Spatial Filtering and Edge

Detection
Grey-scale image (8 bits/pixel)
binary decimal
0

255
2
256 shades of grey
Digital Image (8-bit) and Image Matrix (5x5)

0 255 0 0 255
0 255 0 0 255
0 255 0 0 255
0 255 0 0 255
0 255 0 0 255
Binary Image Image matrix
255 255 255 255 255
100 100 100 100 100
50 50 50 50 50
0 0 0 0 0
50 50 50 50 50
Grey Image Image matrix 3
 Digital Color Image and Image Matrix

255 255 0 0 0
255 255 0 0 0
255 255 0 0 0
255 255 0 0 0 255 0 0 255 0
255 255 0 0 0 255 0 0 255 0
Red component 255 0 0 255 0
255 0 0 255 0
255 0 0 0 255
255 0 0 255 0
Image
255 0 0 0 255 Green component
255 0 0 0 255
255 0 0 0 255
255 0 0 0 255
4
Blue component
Image Filtering
54 100 208 1 1 1

49 106 198 ×1/9 1 1 1

45 90 185 1 1 1
23 42 86
3x3 part of Image 3*3 Smoothing Filter
54 100 208

49 106 198
45 90 185

e = 1/9[106 + 54 +100 + 208 +

49 + 198 +45 + 90 + 185]
= 115
Image f (x, y)
 Smoothing Spatial Filtering
54 100 208 1 1 1

49 106 198 × 1/9 1 1 1

45 90 185 1 1 1
23 42 86
3x3 part of Image 3*3 Smoothing Filter
54 100 208

49 106 198
45 90 185

Image f (x, y) 23 42 86

54 100 208

49 115 198

45 90 185

After averaging
 Smoothing Spatial Filtering
Origin x
23 42 86 1 1 1

54 100 208 ×1/9 1 1 1

49 106 185 1 1 1
23 42 86
3*3 Smoothing
54 100 208

49 106 198
Filter
45 90 185
e = 94.77  95
23 42 86

54 95 208

y Image f (x, y) 49 115 198

45 90 185
 Smoothing Spatial Filtering
Origin x
23 42 86 1 1 1

54 100 208 ×1/9 1 1 1

49 106 185 1 1 1
23 42 86
3*3 Smoothing
54 100 208

49 106 198
Filter
45 90 185
e = 94.77  95
23 42 86

54 95 208

y Image f (x, y) 49 115 198

45 90 185
Smoothing Spatial Filtering

Original Image After averaging by 3x3 filter

 Smoothing Spatial Filtering

After averaging by 3x3 filter After averaging by 5x5 filter

 Smoothing Spatial Filtering

After averaging by 15x15 filter After averaging by 35x35 filter

 Weighted Smoothing Filters

 Provides more effective 2 1

1
smoothing
 Pixels closer to the 2 4 2
A= (1/16)
central pixel are more
important 1 2 1
 Often referred to as a
Gaussian filter Weighted
averaging filter
 Application of average filter

Original Image
 Application of average filter

Original Image Smoothed Image

5×5 mask
 Application of average filter

Original Image Smoothed Image Thresholded Image

5×5 mask G(x,y) =255, if f(x,y)>120
=0, otherwise
• Averaging removes finer details
• Thresholding separates gross details
 Example: Spatial filters

Image matrix is given below. Determine the effect of

1. 3x3 and 5x5 averaging filters
2. 3x3 weighted averaging filter

45 56 42 63 54
20 47 56 28 53
63 59 26 38 47
67 36 27 48 51
43 36 42 65 43
 Example: Spatial filters

Image matrix is given below. Determine the effect of

1. 3x3 and 5x5 averaging filters
2. 3x3 weighted averaging filter

45 56 42 63 54 45 56 42 63 54 45 56 42 63 54
20 47 56 28 53 20 47 56 28 53 20 47 56 28 53
63 59 26 38 47 63 59 26 38 47 63 59 26 38 47
67 36 27 48 51 67 36 27 48 51 67 36 27 48 51
43 36 42 65 43 43 36 42 65 43 43 36 42 65 43
Apply filter on top Move filter to
left corner process each pixel
 Example: Spatial filters
Image matrix is given below. Determine the effect of
1. 3x3 and 5x5 averaging filters
2. 3x3 weighted averaging filter
45 56 42 63 54 45 56 42 63 54
20 47 56 28 53 20 46 46 45 53
1 1 1
63 59 26 38 47 63 44 41 42 47
x (1/9) 1 1 1 =
67 36 27 48 51 67 44 42 43 51
1 1 1
43 36 42 65 43 43 36 42 65 43

45 56 42 63 54 1 1 1 1 1 45 56 42 63 54
20 47 56 28 53 1 1 1 1 1 20 47 56 28 53
x (1/25)
63 59 26 38 47 1 1 1 1 1= 63 59 46 38 47
67 36 27 48 51 1 1 1 1 1 67 36 27 48 51
43 36 42 65 43 1 1 1 1 1 43 36 42 65 43
 Example: Spatial filters

Image matrix is given below. Determine the effect of

1. 3x3 weighted averaging filters
45 56 42 63 54 45 56 42 63 54
20 47 56 28 53 20 47 45 44 53
1 2 1
63 59 26 38 47 x(1/16) 2 4 2 = 63 47 39 40 47
67 36 27 48 51 67 43 38 44 51
1 2 1
43 36 42 65 43 43 36 42 65 43
 Sharpening Spatial Filters

• Remove blurring in images

• Highlight transition in intensity (edges)
• Uses spatial differentiation
• Differentiation measures the rate of change of a pixel
intensities
• Image gradient is based on differentiation
 Image Gradients
• Is fundamental building block of many computer vision and image
processing techniques
• Use gradients to
• detecting edges
• find contours
• outlines of objects
• Compute gradient magnitude and gradient orientation
• Image descriptors such as Histogram of Oriented Gradients and
SIFT are built upon image gradient representations
 Detection of Edges
Computation of Image gradients is a pre-processing
step to detecting edges in images
 Detection of Edges
Computation of Image gradients is a pre-processing
step to detecting edges in images
 Gradient of image
• Image pixels
f(x-1,y)
• Gradient along vertical direction
gy = f(x+1,y) – f(x-1,y)
f(x,y-1) f(x,y) f(x,y+1)
• Gradient along horizontal direction
f(x+1,y)
gx = f(x,y+1) – f(x,y-1)

Mask/ filter for gx -1 0 1

-1
Mask/ filter for gy
0
1
Image Gradient/ Gradient Vector
 gy 
𝑀= 𝑔𝑥 2 + 𝑔𝑦 2  ( x, y)  tan  
1

 gx 
• M is magnitude and α is angle of gradient
• Size of matrix for magnitude and angle is same as image
• Magnitude is the value of rate of change in the direction of
gradient vector
• Magnitude of gradient is used to measure how strong the
change in image intensity is
 Image Gradients
• The gradient orientation is used to determine in which direction
the change in intensity is pointing
• Direction of edge is normal to direction of gradient
• Direction of gradient is also called edge normal
 Image Gradients
• The gradient orientation is used to determine in which direction
the change in intensity is pointing
• Direction of edge is normal to direction of gradient
• Direction of gradient is also called edge normal
 Gradient of image

image gx gy
10 10 50 10 10 -1
-1 0 1 -1
10 10 50 10 10 0
0
10 10 50 10 10 1
1
10 10 50 10 10
gy 𝑀𝑎𝑔 = 𝑔𝑥 2 + 𝑔𝑦 2
gx
10 40 0 -40 10 10 10 50 10 10 10 10 50 10 10
10 40 0 -40 10 0 0 0 0 0 10 40 0 40 10
10 40 0 -40 10 0 0 0 0 0 10 40 0 40 10
10 40 0 -40 10 10 10 50 10 10 10 10 50 10 10
 Sobel Filters for gradient
Filter for gx Filter for gy
image
10 10 50 10 10 -1 0 1 -1 -2 -1
10 10 50 10 10 -2 0 2 0 0 0

10 10 50 10 10 -1 0 1 1 2 1

10 10 50 10 10

gy 𝑀𝑎𝑔 = 𝑔𝑥 2 + 𝑔𝑦 2
gx
10 10 50 10 10 10 10 50 10 10 10 10 50 10 10
10 160 0 -160 10 10 0 0 0 10 10 160 0 160 10
10 160 0 -160 10 10 0 0 0 10 10 160 0 160 10
10 10 50 10 10 10 10 50 10 10 10 10 50 10 10
 Gradient Magnitudes
Filter for gx Filter for gy Magnitude using Sobel
-1 0 1 -1 -2 -1 10 10 50 10 10
-2 0 2 0 0 0 10 160 0 160 10
image
-1 0 1 1 2 1 10 160 0 160 10
10 10 50 10 10 50
10 10 10 10
10 10 50 10 10
10 10 50 10 10 Filter for gx Filter for gy Magnitude
10 10 50 10 10 10 10 50 10 10
-1 0 1 -1
10 40 0 40 10
0
10 40 0 40 10
1
10 10 50 10 10
 Image Gradients

Image
 Image Gradients

Image

gx
 Image Gradients

Image

gx gy
 Image Gradients

Image

Magnitude, M
gx gy
 Laplacian Filter

0 1 0 -1 -1 -1

1 -4 1 -1 8 -1

0 1 0 -1 -1 -1

Detects horizontal Detects horizontal,

and vertical edges vertical and diagonal
edges
 Laplacian Filter
Image Laplacian filter
10 10 100 10 10 0 -1 0
10 10 100 10 10
100 100 100 100 100 -1 4 -1
10 10 100 10 10
*
10 10 100 10 10 0 -1 0

10 10 100 10 10
10 -180 180 -180 10
100 180 0 180 100
=
10 -180 180 -180 10
10 10 100 10 10
 Edge Detection Filters and Noise
Edge detectors are sensitive to noise

Image
after
Noisy image edge
detection

37
 Edge Detection Filters and Noise
Edge detectors are sensitive to noise

Image
after
Noisy image threshold
ing

Image
Image after
after threshold
averaging ing

38
 Edge Detection Filters and Noise
Edge detectors are sensitive to noise

Image
after
Noisy image edge
detection

Image
Image after
after edge
averaging detection

Apply edge detection after image smoothing

39
 Gaussian Filter (Smoothing filter)
• Gaussian filter with mean 0 and standard deviation, σ is

3×3 filter with σ = 1 5×5 filter with σ = 1 5×5 filter with σ = 2

1 2 1 1 4 7 4 1 1 1 1 1 1
4 20 33 20 4 1 2 2 2 1
(1/16) 2 4 2 (1/330) 7 33 54 33 7 (1/34) 1 2 2 2 1
4 20 33 20 4 1 2 2 2 1
1 2 1
1 4 7 4 1 1 1 1 1 1

• Blurring increases with the increase in σ

 Gaussian Filter (Smoothing filter)
Original Image
100 100 100 100 100 100 100
100 100 100 100 100 100 100
100 100 100 100 100 100 100
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 Filtered Image
100 100 100 100 100 100 100
3×3 filter with σ = 1
100 100 100
1 2 1
100 75 100
(1/16) 2 4 2 0 25 0
0 0 0
1 2 1
0 0
0 0 0 0 0 0 0
 Gaussian Filter (Smoothing filter)
Filtered Image for σ = 1
100 100 100 100 100 100 100
5×5 filter with σ = 1
100 100 100 100 100 100 100
1 4 7 4 1 100 100 70 70 70 100 100
4 20 33 20 4 0 0 30 30 30 0 0
(1/330)
7 33 54 33 7 0 0 10 10 10 0 0
4 20 33 20 4 0 0 0 0 0 0 0
1 4 7 4 1 0 0 0 0 0 0 0

5×5 filter with σ = 2 Filtered Image for σ = 2

1 1 1 1 1 100 100 100 100 100 100 100
100 100 100 100 100 100 100
(1/34) 1 2 2 2 1
1 2 2 2 1 100 100 60 60 60 100 100
1 2 2 2 1 0 0 40 40 40 0 0
1 1 1 1 1 0 0 10 10 10 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
 Difference of Gaussian
Filtered Image for σ = 1 • Subtract filtered image with
high sigma from filtered image
100 100 100 100 100 100 100
with low sigma
100 100 100 100 100 100 100 • Called difference of Gaussian
100 100 70 70 70 100 100
0 0 30 30 30 0 0 Difference of Gaussian (DoG)
(Absolute Difference)
0 0 10 10 10 0 0
0 0 0 0 0 0 0 100 100 100 100 100 100 100
0 0 0 0 0 0 0 100 100 100 100 100 100 100
100 100 10 10 10 100 100
Filtered Image for σ = 2
0 0 10 10 10 0 0
100 100 100 100 100 100 100
0 0 0 0 0 0 0
100 100 100 100 100 100 100
0 0 0 0 0 0 0
100 100 60 60 60 100 100
0 0 0 0 0 0 0
0 0 40 40 40 0 0
0 0 10 10 10 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
 Laplacian of Gaussian
• Apply Gaussian filter to reduce noise in the image
• Apply Laplacian filter on the image to detect edges

Image Gaussian filter Laplacian filter

100 100 100 100 100 1 2 1 0 -1
0
100 100 100 100 100
(1/16) 2 4 2 -1 4
-1
10 10 10 10 10
1 2 1 0
10 10 20 10 10 -1 0
10 10 10 10 10
Laplacian of Gaussian
Gaussian filtered image
10 10 100 10 10 10 10 100 10 10
10 77.5 10 10 10
100 33.5 100 100 100
10 11.25 10 10 10
10 10 100 10 10 10 10 100 10 10