Cell phone video three panels

It’s a combination of vertical and horizontal skews. First, you skew the image vertically, so a rectangle is transformed into a parallelogram. Then you skew it horizontally, by twice as much as the first skew. Then you skew it vertically again.

In principle, you could use a larger number of skews, each by a smaller amount, but from what I understand, just the three skews usually produces an acceptable result.

Thanks - it looks like this is what you’re describing, which is interesting, but has some pretty horrible jaggies in the end result. I suspect this could be mitigated a bit by some scaling before and after.

You’d never want to use that algorithm in practice. Even a simple bilinear filter would give better results and is absolutely trivial on any modern hardware. Even much more sophisticated filters are trivial with the amount of horsepower available.

Ideally, a rotation filter would take the Bayer arrangement into account. You gain more information if you know where the color subpixels are.