I’m working on writing some CUDA software to remap the pixels of an image to new floating point coordinate destinations
and then interpolating between them at a grid of integer values so as to retrieve a new image which is distorted in the
way specified by the new floating point coordinate destinations. (I think this is called ‘image warping’ ?)
The problem I have is to find a good way to start, I would normally always use texture memory to interpolate in 2D, but
texture memory requires the new coordinates to be integers and this will rarely be the case…
In summary, I would like to do the same as the matlab command ‘triscatteredinterp’, but on the GPU (as the matlab command
requires about 1.5 second to complete).
Thanks for any tips,