[CSDb] - User Forums - Anti-Aliasing

You are not logged in - nap

CSDb User Forums

Forums > C64 Pixeling > Anti-Aliasing

2006-05-17 16:39

Tch
Account closed

Registered: Sep 2004
Posts: 512

Anti-Aliasing

I thought I understood but I am confused again.

What the hell is Anti-aliasing?

... 44 posts hidden. Click here to view all posts....

2006-10-05 14:58

Oswald

Registered: Apr 2002
Posts: 5022

mace,

I am sorry.

2006-10-30 23:24

Death Demon
Account closed

Registered: Feb 2005
Posts: 68

If you're going to manually anti-alias, you should follow the same algorithmic approach used by computational anti-aliasing. The entire purpose of anti-aliasing is to reduce the visual impact of aliasing (which has already been described above). When doing computational anti-aliasing, we cut a few corners because rendering speed is impacted. However, when doing it manually, you're just creating a static image, so who cares. In any event, anti-aliasing is performed by taking "samples" around the edges of the image you want to anti-alias. Look at the leaf example above. Those "samples" are then blended with each other to create the actual image. So let's take an 8-bit set of pixels that looks like:

000 200 200 250 250 000
000 200 220 230 250 000
000 200 210 220 200 000
000 100 100 210 000 000
000 000 000 000 000 000

That's basically some sort of semi-circle. Now, you want to "smooth" out the stuff on the bottom. The proper way to do this is to pick what are called "sub-samples" for each pixel (the pixel itself is called a "sample"). The sub-samples are the pixels around the pixel itself. So, for the pixel corresponding to coordinate (x,y)=(2,4), you would see a value of 100. If you take samples around that pixel, say four samples in a diagonal grid, you would take samples of (1,3), (3,3), (1,5), and (3,5). Those samples will then be multiplied by some weighting factor, say 25% for each. Then they are added together and that value is your new pixel for coordinate (2,4). That means you replaced the 100 completely without using any part of it for a subsample. Now, you can come up with more interesting grids which include more samples and even include the pixel itself. You can also mess with weighting factors. It's really just dependent on what your image consists of, the angles of the edge off ideal horizontal or vertical, the contrast of the background you're effectively blending with, etc.

2006-10-30 23:43

jailbird

Registered: Dec 2001
Posts: 1576

Ok, now I must quote Tch from the first post in this thread.

I thought I understood but I am confused again.

:)

2006-10-31 00:27

Helm
Account closed

Registered: Jul 2006
Posts: 25

"If you're going to manually anti-alias, you should follow the same algorithmic approach used by computational anti-aliasing."

Short reply: Not always, no.

Complicated reply: automatic anti-alias and manual anti-alias share a characteristic: they're both about smoothing jaggies on small-resolution screens. They also share a difference: one is automatic, one is manual.

When you do something manually, I propose it to be a good idea to not try to approach the procedure as would a machine. Let's say I'm antialiasing a stick. one end is deep inside a hedge, and the other is high in the middle of the sky. One end, inside the hedge, I'd antialias more, to blur the detail inside the conflicting masses of color that make up my representation of leaves, shadows and whatnot. The other end, against the bright contrast of the sky, I might choose to antialias less (1point, even less) so I can retain the sharpened contour that calls the eye upon it for inspection. I'd even change the basic form around to not have any small-angle curves that cause the most alias before I'd add 4 points of AA around it.

Pleasant lack of jaggies is one thing. Clarity and juxtaposition is another. A manual approach befits those that consider it a boon to have the choice to sacrifice one in favour of the other on a case-by-case basis inside a single piece of art.

2006-10-31 06:20

Style

Registered: Jun 2004
Posts: 498

wow, there's some interesting interpretations of 'antialiasing' floating around :)

Here's what it is in a nutshell - imagine you draw a black circle using a sharp pencil and a compass. Now imagine trying to represent that real-world circle in a bitmap, that is, a grid of discrete elements. You end up with 'jaggies'.

Anti-aliasing means quantifying how much of a 'discrete element' (or cell, or pixel) would be covered by our real-world circle, and then choosing an appropriate colour for the pixel based on how much of it would be covered.

For example, if 50% of a pixel would be covered by our (black) real-world circle, then we might give it a 50% grey colour. 5% of a pixel being covered would be very nearly white. 95% of a pixel being covered would be very nearly black.

Thats pretty much it. Its a technique used to give the best reproduction of a real world shape on an imperfect media.

2006-10-31 15:06

ptoing

Registered: Sep 2005
Posts: 271

Yes, Style that is correct of course, but when doing aa manually you have a choice of where to apply how much aa and also well done manmade aa is always crisper than what a machine can come up with from what I have seen so far, esp when you are doing demosceney lowcolour stuff, like c64 where you have a set palette or Atari ST stuff where you have 16 out of 512 colours to choose from.

2006-11-15 22:23

Death Demon
Account closed

Registered: Feb 2005
Posts: 68

Helm : So the algorithm you're talking about there is contrast weighted sub-sampling. And it looks like garbage. The higher the contrast, the more obvious the aliasing. The reason is that there's a much larger frequency differential between both images. Aliasing is worsened the further outside of the Nyquist rate that you sample. The Nyquist rate is twice the frequency you are sampling. When sampling high frequency, you have to do higher frequency samples, ie. more samples per pixel. Algorithmically can do this perfectly. In computational hardware where framerate is important, you can't do this because it takes too much time to calculate the frequency. But when you're doing offline rendering in software, it's not a big deal.

Manually, you have the same capabilities. But I don't think there's any manual method for anti-aliasing that can't be covered by algorithms. If there is, and you write a paper, you're going to receive a lot of money because I'm not aware of any.

2006-11-15 22:48

jailbird

Registered: Dec 2001
Posts: 1576

Yeah that's very nice, but what all this has to do anti-aliasing on Commodore 64? I'd like to see that algorithm which emulates e.g. Electric's or miss Mermaid's anti-aliasing, I could even write a paper about that. Now where can I apply for the lot of money?

2006-11-16 01:14

ptoing

Registered: Sep 2005
Posts: 271

The keyword here is aethetic choice. No algorithm can determine to antialias so that it looks the most pleasing regarding to all parts of the picture, especially not in pixelart.

2006-11-16 02:33

Helm
Account closed

Registered: Jul 2006
Posts: 25

"Helm : So the algorithm you're talking about there is contrast weighted sub-sampling."

No... not at all.

Previous - 1 | 2 | 3 | 4 | 5 | 6 - Next

Refresh

Subscribe to this thread:

You need to be logged in to post in the forum.

Search the forum:
Search for in
All times are CET.

Search CSDb

Advanced

Users Online

A3/AFL
csabanw
Ghost/Quantum
Alakran_64
Sentinel/Excess/TREX
Airwolf/F4CG
Kakka/Extend, Damone..
Guests online: 136

Top Demos

1 Next Level  (9.8)
2 Mojo  (9.7)
3 Coma Light 13  (9.7)
4 Edge of Disgrace  (9.6)
5 Comaland 100%  (9.6)
6 No Bounds  (9.6)
7 Uncensored  (9.6)
8 Wonderland XIV  (9.6)
9 Memento Mori  (9.6)
10 Bromance  (9.5)

Top onefile Demos

1 It's More Fun to Com..  (9.7)
2 Party Elk 2  (9.7)
3 Cubic Dream  (9.6)
4 Copper Booze  (9.5)
5 TRSAC, Gabber & Pebe..  (9.5)
6 Rainbow Connection  (9.5)
7 Dawnfall V1.1  (9.5)
8 Quadrants  (9.5)
9 Daah, Those Acid Pil..  (9.5)
10 Birth of a Flower  (9.5)

Top Groups

1 Nostalgia  (9.3)
2 Oxyron  (9.3)
3 Booze Design  (9.3)
4 Censor Design  (9.3)
5 Crest  (9.3)

Top NTSC-Fixers

1 Pudwerx  (10)
2 Booze  (9.7)
3 Stormbringer  (9.7)
4 Fungus  (9.6)
5 Grim Reaper  (9.3)

Page generated in: 0.05 sec.