[tgb]

I'm not convinced that is necessary. Here's my first thought. If I'm given a higher resolution image and am to downscale, knowing it will be displayed on a particular screen, then I would sample the blue pixel not from the "center of the pixel" but from the position of the blue dot in the screen. Similarly for red and green pixels. Couldn't an approach like that give subpixel precision?

[eridius]

If you're guaranteed that the image is displayed properly aligned to pixel boundaries, then doing that might do something, but it's probably not worth it.

Which it probably will be, but it's certainly not guaranteed.