Week 3

Question 1

What are features?

Answer 1

Features are local invariant descriptors
They are local, meaningful and detectable parts of the image
It is the location of sudden change

Question 2

Why use local features?

Answer 2

1) Information Content is High
2) Invariant to change of view point, illumination
3) They reduce computational burden

Question 3

What are the local features requirements?

Answer 3

1) Repeatable -
a) They should be invariant to rotation, translation and scale changes
b) Robust and covariant to out-of-plane affine transformations
c) Robust to changing lighting variations, blur, noise and quantization
2) Locality - The features are local and therefore robust to occlusion and clutter
3) Quantity - There must be enough features to cover the whole object
4) Distinctiveness - The local features must contain part of the image which has interesting structure
5) Efficiency - close to real-time performance