Lots of methods from sampling theorem are useful for graphics, but they don’t exactly apply:

Samples are infinitesimal points. Pixels in hardware aren’t.

You can’t have a real square wave in sampling. If you shift samples a bit to reveal Gibbs phenomenon, it doesn’t change the wave properties at all. OTOH you can have a hard edge on screen, and people will notice if it’s shifted by a half of a pixel.

If image aliasing behaved like aliasing of sampled signals, we could solve font rendering with a Fourier transform and noise shaping (the way audio resampling is a solved problem). Instead, we have a mess of font hinting hacks, because eyes and brains somehow didn’t get the memo about mathematical beauty of the sampling theorem.

I remember first learning about CG and reading about this - it caused me a lot of confusion, because that’s the theory but in practice no one does this. One of the most frequent ways is down size a double size image, and as @kornel says font hints

Lots of methods from sampling theorem are useful for graphics, but they don’t exactly apply:

If image aliasing behaved like aliasing of sampled signals, we could solve font rendering with a Fourier transform and noise shaping (the way audio resampling is a solved problem). Instead, we have a mess of font hinting hacks, because eyes and brains somehow didn’t get the memo about mathematical beauty of the sampling theorem.

I remember first learning about CG and reading about this - it caused me a lot of confusion, because that’s the theory but in practice no one does this. One of the most frequent ways is down size a double size image, and as @kornel says font hints