Week4 SWIFT

Question 1

Features are (partially) invariant to what?

Answer 1

Image scaling and rotation

Partially: change in illumination and 3D camera viewpoint

Question 2

Why is high distinctive good?

Answer 2

Allows for single feature to be correctly matched in large database. Basis for object and scene recognition.

Question 3

What are the major stages of computation in SIFT?

Answer 3

1. Scale-space extrema detection
2. keypoint localization
3. orientation assignment
4. keypoint descriptor

Question 4

Scale-space extrema detection

Answer 4

Searches all scales and image locations and identify points of interest that are invariant ot scale and orientation.

Question 5

Keypoint localization

Answer 5

At each candidate location a model is fit to determine location and scale. Keypoints are selected by stability.

Question 6

Orientation assignment

Answer 6

One or more orientations assigned to each keypoint based on local image gradient directions.

Question 7

Keypoint descriptor

Answer 7

Local image gradients measured at scale in the region around each keypoint. These are transformed into a representation that can handle shape distortion and change in illumination.

Question 8

Scale space

Answer 8

Focus verleggen binnen beeld. Bijv. focus op raam of op huis / boom. Met gaussian smoothing konden objecten netjes verborgen worden waar niet op gefocust werd.

Question 9

Octave

Answer 9

Verdubbeling van schaal. Belangrijk om te kijken in hoe snelle stappen dat moet gaan.

We must produce s + 3 images in the stack of blurred images for each octave, so that final
extrema detection covers a complete octave.

Question 10

the relationship between D and σ^2∇^2G

Answer 10

Heat diffusion equation.

Question 11

ratio eigenvalues

Answer 11

-

Question 12

trilinear interpolation

Answer 12

-

Question 13

affine changes in illumination

Answer 13

-

Question 14

Scale space maxima

Answer 14

Maxima of scale normalizef Gaussian derivatives at different scales.

Question 15

LoG DoG extremen

Answer 15

Laplacian of Gauss (convolutions)

Lowe gebruikt Difference of Gauss

Question 16

Least Squares werkt slecht bij outliers, oplossing?

Answer 16

Estimation methods that are robust to outliers

Question 17

RANSAC

Answer 17

Random Sample Consensus

• classify points as inliners or outliners
• fit model to inliners disregarding outliners

Question 18

SIFT stappenplan Rein

Answer 18

1. Keypoint in f en f' 
sift(x, y, theta, s, d)
Waarbij:
x,y = position
theta = orientatie
s = scale
d = discriptor 

2. Matching van descriptor

False match: punten die wel op elkaar lijken maar niet in een overlappend beeld zitten.