Introduction to Computer Vision

(ME-4201)

Shape Detection
Shape Detection
Line detection

Shape Detection |2
Shape Detection
Line fitting
Given: Many ( xi , yi )pairs y
y = mx + c
Find: Parameters (m, c) ( xi , yi )

Minimize: Average square distance:

yi − mxi − c
( yi − mxi − c) 2
E= x
i N
Using:
E E c = y −m x
=0 & =0
m c
 ( x − x)( y − y)
i i
Note:
y x m= i

y= i
i
x= i
i
 ( x − x) i
2

N N i
Shape Detection |3
Shape Detection
Line fitting
y

Line that minimizes E!!

Solution: Use a different parameterization

(same as the one we used in computing Minimum Moment of Inertia)

1
E=
N
 (
i
 − xi cos  + yi sin  ) 2

Note: Error E must be formulated carefully!

Shape Detection |4
Shape Detection
Polynomial fitting
y
Find Polynomial:

y = f ( x) = ax3 + bx 2 + cx + d
that best fits the given points ( xi , yi )
Minimize:
x
1
 i − + + +
3 2 2
[ y ( axi bxi cxi d )]
N i

Using: E E E E
=0 , =0 , =0 , =0
a b c d

Note: f (x) is LINEAR in the parameters (a, b, c, d)

Shape Detection |5
Shape Detection
Circle fitting
Find Circle:

that best fits the given points ( xi , yi )

Minimize:

Using:

𝜕𝐸 𝜕𝐸 𝜕𝐸
= 0, = 0, =0
𝜕𝑎 𝜕𝑏 𝜕𝑟

Shape Detection |6
Shape Detection
Detection?

Slide credit: David Jacobs

Shape Detection |7
Hough Transform
Parameter space and voting

y = mx + c y
Equation of Line: y = mx + c
Find: (m, c)

Consider point: ( xi , yi ) ( xi , yi )
x
Image Space
yi = mxi + c or c = − xi m + yi
m
Parameter space also called Hough Space

(m, c)
c
Shape Detection |8
Hough Transform
Parameter space and voting
• A point can belong to many lines.
• If there is a line crosses many points, that is the line we are looking for.

c
Shape Detection |9
Hough Transform
Hough Line fitting y
Algorithm:

• Quantize Parameter Space (m, c)

(m, c)
• Create Accumulator Array A(m, c)
x
• Set A(m, c) = 0 m, c Parameter Space
A(m, c)
• For each image edge ( xi , yi ) increment: 1 1
1 1
A(m, c) = A(m, c) + 1 1 1
2
• If (m, clies
) on the line: 1 1
1 1
c = − xi m + yi 1 1

• Find local maxima in A(m, c)

Shape Detection |10

Hough Transform
Hough Line fitting
NOTE: −  m   y
Large Accumulator ( xi , yi )
More memory and computations

Improvement: (Finite Accumulator Array Size)

x
Line equation:  = − x cos + y sin  Image Space

Here 0    2 
0     max
Given points ( xi , yi ) find (  , ) ?

Hough Space Sinusoid

Hough Space

Shape Detection |11
Hough Transform
Hough Line fitting

Image space Votes

Horizontal axis is θ,
vertical is rho.
Shape Detection |12
Hough Transform
Hough Line fitting

Image votes
space

Shape Detection |13

Hough Transform
Hough Line fitting

Image votes
space

Shape Detection |14

Hough Transform
Hough Line fitting

Shape Detection |15

Hough Transform
Hough Line fitting
Move to the center
𝑛𝑐𝑜𝑙 𝑛𝑟𝑜𝑤𝑠
𝑟= 𝑥− cos 𝜃 + 𝑦 − sin 𝜃
2 2
𝜃: 0 → 180𝑜
1 2 2
𝑟: 0 → 𝑛𝑐𝑜𝑙 + 𝑛𝑟𝑜𝑤
2

Shape Detection |16

Hough Transform
Hough Circle fitting
Equation of Circle:

( xi − a) 2 + ( yi − b) 2 = r 2

If radius is known: (2D Hough Space)

Accumulator Array A(a, b)

Shape Detection |17

Hough Transform
Hough Circle fitting

Equation of Circle:

( xi − a) 2 + ( yi − b) 2 = r 2

If radius is not known: 3D Hough Space!

Use Accumulator array A(a, b, r )

What is the surface in the hough space?

Shape Detection |18

Hough Transform
Hough Circle fitting

Edge Location ( xi , yi )
Edge Direction i

Assume radius is known:

a = x − r cos 
b = y − r sin 

Need to increment only one point in Accumulator!!

Shape Detection |19
Hough Transform
General Hough Transform
• Model Shape NOT described by equation

Shape Detection |20

Hough Transform
General Hough Transform
• Model Shape NOT described by equation

Shape Detection |21

Hough Transform
General Hough Transform
Find Object Center ( xc , yc ) given edges ( xi , yi , i )
Create Accumulator Array A( xc , yc )
Initialize: A( xc , yc ) = 0 ( xc , yc )
For each edge point ( xi , yi , i )
For each entry rki in table, compute:

xc = xi + rki cos  ki
yc = yi + rki sin  ki
Increment Accumulator: A( xc , yc ) = A( xc , yc ) + 1
Find Local Maxima in A( xc , yc )
Shape Detection |22
Hough Transform
General Hough Transform

Shape Detection |23

Hough Transform
General Hough Transform

• Works on Disconnected Edges

• Relatively insensitive to occlusion

• Effective for simple shapes (lines, circles, etc)

• Trade-off between work in Image Space and Parameter Space

• Handling inaccurate edge locations:

• Increment Patch in Accumulator rather than a single point

Shape Detection |24