The other method is to start from the target image, and then "work
backwards" and determine which pixel to "plop" from in the source image. You
might do "nearest-neighbour" approximation, where the same pixel from the
source might be used for two or more pixels in the target, or some sort of
weighed-average of the neighbouring pixels in the source image depending on
what your sin/cosine functions return as the source coordinates.

   Depends on how much processing power you have available to you, and how
patient your users are.

   - Oliver