7 H8 H9 H1 H2 H3
H  Kernel H4 H5 H6 H4 H5 H6
H1 H2 H3 H7 H8 H9

I Y  flip
I1 I2 I3 H9 H8 H7 I * H  I1 H 9  I 2 H 8  I 3 H 7
I4 I5 I6  H6 H5 H4  I 4 H 6  I5 H5  I6 H 4
I7 I8 I9 H3 H2 H1  I 7 H 3  I 8 H 2  I 9 H1
42
 Gaussian Filter
Special type of smoothing kernel

G x, y  
1  x2  y2
exp  
 
2 2
 2 
 2 

H i, j  
1
exp  

 i  k  12   j  k  12 
2 2
 2 
 2 

where H i, j  is 2k  1 2k  1 array

43
1-D Gaussian Filter

 x2

g ( x)  e 2 2

g ( x)  .011 .13 .6 1 .6 .13 .011

44
Smoothing with Gaussian filter

45
Smoothing with box filter

46
 Gaussian filters
• Remove “high-frequency” components from the
image (low-pass filter)
– Images become more smooth
• Convolution with self is another Gaussian
– So can smooth with small-width kernel, repeat, and
get same result as larger-width kernel would have
– Convolving two times with Gaussian kernel of width σ
is same as convolving once with kernel of width σ√2
• Separable kernel
– Factors into product of two 1D Gaussians

47
Source: K. Grauman
 Separability example
2D convolution
(center location only)

The filter factors

into a product of 1D
filters:

Perform convolution =
along rows: *

Followed by convolution =
along the remaining column: *
48
Source: K. Grauman
 Properties of Gaussian
• Most common natural model
• Smooth function, it has infinite number of
derivatives
• Fourier Transform of Gaussian is Gaussian.
• Convolution of a Gaussian with itself is a
Gaussian.
• Gaussian kernels is separable.

49
 Linear filtering and convolution

Original Noisy

Filtered
Filtered
σ=3.0
σ=1.5
50
 Gaussian filters

= 1 pixel = 5 pixels = 10 pixels = 30 pixels

51
Types of Linear Filters

52
 Practical matters
• What about near the edge?
– the filter window falls off the edge of the image
– need to extrapolate
– methods:
• clip filter (black)
• wrap around
• copy edge
• reflect across edge

53
Source: S. Marschner
 Examples
• I2 = cv2.blur(I,(5,5))
• I2 = cv2.blur(I,(15,15))
• I2 = cv2.blur(I,(50,50))

cv2.GaussianBlur(I, (3,3),1.5)

Based on A Practical Introduction to Computer Vision with OpenCV 2 by

Images Slide 54
Kenneth Dawson-Howe © Wiley & Sons Inc. 2014
 Smoothing – Examples

Based on A Practical Introduction to Computer Vision with OpenCV by

Images Slide 55
Kenneth Dawson-Howe © Wiley & Sons Inc. 2014
 Smoothing – Median Filter (non-linear)
• Use the median value… 11 18 20 21 23 25 25 30 250
Median = 23 Average = 47
• Not affected by noise
• Doesn’t blur edges much
• Can be applied iteratively
• Damages thin lines and sharp corners
– Change region shape

• Computational expensive
– Standard – O(r2log r)
– Huang – O(r)
– Perreault (2007) – O(1)
medianBlur(image, smoothed_image, 5);

Based on A Practical Introduction to Computer Vision with OpenCV by

Images Slide 56
Kenneth Dawson-Howe © Wiley & Sons Inc. 2014
 Smoothing – Examples

noisy 3x3 smoothing Gaussian Median

Based on A Practical Introduction to Computer Vision with OpenCV by

Images Slide 57
Kenneth Dawson-Howe © Wiley & Sons Inc. 2014
 Take-home messages
• Linear filtering is sum of dot
1 1 1
product at each position 1 1 1

– Can smooth, sharpen, translate 1 1 1

(among many other uses)

• Be aware of details for filter size,

extrapolation, cropping

58
 Suggested Reading

• Chapter 3, Emanuele Trucco, Alessandro

Verri, "Introductory Techniques for 3-D
Computer Vision”
• Chapter 7,8, David A. Forsyth and Jean
Ponce, "Computer Vision: A Modern
Approach“

59