Transformations and Image Alignment

Question 1

What’s the difference between image filtering and image warping?

Answer 1

Image filtering: changing the range of the image
Image warping: changing the domain of the image

Question 2

Parametric/Global Warping

Answer 2

Translation, rotation or aspect ratio change.
Global warping is done by taking a function that takes each pixel from the original image as input and maps it to its location in the new image

Question 3

Affine transformation

Answer 3

Any transformation that occurs with a 3x3 matrix where the last 3 entries in the matrix are [0 0 1]

Question 4

What property of linear transformations changes when we use affine transfomation

Answer 4

The origin no longer always maps to the origin

Question 5

Homography

Answer 5

A 3x3 non-affine matrix i.e. the last row = [a b 1]

Question 6

What can we do with homographies that we can’t do with affine transformation

Answer 6

Projective warps - parallel lines are no longer always parallel and ratios are not preserves

Question 7

Forward Warping

Answer 7

If a pixel lands in a non-integer location after a transformation, all neighboring pixels “contribute” to the value of the pixel. This can still result in holes

Question 8

Inverse Warping

Answer 8

Given a pixel location in I’, get the corresponding pixel location from T’(I’(x’, y’)) and use that color