In class a student suggested using a box filter to smooth the edges of the triangles; I think it's interesting that the supersampling approach from this slide seems to generate the same signal that a box filter would've done.

jerrypiglet

Could this approach be explained as: Supersampling at a far denser interval than the pixel density to preserve as many high frequency component in the frequency domain as possible, and at the same time to smooth the edge a little with reconstruction. And then perform resample at pixel density. In this way pink pixels can reduce zigzag along the edge of the red color block.

kayvonf

@jerrypiglet. You're on the right track. A triangle, by its definition, has edges, so as @lucida points out on slide 47 the coverage signal for a triangle will have discontinuities (which you can think of as infinite frequency content).

If you look at the next couple of slides, you'll see that we need to "resample" the sampled signal so that there is one sample per display pixel. This means we'll reconstruct an approximation to the original signal, and then sample the reconstructed signal at the desired rate. It's the reconstruction where the smoothing happens. Can you tell me why?

In class a student suggested using a box filter to smooth the edges of the triangles; I think it's interesting that the supersampling approach from this slide seems to generate the same signal that a box filter would've done.

Could this approach be explained as: Supersampling at a far denser interval than the pixel density to preserve as many high frequency component in the frequency domain as possible, and at the same time to smooth the edge a little with reconstruction. And then perform resample at pixel density. In this way pink pixels can reduce zigzag along the edge of the red color block.

@jerrypiglet. You're on the right track. A triangle, by its definition, has edges, so as @lucida points out on slide 47 the coverage signal for a triangle will have discontinuities (which you can think of as infinite frequency content).

If you look at the next couple of slides, you'll see that we need to "resample" the sampled signal so that there is one sample per display pixel. This means we'll reconstruct an approximation to the original signal, and then sample the reconstructed signal at the desired rate. It's the reconstruction where the smoothing happens. Can you tell me why?