Canny-Harris

Question 1

Criteria for a good edge detector

Answer 1

Low error rate
Edge points should be well localized
Only one response to a single edge

Question 2

Canny edge detector properties

Answer 2

Optimal kernel

Non-maximum suppression

Hysteresis thresholding

Question 3

Non maximum suppression

Answer 3

Check if pixel is local maximum along gradient direction, select a single max across width of the edge (keep checking if pixels next to current pixel is higher)

Question 4

Hysteresis thresholding

Answer 4

If the gradient of the pixel is:
High- strong edge pixel
Low- non edge pixel
In between-based in neighbor pixels

Question 5

Effect of theta on canny

Answer 5

Theta is Gaussian spread size

Large theta detects large scale edges

Small theta detects fine features

Question 6

Gaussian pyramid

Answer 6

Take original image, shrink and apply Gaussian filter

They are useful because they can help with feature tracking and search over scale. Also search for correspondence (coarse to fine matching)

Question 7

Interest point detection characteristics

Answer 7

Repeatability
Same interest points

Saliency
Each point is distinctive

Locality
A point occupies a small area of the image

Compactness and efficiency
Need a sufficient amount but fewer than image pixels

Question 8

Key points are used for

Answer 8

Image alignment
3D reconstruction
Motion tracking 
Robot navigation
Indexing and database retrieval

Question 9

Interpreting Harris corner detection matrix

Answer 9

M = a 0
0 b

If either a or b is close to 0, it is not a corner
If it is a corner they will be large