Can ClearType-ish techniques be applied to 3D AA?

[Hoenlere]

Just thinking out loud here...

Could the technique ClearType uses to anti-alias fonts in 2D be applied somehow to 3D anti-aliasing?

A friend of mine asked me this the other day, and I really didn't have an answer for him, so I thought I'd check here.

For those of you not familiar w/ ClearType, it relies upon knowing the exact layout of the pixels on an LCD display to "pseudo" increase the resolution of fonts and provide anti-aliasing.

I suspect that ClearType is really only effective on a single coloured background, but I am not informed enough to know for sure.

What do you guys think?

Thanks!

-Ryan

 

[Nick]

ClearType works on almost every LCD and for every application. It doesn't matter whether the elements are in RGB, GBR or BRG order, your eyes won't see the difference, unless you have an old LCD where the space between pixel color elements isn't equal to the space between adjecent pixels.

What I'm not sure about is if this would further increase quality if you're already using anti-aliasing (probably yes), and if it can easily be implemented in hardware (also probably yes). Unfortunately I think it's patented...

 

[Hoenlere]

Actually, In my experience it must know the sub pixel order in order to function properly.

You're right, it doesn't matter WHICH order they are in, but it needs to know... This is why Microsoft was forced to update the ClearType tuner to support BGR displays such as the Dell 1800FP (Speaking of which, if you have one of these and you haven't tuned it to BGR, you are going to be quite happy w/ the results

-Ryan

 

[CosmoKramer]

It gives a marked improvement on any old CRT as well.

 

[DemoCoder]

Wow, that new ClearType tuner fixed everything. I had experienced ClearType on my IPAQ, but whenever I used it on my flatplanel, the result was worse than standard font smoothing, the fonts looked DROP SHADOWED/gaussian blurred for example. I chalked it up to it working better on small displays like the IPAQ than 1600x1200 LCD, but when I used this new tuner to select BGR, the result was amazing!

Man, wish MS had put this into a Windows Update or something, I never knew it existed.

 

[Hoenlere]

Isn't it a HUGE difference? I can't believe you need SP1 to support it too!

They should really tell people this stuff

 

[Althornin]

Is the tuner you are refering to the online one?

 

[Hoenlere]

Yup.

http://www.microsoft.com/typography/cleartype/tuner/1.htm?fname= &fsize=

 

[pcchen]

ClearType is supported on pre-SP1 WinXP. However, pre-SP1 does not support BGR configuration (only the more common RGB configuration).

I read a paper about ClearType from Microsoft few months ago. It does mention about using such techniques for general anti-aliasing. However, it concludes that this technique work best for high contrast objects (such as text). I also did some simple experiments and I can't see significant advantage of using such technique over general anti-aliasing on general color images.

 

[Colourless]

Actually such an idea might work. However, tests would need to be done to check it's effectiveness. In a sense it would be a primitve version of colour compression.

Basically, think of 3x1 multisampling. However instead of writing all 3 colour components of every subsample to the framebuffer, you just write one. However, it might cause some really odd looking edges if it was all that was done. If it was done at the same time as say normal 4x multisampling it might actually be quite nice. You in effect have 12x multisampling. Memory requirements for the Z buffer might be a killer here though.

Trying to do with with supersampling would likely be unwise. Chances are it would cause some odd bluring to occur.

The idea however, does have an issue with Dest Alpha. There would only be 1 Dest Alpha component for every 3 subsamples.

 

[nutball]

The major short-coming of ClearType is that it doesn't work for vertical edges (for obvious reasons).

 

[Babel-17]

I just gained a marked improvement on my system. Thank you very much. I posted a link to this thread over at slackercentral and it's helped at least one other user.

 

[Basic]

One place where ClearType could be nice is stereo 3D. Especially in HMDs, where high res displays are quite expensive. Since the depth information comes as small horizontal displacements, the increased horizontal resolution should come in handy. Combine it with regular AA on top of that to make it even better.

I've noticed that on at least some HMDs, the speced horizontal resolution is three times larger than expected (like 2400x600 instead of 800x600). Maybe this is because they expect the images to be rendered in ClearTye-style.

[Edit]
Hmm, it should be possible to simulate it by rendering to a three times wider texture, and then do a PS2.0 postfilter.

But thinking at it again, I agree with pcchen. It's got obvious problems with color images. (I'm not sure why it should be best for high contrast images though.) The problem is that objects with different colors could get different AA. White on black is optimal, but ie red on black wouldn't get any AA at all. More saturated colors (especially red/green/blue) means worse AA.

 

[Ailuros]

I just gained a marked improvement on my system. Thank you very much. I posted a link to this thread over at slackercentral and it's helped at least one other user.

That's two. I just forgot to mention it

 

[pcchen]

Speaking of color images... I was thinking about this: would it better to use an alternative pattern for LCDs which are mainly for displaying graphics (such as game consoles)?

The pattern should be like:

R G R G R G
G B G B G B
R G R G R G
G B G B G B

or something similar. Will such patterns display better?

 

[mboeller]

Speaking of color images... I was thinking about this: would it better to use an alternative pattern for LCDs which are mainly for displaying graphics (such as game consoles)?

The pattern should be like:

R G R G R G
G B G B G B
R G R G R G
G B G B G B

or something similar. Will such patterns display better?

You mean something like this?

http://www.clairvoyante.com/

 

[pcchen]

Hmm... I remembered I've read that before
However, I never see any real product utilizing such "alternative" patterns. Even on handheld game consoles or digital cameras.

 

[Simon F]

Hmm... I remembered I've read that before
However, I never see any real product utilizing such "alternative" patterns. Even on handheld game consoles or digital cameras.

I believe digital (still) cameras do it as a means of trying to do more with fewer samples by using more green sensors.

 

[mangrove]

I believe digital (still) cameras do it as a means of trying to do more with fewer samples by using more green sensors.

This cannot be applied to output technology however. The thing with digital cameras is that they don't take a full color reading per pixel; they only take R, G, or B. A LCD screen "pixel" consists of all three components; a 1600x1200 pixel screen contains three times as many color elements (another new short form: "cexels" ;-) while a 1600x1200 CCD in a digital camera contains only 1600x1200 "cexels".

A digital camera captures, essentialy, monochrome luminance values. In front of the sensor they have fitted a filter with a color pattern, often of the "Bayer pattern" type. This is
GRGRGR...
BGBGBG...
(etc)
That green information is more important is because of the eye's higher sensitivity to green resolution. Essentially, what the camera is doing is interpolating the missing color information from the neighbor pixels -- it only has access to one color component per pixel.

I cannot see any relevance for this technology in computer screens... then again, I don't see everything (too few forward-looking cexels)

 

[OpenGL guy]

A digital camera captures, essentialy, monochrome luminance values. In front of the sensor they have fitted a filter with a color pattern, often of the "Bayer pattern" type. This is
GRGRGR...
BGBGBG...
(etc)
That green information is more important is because of the eye's higher sensitivity to green resolution. Essentially, what the camera is doing is interpolating the missing color information from the neighbor pixels -- it only has access to one color component per pixel.

This isn't true of all digital cameras, but close. Check out the Foveon CCD used in the Sigma SD9. While the Sigma implementation seems lacking, future products based on Foveon technology should be interesting.