Antialiasing the Mandelbrot Set at sci.fractals

"The best way I found was to use a modified supersample and then use a 3x3 grid. Do this: Divide the pixel to render in 3x3 sub-pixels.
Iterate them all and save the results. After all nine sub-points are examined then compare the results and discharge all but the one that used the least number of iterations and write that one as the pixel on the screen. If you do this you do not need to anti-anilise. The drawback is that the method is slow as ... (nine times slower than normally), but the result is pretty nice.
....
(I also used a palette of 8192 colours and the same amount of iterations to create the images.)
Regards: SunCode "
Just curious: why do that rather than compute the average color of the subpixels? Edwin Young

"Just because this make the chance bigger two adjecting points will get nearby colours. This makes a sharper, (less blury) image than the average. Think about a Julia spiral that hat got spirals, spirals ... Then two nearby points in one of the smaller spirals can have a wery big diffrence in used iterations, (this makes a pixly image if super-sample is not used). The average will still make a pixly image - just less pixly. If one uses least number of iterations, then (maybe?) one of the points is not to deep into the spiral. 2x2 sampling is not enough but 3x3 is pretty good. 4x4 makes a slight better image still but if one consider the time it takes to calculate it - then it is not woth it (16 times slower than normally)"

SunCode.