Pixelation by oversampling.
This work here is striking of realism on two counts:
1. simplicity
2. trustworthy
Considering that cameras aren't calibrated, unless of very high pro quality, whether pixels aren't true originally and carried as such in Fourier or not true from the diagonal interpolator, none of the method will correct the camera inexact captured data. From there on and from previous demonstrations that "oversample2D" is unsatisfactory, this other avenue is exemplified. If the image would be rectangular, the rectangular spline would be used instead of the square one here. The other great method is pixelation by zoom/kernel/convolve/center. This method suffers lack of visibility from the unknown zoom. Maths are like real life: two twin branches of the tree don't carry twin leaves and unlikely the same counts. However, either branch will produce undistinguishable cabinets.
The other point in this thread is about the Mathcad Improc not introducing the oversampling as applied to pixelation. In fact it's not needed on the built-in zoom(,,,), but it still raises curiosity for the developers and advanced users.
jmG
