2D Colour Image Warping

2D image warping can be achieved by identifying homologous landmarks between two images, that is pairs of points which map from one image to the other. An RBF is fitted to the displacement field which relates the two sets of landmarks. A separate RBF is fitted to each component of the displacement field. At each pixel in the warped image the RBF is evaluated to determine its corresponding location in the original image. The value at this location determines the `warped' pixel's RGB value. The warped image can be evaluated over an arbitrary region, at any resolution specified by the user.

Features

  • No need to define where every point maps to
  • A function describes the warping
  • The warping (in general) is not reversible since `many to one' mappings are allowed.

Example - warping of a scanned photo

No warpingWarped image
(a) No warping (b) warped image
(The original photo was taken by Jonathan Carr on a climbing trip to Mt Hooker, South West New Zealand)