CSE 185

Introduction to Computer Vision

Fitting and Alignment
Fitting and alignment
• RANSAC
• Transformation
• Iterative closest point

• Reading: Chapter 4
Correspondence and alignment
• Correspondence: matching points, patches,
edges, or regions across images

≈
Fitting and alignment: Methods
• Global optimization / Search for parameters
– Least squares fit
– Robust least squares
– Iterative closest point (ICP)
• Hypothesize and test
– Hough transform
– RANSAC
Hough transform
Good
• Robust to outliers: each point votes separately
• Fairly efficient (much faster than trying all sets of parameters)
• Provides multiple good fits
Bad
• Some sensitivity to noise
• Bin size trades off between noise tolerance, precision, and
speed/memory
– Can be hard to find sweet spot
• Not suitable for more than a few parameters
– grid size grows exponentially
Common applications
• Line fitting (also circles, ellipses, etc.)
• Object instance recognition (parameters are affine transform)
• Object category recognition (parameters are position/scale)
 RANSAC
RANdom SAmple Consensus:
Learning technique to estimate
parameters of a model by random
sampling of observed data

Fischler & Bolles in ‘81.


 RANSAC

Algorithm:
1. Sample (randomly) the number of points required to fit the model
2. Solve for model parameters using samples
3. Score by the fraction of inliers within a preset threshold of the model

Repeat 1-3 until the best model is found with high confidence
 RANSAC
Line fitting example

Algorithm:
1. Sample (randomly) the number of points required to fit the model (#=2)
2. Solve for model parameters using samples
3. Score by the fraction of inliers within a preset threshold of the model

Repeat 1-3 until the best model is found with high confidence
 RANSAC
Line fitting example

Algorithm:
1. Sample (randomly) the number of points required to fit the model (#=2)
2. Solve for model parameters using samples
3. Score by the fraction of inliers within a preset threshold of the model

Repeat 1-3 until the best model is found with high confidence
 RANSAC
Line fitting example

NI  6 

Algorithm:
1. Sample (randomly) the number of points required to fit the model (#=2)
2. Solve for model parameters using samples
3. Score by the fraction of inliers within a preset threshold of the model

Repeat 1-3 until the best model is found with high confidence
 RANSAC


Algorithm: N I  14
1. Sample (randomly) the number of points required to fit the model (#=2)
2. Solve for model parameters using samples
3. Score by the fraction of inliers within a preset threshold of the model

Repeat 1-3 until the best model is found with high confidence
How to choose parameters?
• Number of samples N
– Choose N so that, with probability p, at least one random sample is free
from outliers (e.g. p=0.99) (outlier ratio: e )
• Number of sampled points s
– Minimum number needed to fit the model
• Distance threshold 
– Choose  so that a good point with noise is likely (e.g., prob=0.95) within threshold
– Zero-mean Gaussian noise with std. dev. σ: t2=3.84σ2
proportion of outliers e


N  log1  p  / log 1  1  e 
s
 s
2
5%
2
10%
3
20%
5
25%
6
30%
7
40%
11
50%
17
3 3 4 7 9 11 19 35
4 3 5 9 13 17 34 72
5 4 6 12 17 26 57 146
6 4 7 16 24 37 97 293
7 4 8 20 33 54 163 588
8 5 9 26 44 78 272 1177
RANSAC
Good
• Robust to outliers
• Applicable for larger number of objective function parameters
than Hough transform
• Optimization parameters are easier to choose than Hough
transform
Bad
• Computational time grows quickly with fraction of outliers
and number of parameters
• Not good for getting multiple fits
Common applications
• Computing a homography (e.g., image stitching)
• Estimating fundamental matrix (relating two views)
How do we fit the best alignment?
Alignment
• Alignment: find parameters of model that
maps one set of points to another
• Typically want to solve for a global
transformation that accounts for *most* true
correspondences
• Difficulties
– Noise (typically 1-3 pixels)
– Outliers (often 50%)
– Many-to-one matches or multiple objects
Parametric (global) warping

p = (x,y) p’ = (x’,y’)
Transformation T is a coordinate-changing machine:
p’ = T(p)

What does it mean that T is global?

– Is the same for any point p
– can be described by just a few numbers (parameters)

For linear transformations, we can represent T as a matrix

p’ = Tp
 x'  x
 y '  T  y 
   
Common transformations

original

Transformed

translation rotation aspect

affine perspective
Scaling
• Scaling a coordinate means multiplying each of its components by a
scalar
• Uniform scaling means this scalar is the same for all components:

2
Scaling
• Non-uniform scaling: different scalars per component:

X  2,
Y  0.5
Scaling
• Scaling operation:
x '  ax
y '  by
• Or, in matrix form:
 x '   a 0  x 
 y '   0 b   y 
    
scaling matrix S
2D rotation
(x’, y’)

(x, y)

x’ = x cos() - y sin()
 y’ = x sin() + y cos()
2D rotation
Polar coordinates…
x = r cos ()
y = r sin ()
x’ = r cos ( + )
(x’, y’) y’ = r sin ( + )

Trig Identity…
(x, y) x’ = r cos() cos() – r sin() sin()
y’ = r sin() cos() + r cos() sin()

  Substitute…
x’ = x cos() - y sin()
y’ = x sin() + y cos()
2D rotation
This is easy to capture in matrix form:

 x '  cos   sin     x 

 y '   sin   cos    y 
  

R
Even though sin() and cos() are nonlinear functions of ,
– x’ is a linear combination of x and y
– y’ is a linear combination of x and y

What is the inverse transformation?

– Rotation by –
– For rotation matrices R 1  R T
Basic 2D transformations
 x'   s x 0  x  x'   1  x   x 
 y '   0 s y   y   y '    
      y 1  y
Scale Shear

 x
 x'  cos   sin   x   x  1 0 t x   
 y '   sin  cos    y   y   0 1 t   y 
        y
 1 
Rotate
Translate
 x
 x   a b c  
 y   d e 
f 
y Affine is any combination of
  
 1  translation, scale, rotation, shear
Affine
 Affine transformation

Affine transformations are combinations of  x   x

a b c  
• Linear transformations, and  y   d e 
f 
y
  
 1 
• Translations
Properties of affine transformations:
or
• Lines map to lines
• Parallel lines remain parallel  x'   a b c  x 
• Ratios are preserved  y '   d e f  y
    
• Closed under composition  1   0 0 1   1 
 Projective transformations
Projective transformations are combos of  x'   a b c  x 
• Affine transformations, and  y '  d e f  y 
• Projective warps  w'  g h i   w
Properties of projective transformations:
  
• Lines map to lines
• Parallel lines do not necessarily remain
parallel
• Ratios are preserved
• Closed under composition
• Models change of basis
• Projective matrix is defined up to a scale
(8 DOF)
2D image transformations
Example: solving for translation

A1

A2 A3 B1

B2 B3

Given matched points in {A} and {B}, estimate the translation of the object
 xiB   xiA  t x 
 B    A   t 
 yi   yi   y 
Example: solving for translation

A1

A2 A3 (tx, ty) B1

B2 B3

Least squares solution

 xiB   xiA  t x  1 0  x1B  x1A 
 B 
1. Write down objective function
 B    A   t  0

1
 t x   y1  y1A 
2. Derived solution
a) Compute derivative
 yi   yi   y         
  ty  
b) Compute solution
1 0    xnB  x nA 
3. Computational solution  y nB
a) Write in form Ax=b
0 1  y nA 
b) Solve using pseudo-inverse or eigenvalue
decomposition
Example: solving for translation

A1
A5
B4
A2 A3 (tx, ty) B1

A4 B2 B3
B5

Problem: outliers
RANSAC solution
1. Sample a set of matching points (1 pair)  xiB   xiA  t x 
2. Solve for transformation parameters  B    A   t 
3. Score parameters with number of inliers  yi   yi   y 
4. Repeat steps 1-3 N times
Example: solving for translation
B4
A1 B5 B6

A2 A3 (tx, ty) B1

A4 B2 B3
A5 A6

Problem: outliers, multiple objects, and/or many-to-one matches

Hough transform solution
1. Initialize a grid of parameter values  xiB   xiA  t x 
2. Each matched pair casts a vote for consistent  B    A   t 
values  yi   yi   y 
3. Find the parameters with the most votes
4. Solve using least squares with inliers
Example: solving for translation

(tx, ty)

Problem: no initial guesses for correspondence

 xiB   xiA  t x 
 B    A   t 
 yi   yi   y 
When no prior matched pairs exist
• Hough transform and RANSAC not applicable
• Important applications

Medical imaging: match Robotics: match point clouds

brain scans or contours
Iterative Closest Point (ICP)
Goal: estimate transform between two dense sets of
points

1. Initialize transformation (e.g., compute difference in means and

scale)
2. Assign each point in {Set 1} to its nearest neighbor in {Set 2}
3. Estimate transformation parameters
– e.g., least squares or robust least squares
4. Transform the points in {Set 1} using estimated parameters
5. Repeat steps 2-4 until change is very small

https://www.youtube.com/watch?v=m64E47uvPYc
https://www.youtube.com/watch?v=uzOCS_gdZuM
Example: aligning boundaries

p q
Example: solving for translation

(tx, ty)

Problem: no initial guesses for correspondence

ICP solution
 xiB   xiA  t x 
1. Find nearest neighbors for each point
 B    A   t 
2. Compute transform using matches  yi   yi   y 
3. Move points using transform
4. Repeat steps 1-3 until convergence
Algorithm summary
• Least Squares Fit
– closed form solution
– robust to noise
– not robust to outliers
• Robust Least Squares
– improves robustness to noise
– requires iterative optimization
• Hough transform
– robust to noise and outliers
– can fit multiple models
– only works for a few parameters (1-4 typically)
• RANSAC
– robust to noise and outliers
– works with a moderate number of parameters (e.g, 1-8)
• Iterative Closest Point (ICP)
– For local alignment only: does not require initial correspondences
Object instance recognition
B3
A1
A3
1. Match keypoints to A2
B2
object model Matched B1
keypoints
2. Solve for affine
transformation Affine
parameters Parameters

3. Score by inliers and

# Inliers
choose solutions with
score above threshold Choose hypothesis with max
score above threshold
Keypoint matching 1. Find a set of
distinctive key-
points
A1 B3 2. Define a region
around each
keypoint
A2 A3 3. Extract and
B2 normalize the
B1 region content
4. Compute a local
fA fB descriptor from the
normalized region
5. Match local
descriptors
d ( f A, fB )  T
Finding the objects

Input
Image Stored
Image

1. Match interest points from input image to database image

2. Matched points vote for rough position/orientation/scale of object
3. Find position/orientation/scales that have at least three votes
4. Compute affine registration and matches using iterative least squares
with outlier check
5. Report object if there are at least T matched points
Object recognition using SIFT descriptors

1. Match interest points from input image to database image

2. Get location/scale/orientation using Hough voting
– In training, each point has known position/scale/orientation wrt whole object
– Matched points vote for the position, scale, and orientation of the entire
object
– Bins for x, y, scale, orientation
• Wide bins (0.25 object length in position, 2x scale, 30 degrees orientation)
• Vote for two closest bin centers in each direction (16 votes total)
3. Geometric verification
– For each bin with at least 3 keypoints
– Iterate between least squares fit and checking for inliers and outliers
4. Report object if > T inliers (T is typically 3, can be computed to match some
probabilistic threshold)
Examples of recognized objects