#1

Lecture 9:
Hough Transform and Thresholding

Saad Bedros
[email protected]

ME5286 – Lecture 9
 Hough Transform
#2
• Robust method to find a shape in an image
• Shape can be described in parametric form
• A voting scheme is used to determine the correct
parameters

ME5286 – Lecture 9
 Example: Line fitting
• Why fit lines?
Many objects characterized by presence of straight lines

• Can we do it with edge detection? Use edge information

ME5286 – Lecture 9
 Difficulty of line fitting
• Extra edge points (clutter),
multiple models:
– which points go with which
line, if any?
• Only some parts of each line
detected, and some parts
are missing:
– how to find a line that bridges
missing evidence?
• Noise in measured edge
points, orientations:
– how to detect true underlying
parameters?
ME5286 – Lecture 9
 Voting
• It’s not feasible to check all combinations of features by fitting
a model to each possible subset.

• Voting is a general technique where we let the features vote

for all models that are compatible with it.
– Cycle through features, cast votes for model parameters.
– Look for model parameters that receive a lot of votes.

• Noise & clutter features will cast votes too, but typically their
votes should be inconsistent with the majority of “good”
features.

ME5286 – Lecture 9
 Fitting lines: Hough transform
• Given points that belong to a line, what is
the line?
• How many lines are there?
• Which points belong to which lines?

• Hough Transform is a voting technique

that can be used to answer all of these
questions.
Main idea:
1. Record vote for each possible line on
which each edge point lies.
2. Look for lines that get many votes.

ME5286 – Lecture 9
 Finding lines in an image: Hough space
y b

b0

x m0 m
image space Hough (parameter) space

Connection between image (x,y) and Hough (m,b)

spaces
– A line in the image corresponds to a point in Hough space
– To go from image space to Hough space:
• given a set of points (x,y), find all (m,b) such that y = mx + b

ME5286 – Lecture 9 Slide credit: Steve Seitz

 Finding lines in an image: Hough space
y b
y0

x0 x m
image space Hough (parameter) space
Connection between image (x,y) and Hough (m,b) spaces
– A line in the image corresponds to a point in Hough space
– To go from image space to Hough space:
• given a set of points (x,y), find all (m,b) such that y = mx + b

– What does a point (x0, y0) in the image space map to?

– Answer: the solutions of b = -x0m + y0

– this is a line in Hough space
ME5286 – Lecture 9 Slide credit: Steve Seitz
 Finding lines in an image: Hough space
y b
(x1, y1)
y0
(x0, y0)

b = –x1m + y1
x0 x m
image space Hough (parameter) space

What are the line parameters for the line that

contains both (x0, y0) and (x1, y1)?
– It is the intersection of the lines b = –x0m + y0 and
b = –x1m + y1
ME5286 – Lecture 9
 Finding lines in an image: Hough algorithm
y b

x m
image space Hough (parameter) space
How can we use this to find the most likely parameters (m,b) for
the most prominent line in the image space?
• Let each edge point in image space vote for a set of possible
parameters in Hough space
• Accumulate votes in discrete set of bins; parameters with the
most votes indicate line in image space.
ME5286 – Lecture 9
 Hough Transform for Line Detection
#11

Find a subset of n points on an image

that lie on the same straight line.
Write each line formed by a pair of
these points as
yi = axi + b
Then plot them on the parameter space
(a, b):
b = xi a + yi
All points (xi, yi) on the same line will
pass the same parameter space point (a,
b).
Quantize the parameter space and tally
# of times each points fall into the same
accumulator cell. The cell count = # of
points in the same line.

ME5286 – Lecture 9
 Polar representation for lines
Issues with usual (m,b) parameter space: can take on
infinite values, undefined for vertical lines.

Image columns
x d : perpendicular distance
[0,0]
 from line to origin
Image rows

d  : angle the perpendicular

y makes with the x-axis

x cos   y sin   d

Point in image space  sinusoid segment in Hough space

ME5286 – Lecture 9 Kristen Grauman
 Hough Transform in (r, ) plane
#13

To avoid infinity slope,

use polar coordinate to
represent a line.
x cos   y sin   r
Q points on the same
straight line gives Q
sinusoidal curves in (r,
) plane intersecting at
the same (ri, i) cell.

ME5286 – Lecture 9
 Hough transform algorithm
Using the polar parameterization: H: accumulator array (votes)

x cos   y sin   d
Basic Hough transform algorithm d
1. Initialize H[d, ]=0
2. for each edge point I[x,y] in the image
for  = [min to max ] // some quantization
d  x cos   y sin  
H[d, ] += 1
3. Find the value(s) of (d, ) where H[d, ] is maximum
4. The detected line in the image is given by d  x cos   y sin 

Time complexity (in terms of number of votes per pt)?

ME5286 – Lecture 9 Source: Steve Seitz
 Hough Transform for Lines
#15

r  x1 cos( )  y1 sin( ) r (r0 ,0 )

2
( x0 , y0 )
y 1

( x2 , y2 ) -3 -2 -1 1 2 ( x1 , 3y1 )
-1
( x1 , y1 ) ( x2 , y2 )
-2

r0 ( x0 , y0 ) (r ,  )
0 0

0
ME5286 – Lecture 9
x
 Hough Transform for Lines
r 4
#16

r  x1 cos( )  y1 sin( ) (r1 ,1 )

2

y 
-3 -2 -1 1 2 3
( x , y
( x3 , y3 )
1 1 )
-2
( x4 , y4 )
( x1 , y1 ) -4

r1 ( x3 , y3 ) ( x , y )
4 4 (r1 , 1 )
1
ME5286 – Lecture 9
x
 #17

Peak in the parametric

space that corresponds
to the line

ME5286 – Lecture 9
 Hough Transform for Lines
#18

• Domain of the parametric space:

 2   
r   M  N , M  N ,   , 
2 2 2
   2 2
M and N image resolution

Not just lines, any parametric curve!

However increase of dimensions of the
parametric space
ME5286 – Lecture 9
 Original image Canny edges

Vote space and top peaks

ME5286 – Lecture 9 Kristen Grauman

 Showing longest segments found
ME5286 – Lecture 9 Kristen Grauman
 Impact of noise on Hough

y d

x 
Image space Votes
edge coordinates
What difficulty does this present for an implementation?
ME5286 – Lecture 9
 Impact
Impact of
of noise onHough
noise on Hough

Image space Votes

edge coordinates
In this case, everything appears to be “noise”, or random
edge points, but we still see some peaks in the vote space.
ME5286 – Lecture 9
 Extensions
Extension 1: Use the image gradient
1. same
2. for each edge point I[x,y] in the image
 = gradient at (x,y)
d  x cos   y sin 
H[d, ] += 1
3. same
4. same
(Reduces degrees of freedom)

Extension 2
– give more votes for stronger edges
Extension 3
– –change
ME5286 the sampling
Lecture 9 of (d, ) to give more/less resolution
 Extensions
Extension 1: Use the image gradient
1. same
2. for each edge point I[x,y] in the image
compute unique (d, ) based on image gradient at (x,y)
H[d, ] += 1
3. same
4. same
(Reduces degrees of freedom)

Extension 2
– give more votes for stronger edges (use magnitude of gradient)
Extension 3
– change the sampling of (d, ) to give more/less resolution
Extension 4
– The same procedure can be used with circles, squares, or any other
shape…

ME5286 – Lecture 9 Source: Steve Seitz

 Hough transform for circles
• Circle: center (a,b) and radius r
( xi  a) 2  ( yi  b) 2  r 2
• For a fixed radius r, unknown gradient direction

Image space Hough space a

ME5286 – Lecture 9 Kristen Grauman
 Hough transform for circles
• Circle: center (a,b) and radius r
( xi  a) 2  ( yi  b) 2  r 2
• For a fixed radius r, unknown gradient direction

Intersection:
most votes
for center
occur here.

Image space Hough space

ME5286 – Lecture 9 Kristen Grauman
 Hough transform for circles
• Circle: center (a,b) and radius r
( xi  a) 2  ( yi  b) 2  r 2
• For an unknown radius r, unknown gradient direction

b
a
Image space Hough space

ME5286 – Lecture 9 Kristen Grauman

 Hough transform for circles
• Circle: center (a,b) and radius r
( xi  a) 2  ( yi  b) 2  r 2
• For an unknown radius r, unknown gradient direction

b
a
Image space Hough space

ME5286 – Lecture 9 Kristen Grauman

 HT for Circles: Search with fixed R
Equation of Circle:

( xi  a) 2  ( yi  b) 2  r 2

ME5286 – Lecture 9
 31

Multiple Circles with known R

• Multiple circles with the same radius can be found with the same
technique. The centerpoints are represented as red cells in the
parameter space drawing.
• Overlap of circles can cause spurious centers to also be found, such as
at the blue cell. Spurious circles can be removed by matching to circles
in the original image.

ME5286 – Lecture 9
 HT for Circles: Search with unknown R

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 )

ME5286 – Lecture 9
 Hough transform for circles
For every edge pixel (x,y) :
For each possible radius value r:
For each possible gradient direction θ:
// or use estimated gradient at (x,y)
a = x – r cos(θ) // column
b = y + r sin(θ) // row
H[a,b,r] += 1
end
end

ME5286 – Lecture 9
Example: detecting circles with Hough
Original Edges Votes: Penny

Note: a different Hough transform (with separate accumulators)

was used for each circle radius (quarters vs. penny).
ME5286 – Lecture 9
Example: detecting circles with Hough
Original Edges Votes: Quarter

Combined detections

ME5286 – Lecture 9
 Example: iris detection

Gradient+threshold Hough space Max detections

(fixed
radius)

ME5286 – Lecture 9 Kristen Grauman

 Voting: practical tips
• Minimize irrelevant tokens first
• Choose a good grid / discretization
Too fine ? Too
coarse
• Vote for neighbors, also (smoothing in accumulator
array)
• Use direction of edge to reduce parameters by 1
• To read back which points voted for “winning” peaks,
keep tags on the votes.

ME5286 – Lecture 9 Kristen Grauman

 Hough transform: pros and cons
Pros
• All points are processed independently, so can cope with
occlusion, gaps
• Some robustness to noise: noise points unlikely to contribute
consistently to any single bin
• Can detect multiple instances of a model in a single pass
Cons
• Complexity of search time increases exponentially with the
number of model parameters
• Non-target shapes can produce spurious peaks in parameter
space
• Quantization: can be tricky to pick a good grid size
ME5286 – Lecture 9 Kristen Grauman
 Generalized Hough Transform
• What if we want to detect arbitrary shapes?
• Detect any arbitrary shape
– Requires specification of the exact shape of the object

ME5286 – Lecture 9
 Generalized Hough Transform
• Define a model shape by its boundary points and a reference
point.

Offline procedure:
x
a At each boundary point,
compute displacement
θ
p1
θ
p2 vector: r = a – pi.

Model shape Store these vectors in a

table indexed by gradient
θ … orientation θ.
θ …
…

ME5286 – Lecture 9
[Dana H. Ballard, Generalizing the Hough Transform to Detect Arbitrary Shapes, 1980]
 Generalized Hough Transform

R-Table

ME5286 – Lecture 9
 Generalized Hough Transform
Detection procedure:
x
For each edge point:
θ xx
• Use its gradient orientation θ to xx
index into stored table θ θ
p1
θ
• Use retrieved r vectors to vote θ

for reference point

Novel image
θ …
θ …
…

Assuming translation is the only transformation here, i.e., orientation and scale are fixed.

ME5286 – Lecture 9
 Generalized Hough Transform

ME5286 – Lecture 9
 Generalized Hough Transform

ME5286 – Lecture 9
 Rotation and Scale Solution

ME5286 – Lecture 9
 Rotation and Scale Solution

ME5286 – Lecture 9
 #47

Segmentation of Objects
Using Thresholding Method

ME5286 – Lecture 9
 Thresholding based Segmentation
#48
• Goal is to identify an object based on uniform
intensity
• Use the Histogram to compute the best threshold that
can separate the object intensity

ME5286 – Lecture 9
 Thresholding Methods
#49

ME5286 – Lecture 9
 Thresholding Principles
#50

ME5286 – Lecture 9
 Thresholding Example
#51

ME5286 – Lecture 9
 Thresholding Examples
#52

ME5286 – Lecture 9
 Histogram Calculation
#53

ME5286 – Lecture 9
 Histogram Profiles
#54

ME5286 – Lecture 9
 Good and Bad Histograms
#55

ME5286 – Lecture 9
 Maximum Separation
#56

ME5286 – Lecture 9
 Adaptive Thresholding
#57

ME5286 – Lecture 9
 Two Types of Variance
#58

ME5286 – Lecture 9
 Threshold selection via optimization
#59

ME5286 – Lecture 9
 Recursive Procedure
#60

ME5286 – Lecture 9
 Properties
#61

ME5286 – Lecture 9
 Gaussian Mixture Modeling of Histograms
#62

ME5286 – Lecture 9
 Fitting Model Distribution
#63

ME5286 – Lecture 9
 Fitting Model Distribution - 2
#64

ME5286 – Lecture 9
 Derivation of Optimal Threshold
#65

ME5286 – Lecture 9
 Derivation of Optimal Threshold - 2
#66

ME5286 – Lecture 9
 Cases for Optimal Threshold
#67

ME5286 – Lecture 9
 Algorithm for Gaussian Threshold Detection
#68

ME5286 – Lecture 9
 Properties of Gaussian Mixture Approach
#69

ME5286 – Lecture 9
 Examples
#70

ME5286 – Lecture 9
 Otsu vs Gaussian Approach
#71

ME5286 – Lecture 9
 Gaussian Gives Poor Results
#72

ME5286 – Lecture 9
 Gaussian Mixture – a Fail Case
#73

ME5286 – Lecture 9
 Issues with Thresholding
#74
• Histogram based thresholding is very effective
• Even with low noise, if one class is much smaller
than the other we might still be in trouble.
• Remember also that both these images have the same
histogram:

ME5286 – Lecture 9