Lectures 6- 7-8: Edge Detection

Dr. V Masilamani
[email protected]
Department of Computer Science and Engineering
IIITDM Kancheepuram
Chennai-127

Computer Vision Lectures 6-7-8: Edge Detection 1 / 34

Overview

1 What is edge?

2 Why to detect edge?

3 Which features of object cause edges in its image

4 How to detect edge?

Edge detection using first derivative operators
Edge detection using second derivative operators

5 Pre-processing before applying edge detection operator

6 Some applications of edge detection

7 Acknowledgements

Computer Vision Lectures 6-7-8: Edge Detection 2 / 34

What is edge

Computer Vision Lectures 6-7-8: Edge Detection 3 / 34

What is edge (cont.)

Computer Vision Lectures 6-7-8: Edge Detection 4 / 34

Why to detect edge?

I Edge in image provides structural information of object in the scene

Computer Vision Lectures 6-7-8: Edge Detection 5 / 34

Why to detect edge? (cont.)

Computer Vision Lectures 6-7-8: Edge Detection 6 / 34

Why to detect edge? (cont.)

I Structural information of
object is more important than
the other information such as
colour

Computer Vision Lectures 6-7-8: Edge Detection 7 / 34

Which features of object cause edges in its image

I Discontinuity in surface
normal of the object
I Discontinuity in depth
I Discontinuity in colour of the
surface of the object
I Discontinuity in light intensity

Computer Vision Lectures 6-7-8: Edge Detection 8 / 34

Models of edge

I Edge Point: If intensity profile

changes at a point p
significantly, then p is called
as edge point
I Edge Map: Set of edge points
is called as edge map

Computer Vision Lectures 6-7-8: Edge Detection 9 / 34

Edge detection using first derivative operators

I Let f (x) be a function

I
f 0 (x) = lim (f (x + h) − f (x))/h
h→0

I Discrete Approximation 1: f 0 (x) = f (x + 1) − f (x) (Sub h=1)

• f 0 (x) = Cov (f (x), w (x)), where w = (1, −1) —(1)
I f = (5, 5, 5, 5, 5, 10, 10, 10, 10, 10)
I f 0 = 0, 0, 0, 0, 5, 0, 0, 0, 0, 0
• An Approximation to (1), f 0 (x) = Cov (f (x), w (x), where
w = (1, 0, −1) ——(2)
I f = (5, 5, 5, 5, 5, 10, 10, 10, 10, 10)
I f 0 = 0, 0, 0, 0, 5, 5, 0, 0, 0, 0

I Discrete Approximation 2: f 0 (x) = f (x) − f (x − 1) (Sub h=-1)

Computer Vision Lectures 6-7-8: Edge Detection 10 / 34

Edge detection using first derivative operators (cont.)

Computer Vision Lectures 6-7-8: Edge Detection 11 / 34

Edge detection using first derivative operators (cont.)
I How to detect edge point in 1D signal f?
if (f 0 (x) > T0 ) then x is an edge point
First Derivative Filters for Images
I Robert
Along X direction
 
1
−1
Along Y direction


1 −1
Along diagonal direction
 
1 0
0 −1
Along anti diagonal direction

Computer Vision Lectures 6-7-8: Edge Detection 12 / 34

Edge detection using first derivative operators (cont.)

 
0 1
−1 0

I How to detect edge point in 2D signal f?

• If |∇(f (x, y ))| > T0 , then (x, y ) is an edge pixel

Computer Vision Lectures 6-7-8: Edge Detection 13 / 34

Edge detection using first derivative operators (cont.)

Issues with Robert Filters

I Not center aligned
I Noise pixels are also detected as edge pixel
Approximating to make it center aligned, and remove noise before
taking derivatives
I Instead of  
1
−1
, consider  
1
 0 
−1

I Remove noise using sum filter before applying derivatives

Computer Vision Lectures 6-7-8: Edge Detection 14 / 34

Edge detection using first derivative operators (cont.)

I Prewitt filter along X -dir (to detect horizontal edge) is defined as

 
1 1 1
Px =  0 0 0 
−1 −1 −1

Prewitt in X - dir filter can be seen as convolution of smoothing and

differentiating operator as follows.
 
1

 0 ∗ 1 1 1
−1

Computer Vision Lectures 6-7-8: Edge Detection 15 / 34

Edge detection using first derivative operators (cont.)

I Similarly along Y dir (to detect vertical edge), the Prewit filter will
be
 
1 0 −1
Py =  1 0 −1 
1 0 −1

Prewitt in Y -dir filter can be seen as convolution of smoothing and

differentiating operator as follows.
 

1
1 0 −1 ∗  1 
1

I Horizontal Edge Detection: fx = Eh (f ) = f ∗ Px

I Vertical Edge Detection: fy = Ev (f ) = f ∗ Py

Computer Vision Lectures 6-7-8: Edge Detection 16 / 34

Edge detection using first derivative operators (cont.)

Edge detection using gradient:

Computer Vision Lectures 6-7-8: Edge Detection 17 / 34

Edge detection using first derivative operators (cont.)

Edge detection using gradient:

I Edge Detection:
if( |∇(f (x, y ))| > T ) then E (x, y ) = 1
else E (x, y ) = 0

Computer Vision Lectures 6-7-8: Edge Detection 18 / 34

Edge detection using first derivative operators (cont.)

Edge detection using gradient:

Computer Vision Lectures 6-7-8: Edge Detection 19 / 34

Edge detection using first derivative operators (cont.)
• Sobel filter along X -dir (to detect horizontal edge) is defined as
 
1 2 1
Sx =  0 0 0 
−1 −2 −1

Sobel in X - dir filter can be seen as convolution of smoothing and

differentiating operator as follows.
 
1

 0 ∗ 1 2 1
−1

• Similarly along Y dir (to detect vertical edge), the Sobel filter will be
 
1 0 −1
Sy =  2 0 −2 
1 0 −1

Sobel in Y -dir filter can be seen as convolution of smoothing and

differentiating operator as follows.
Computer Vision Lectures 6-7-8: Edge Detection 20 / 34
Edge detection using first derivative operators (cont.)

 

1
1 0 −1 ∗ 2 
1

• Horizontal Edge Detection: fx = Eh (f ) = f ∗ Sx

• Vertical Edge Detection: fy = Ev (f ) = f ∗ Sy
• Edge Detection Using gradient:

if( |∇(f (x, y ))| > T ) then E (x, y ) = 1

else E (x, y ) = 0

Computer Vision Lectures 6-7-8: Edge Detection 21 / 34

Edge detection using first derivative operators (cont.)

Edge detection using gradient:

Computer Vision Lectures 6-7-8: Edge Detection 22 / 34

Edge detection using first derivative operators (cont.)

Summary of edge detection using gradient:

Computer Vision Lectures 6-7-8: Edge Detection 23 / 34

Edge detection using second derivative operators

What are the issues with first

derivative based filters

I Even when intensity profile is

slowly varying at a point p, p
may be be declared as edge
point

I Localization of edge point is

not effective

Computer Vision Lectures 6-7-8: Edge Detection 24 / 34

Edge detection using second derivative operators
(cont.)
Laplacian Filter Design

Computer Vision Lectures 6-7-8: Edge Detection 25 / 34

Edge detection using second derivative operators
(cont.)

Properties of Lapacian filter

I The output of the filter is 0, when applied on homogeneous region
I The magnitude of output of the filter is high, when applied on edges
I Zero Crossing(+ve/-ve followed by -ve/+ve) indicates the presence
of edges
I Zero Crossing can be used to located the edge point effectively

Computer Vision Lectures 6-7-8: Edge Detection 26 / 34

Edge detection using second derivative operators
(cont.)

Other variants of Laplacian Filter -satisfying above

properties

Computer Vision Lectures 6-7-8: Edge Detection 27 / 34

Edge detection using second derivative operators
(cont.)

Computer Vision Lectures 6-7-8: Edge Detection 28 / 34

Edge detection using second derivative operators
(cont.)

Smoothing followed by Laplacian Filter Design

-Laplacian of Gaussian Filter(LOG)

Computer Vision Lectures 6-7-8: Edge Detection 29 / 34

Pre-processing before applying edge detection operator

I To obtain accurate edges, do pre-processing such as

• Noise removal
• Contrast enhancement
• Sharpen the image

Computer Vision Lectures 6-7-8: Edge Detection 30 / 34

Pre-processing before applying edge detection operator
(cont.)

Computer Vision Lectures 6-7-8: Edge Detection 31 / 34

Some applications of edge detection

I Object detection
I Measure computation

Computer Vision Lectures 6-7-8: Edge Detection 32 / 34

Acknowledgements

I Images are downloaded from internet sources

Computer Vision Lectures 6-7-8: Edge Detection 33 / 34

 Thank You! :)

Computer Vision Lectures 6-7-8: Edge Detection 34 / 34