END OF AN ERA, FRACTALFORUMS.COM IS CONTINUED ON FRACTALFORUMS.ORG

um, what is my viewpoint on antialiasing? and erik reckase's on images?


good antialiasing has everything to do with sharpness and edges, really: http://graphics.stanford.edu/~mmp/chapters/pbrt_chapter7.pdf

in fact it's probably because almost no one on earth takes proper care wrt antialiasing that you associate it with blurring. it really is both a science and an art, not to be mocked...


there are few things i dislike more than blurred images. antialiasing is first and foremost a matter of accuracy and precision, and blurring is totally out of the question.

the fundamental distinction is this: you can bandlimit a signal, performing proper reconstruction up to the nyquist limit, without blurring the image needlessly.

that's exactly it; aliasing is just a fancy word (in the context of spectral/fourier analysis) for jaggies


the "gaussian hump" (kinda lewd name for a filter kernel if you ask me) is actually one of the better filters to use for fractals, because:

1. filters having negative lobes are unsuited to fractal boundaries <insert long fourier analysis here>
2. the gaussian filter is radially symmetric (cf. separable like most other filters) and much smoother than the box or tent filters, making it well suited to combat the (infinitely) high frequency variation at fractal boundaries.

i can almost hear people in the audience unfamiliar with signal processing thinking "but gaussian filters blur". this is where one has to fight the fame of photoshop: yes, it is a blurring filter in photoshop, but that's not how it's supposed to be used here! as a photoshop filter, you take a very wide gaussian (say 10+ pixels in diameter) and convolve all the surrounding pixels with it. as an antialiasing filter, you take a very narrow gaussian (say sqrt(2) pixels in diameter) and convolve the underlying continuous function with it, so as to produce a bandlimited sample which will not alias.

i really cannot overstate the difference between the two approaches; one is a process which destroys information, and the other is one which greatly increases the information content. aliasing is not only ugly, it's information massacre!

<rant>
life becomes more complicated once you've studied information theory and signal processing. i nearly got thrown out of the cinema while watching "enemy of the state" (quite a few years back as a hormonal teenager, mind ) - in the scene where they take a photo, zoom in 1000x and spin it around in 3d, i just started yelling "booooollocks!" etc. i know it wasn't appropriate, but sometimes i just can't help myself; fortunately i got it under control
</rant>

resampling the image didn't cause any blurring, it caused antialiasing... oh well, i give up.

i'm not angry at all, just rolling-eyes-disappointed perhaps   just supersampling instead of increasing the resolution will do much to reduce your benchmark's i/o requirements, for example.

i know a decent bit about parallel programming, having writing a distributed ray tracer in high school to run on a cluster of amd machines in my computer lab at the time, and then all the way up to an mpi-based cluster at a physics department's supercomputing lab. i suspect you may have conducted your research on such a system; i'm actually a ray tracing guy, not a fractal guy, and parallel computing comes naturally to ray tracers (it's a so-called "embarassingly parallel" algorithm, in your terminology).

please don't get the impression that all people posting on forums about fractals know only about fractals many of us post on several forums, and even do stuff for a living