Am 18.02.2016 um 23:55 schrieb Alain:
> Is it possible that using antialiasing method 2 could reduce that kind
> of artefacts?
> With method 1, you sample the center of each pixel, and compare that
> with the neibors. With method 2, you start with samples at each corners
> that are then compared.

That may well be.

Note that even anti-aliasing mode 1 already does some extra work to try
to prevent such aliasing artifacts, by silently rendering the pixels
immediately to the top and left of each block again to check if
oversampling is necessary. This way, the same comparisons normally done
for pixels in the bulk of the block can also be applied to the pixels at
the boundary. But in the bulk the neighboring pixels would also have
been subject to potential oversampling already, while the extra pixels
at the boundary will never be oversampled. This /can/ lead to aliasing
artefacts at the boundaries that wouldn't be there in the bulk.

In mode 2, this difference can't happen because the oversampling
decision is based on original trace results, not on oversampled pixels.

Post a reply to this message