Feature matching and Hough

Question 1

What is Feature matching used for?

Answer 1

Generally for finding shapes in images.

Question 2

What are important measures of how good feature matching is?

Answer 2

• Goodness of fit

- Optimized

Question 3

What are the different methods used in feature matching?

Answer 3

• Least squares fit
• Generalized hough transform
• RANSAC

Question 4

Briefly explain Hough transform.

Answer 4

Want to find a line/curve that best represents a set of datapoints.
Steps:
- Create a grid of parameter values
- Each point votes for a set of values, incrementing them
- Find max or local max in grid

Question 5

What are the additions to Hough that makes it perform better?

Answer 5

• Use polar coordinates for parameter space

- Incorporate gradients, since lines are uniquely determined

Question 6

What are pros/cons of Hough?

Answer 6

Pros: Robust to outliers, efficient, multiple good fits
Cons: sensitive to noise, hard to determine bin size

Question 7

How can you generalize Hough?

Answer 7

You need to train the program. Can be done by using boundary and reference points

Question 8

Explain how RANSAC works.

Answer 8

Kinda similar to SVM.

• Sample the number of points to fit a model
• Solve for paramtere
• Score by fraction of inliers within a threshold

Question 9

What are pros/cons of RANSAC?

Answer 9

Pros: Robust to outliers, easy to choose parameters
Cons: Not efficient, not multiple fits