What to do next
===============

* Add support for more images (make the center determination more precise).

* Use a better method for problem solution (employ ``sympy`` and work with
  Jacobian and Hessian).

* Think of better distribution of points we use in the distortion calculation.

* Make a decent regression-test case (i.e. generate image composed of points
  that is supposed to be distorted and the image we suppose being OK.
  Then, ensure that the imagemagick's convert will produce results that are
  consistent with ours.

* Think: Maybe use something else than quadratic polynomial?
* Use a different r -> r model.