Sub pixel rendering

tsprowl's picture

2 years ago I read an article in Baseline about it, how its being used by font system manufacturer's and so on. we're talking ebooks, blackberry's, palm pilots, cell phones ect. suddenly bitmap fonts were all the rage...then SVG stuff sneaked out and I thought bitmap was going to be a thing of the past but no we're back to bitmap again.

so I'm wondering if there's any allowance/ concern/ design being considered to exploit all this sub-pixeling.

It would seem funny that techology is solving vector to grid problems for us, instead of the the other way around.

So the gist of it being that 1 pixel is actually split into 3 columns and important parts are given priority with %100 black while the other 2 may decrease in shade.

it seems awkward that the rendering of someone's design be left up to the manufacturer's when most likely we can create some pretty spiffy stuff for it specifically.

anyone out there with more info? cause I'd sure like to know how/ if this sub-pixelling is being used or affecting stuff



John Hudson's picture

Sub-pixel positioning is certainly being used (e.g. for pretty much every bit of text I look at on my monitor). Are there any type designers deliberately taking exploiting the technology in their designs to improve screen readability? Yes. However, the nature of the a rendering technologies is that they change and improve over time, so it is a dangerous to try to design very specifically for a single rendering technology. For example, the version of the MS ClearType renderer that I use on my XP system is different and better than the version that shipped with the first version of the MS Reader. I've seen versions of the ClearType renderer in development that introduce pretty radical improvements, so trying to address the current state of the technology in design isn't typically a good idea.

That said, if one really understands how the technology works -- i.e. understands the basic principles and the details of the colour filtering model used -- then it is possible to take advantage of the technology in design and to produce outlines and make hinting decisions that will produce optimal onscreen rendering results that will likely be adaptive across several generations of that particular technology. My guess is that there are probably half a dozen type designers in the world who have a good grasp of this stuff.

Another thing to bear in mind is that, while subpixel rendering can improve the screen quality of pretty much any typeface (with the caveat that some kinds of hinting for b/w bitmaps can cause problems for subpixel rendering), a typeface that is optimised to take advantage of such rendering might not be the best choice for documents that will be read onscreen and printed. If you optimise a design for subpixel rendering, you are making the decision that this is going to be a screen font, and other considerations are going to be ignored.

Yet another thing to bear in mind is that there are different subpixel rendering technologies. Compare, for example, Microsoft's ClearType to the Adobe CoolType rendering in Acrobat 5. Microsoft's is vastly superior for actual, immersive screen reading because they maintain the contrast. Adobe's subpixel rendering tends to grey the text, which is a mistake if your goal is screen readability (but might be okay if your goal is wysiwyg display of documents that you're still expecting someone to print).

I've mentioned hinting a couple of times. ClearType rendering greatly simplifies the job of TrueType hinting, because far fewer hints and very few deltas are required to produce a decent, legible shape. Indeed, over-hinting can be a problem for ClearType. I agree very strongly with experienced hinter Laurence Penney that hinting is design. When you are hinting TT, you are designing glyph shapes for different output devices and resolutions. The role is similar to that of the traditional punchcutter, interpreting the drawings of a 'designer' to produce an appropriate impression at different sizes. So the design job of a hinter is changed if the target devices are limited to subpixel rendering. Today, the situation is difficult, because there is a great diversity of screen technologies and rendering technologies at work, and it is hard to optimise for any one of them without downgrading performance on another.

hrant's picture

Nice analysis.

> Are there any type designers deliberately taking exploiting the technology in their designs to improve screen readability? Yes.

Encouraging. Who?

> a typeface that is optimised to take advantage of such rendering might not be the best choice for documents that will be read onscreen and printed.

Unless there's some oscure limitation inherent in designing for subpixel rendering that eludes me, it seems that that's not such an issue.

hhp

tsprowl's picture

If you optimise a design for subpixel rendering, you are making the decision that this is going to be a screen font, and other considerations are going to be ignored.

yes that was my idea specifically. Seems there's thousands of bitmaps out there being produced now-a-days...If subpixels stuff were involved there could be a whole slew of new designs/usages that can get really intricate. think of oblique bitmaps or scripts - with subpixels, instances of Shelley Volante might work in bitmap. well....you get my drift, other categories of type might have a chance on screen rather then the proliferation of harsh grid sans-serifs or techy-styles.

but maybe I should learn more still - thanks for the explanation


hrant's picture

No, it's not so black-and-white.
Design is a balance of compromises.

hhp

John Hudson's picture

In designing type for screen display using subpixel rendering, I've found that quite minor changes in stem weight affect type colour radically (unlike b/w or greyscale, in which radical changes in stem weight don't necessarily make much difference to the pixel pattern at small sizes). I discussed this with colleagues in the ClearType team at MS, and they noted that they had needed to make the stems of Berling Antique heavier for the Reader than they had been in the original design, in order to get the kind of contrast they wanted. This is what I mean when I say that a typeface that has been optimised for this technology may no longer be optimum for printing. It probably won't look terrible printed, but it would not look as good as it might if it were optimised for that technology.

hrant's picture

> they had needed to make the stems of Berling Antique heavier

But not through hinting? Or does CT [still] ignore hinting?

hhp

aluminum's picture

I'm confused. Isn't sub-pixel rendering simply a 'hack' to get more resolution out of an existing display? As such, how is optimizing for sub-pixel rendering really any different than optimizing for anti-aliasing?

Also note that as more and more people go with LCD vs. CRT, the whole sub-pixel argument will be moot anyhow.

John Hudson's picture

But not through hinting? Or does CT [still] ignore hinting?

Not through hinting: the outlines were adjusted.

ClearType does not ignore hinting: it only completely ignores x-direction deltas. With increasing screen resolution, the question arises whether fonts for CT need any x-direction hints, however, since the interpolation across subpixels is so good. So I see a value in adjusting the weight of stems during the design phase, to optimise for screen reading, since this is much less time consuming and reliable for this technology than trying to control subpixel rendering using x-direction hints.

hrant's picture

Great info! Thanks.

hhp

John Hudson's picture

Darrel,

I certainly would not characterise ClearType rendering as a 'hack'. If you read the papers MS have produced on colour filtering, I think you'll agree that this is a pretty impressive and well thought out piece of work. I'm not sure I would say the same about all subpixel rendering: Acrobat's blur everything until it looks smooth but grey approach might be considered a hack.

Antialiasing is very coarse compared the x-directuon resolution gain in ClearType, so the difference is that in CT you can get a much better approximation of the real vertical stem weights. However, due to gamma correction and other screen issues, the kind of stem weights that work best on screen are not necessarily the same as work best on paper. Remember, I'm talking about optimisation for the technology here: most traditional fonts look great in ClearType, even though they were originally designed for print. I'm talking about a minority activity of designing fonts specifically for screen reading, and for which printing is not a significant consideration.

I'm not sure that I understand your comment about LCD vs CRT screens. Subpixel rendering specifically benefits LCD screens, which are the only kind of screen with subpixels: so the issue is far from moot, rather it is becoming more important. [Note that there is also a small but observable benefit of ClearType rendering on CRT screens, but this has to do with very subtle colour bleed between pixels, and not with subpixels. I don't think other subpixel rendering systems would display this particular benefit, since it is a direct result of the CT colour filtering.]

tsprowl's picture

John your a fountain of info on this aren't you.

I think I'm going to experiment and try something with all this new info. then I'm sending it to you. hehe

matthew_dob's picture

Cleartype has many disadvantages over other sub-pixel systems but wins through in its rendering at small (<16pt) sizes. It appears that it works by using an un-anti-aliased rendering of the font stretched horizontally by a factor of three, then applies a filter to it to even out the colour balance, and then "un-streches" the image, leaving the sub-pixel rendering. This has the disadvantage of neglecting near to horizontal lines. Where other systems might give a smoother, anti-aliased line, Cleartype renders an un-aliased version. The diagram below demonstrates this...
1) my sub-pixel rendering using Paint Shop Pro
2) PSP's traditional anti-aliasing
3) Cleartype's version

cleartype

Vive la difference!

Matthew

John Hudson's picture

I agree that ClearType wins principally at small sizes, which is what it was designed to do: it is not a technology made for graphic designers, it is a technology made for readers. However, you should remember that we've only seen the first two generations of ClearType, and that a group of very clever people are continuing to develop it in new ways. Suffice to say they are aware that it currently looks like crap at large sizes.

Your analysis of how ClearType renders to subpixels is not quite accurate, although you understand the principle. The text is not stretched (supersampled) by a factor of three: it is stretched by a factor of 16, so that different colours can be weighted according to optical sensitivity. I can't remember exactly how many 'stripes' each colour gets, but I think it is something like R=5, G=9 and B=2. The human eye is very much more sensitive to shades of green than to shades of blue.

rcapeto's picture

The human eye is very much more sensitive to shades
of green than to shades of blue.


The classic luminosity formula:
R = 30%
G = 59%
B = 11%

hrant's picture

> we've only seen the first two generations

Exactly.
I remember when I first started understanding the basic mechanics of CT, and then looked more closely at actual renderings, I thought: wow, this is nothing yet - the really good stuff is yet to come.

In fact, there are improvements that have yet to be done in "regular" screen rendering, and if/when those are applied to CT, watch out!

> it is stretched by a factor of 16

Wow, more great info! Isn't there an NDA applicable here?... :-)

hhp

hrant's picture

> we've only seen the first two generations

Exactly.
I remember when I first started understanding the basic mechanics of CT, and then looked more closely at actual renderings, I thought: wow, this is nothing yet - the really good stuff is yet to come.

In fact, there are improvements that have yet to be done in "regular" screen rendering, and if/when those are applied to CT, watch out!

> it is stretched by a factor of 16

Wow, more great info! Isn't there an NDA applicable here?... :-)

hhp

John Hudson's picture

Isn't there an NDA applicable here?... :-)

I sure as heck hope not! I'm being very careful to repeat only things that I have heard Greg Hitchcock say in public lectures, e.g. at the Linotype tech forum in Heidelberg and the MS complex script seminar last summer. MS have their patents now, and have published papers on this subjct, so they're much more willing to talk about the details than they were when CT was first announced.

There's a lot of other very cool stuff I can't talk about. :-)

aluminum's picture

"I certainly would not characterise ClearType rendering as a 'hack'."

I'm using hack in the technology sense. It's modifying a default behavior of software/hardware.

"I'm not sure that I understand your comment about LCD vs CRT screens. Subpixel rendering specifically benefits LCD screens, which are the only kind of screen with subpixels:"

That's me being dumb. I didn't realize that LCD pixels were made up that way. 'doh. I do now! ;)

Now, I wish instead of focusing on sub-pixel rendering, display manufacturers would just get around to making high res displays ;o)

(BTW...to give credit where credit is due...Apple came out with this long before MS: http://ww.grc.com/ctwho.htm )

hrant's picture

And to give clubbing where clubbing is due: Apple royally failed to protect/use the invention. They were too busy getting the right sheen on their titanium toys.

hhp

John Hudson's picture

Apple came out with this long before MS

Indeed. MS never claimed to have invented the idea of subpixel rendering. But Apple gave up on the idea very quickly, probably because of negative feedback about the colour fringe on the letters. The most brilliant aspect of ClearType is not the supersampling to increase x-direction resolution but the filtering MS developed to reduce the colour fringe to acceptable levels while maintaining overall contrast.

Syndicate content Syndicate content