Matthew Carter's Georgia is a work of genius, rightly acknowledged as the state of the art for b/w screen rendering, so I'm starting the visuals with a comparison of Georgia and Constantia in two different CT rendering environments, to show how the state of the art for b/w rendering is not necessarily ideally suited to the new rendering systems. This is primarily a comparison of a font optimised for one renderering technology and one optimised for another.

John, I'm not sure I'm seeing the right picture here. The XP looks better than CT on my iBook -- for instance comparing the weight of the vertical strokes of the "T" and "h" in "The" -- in the CT rendering of Constantia, it's bolder in the h than the T.


XP. Left, Georgia, right Constantia


CT. Left, Georgia, right Constantia

The 'Longhorn CT' example (more accurately the Avalon WinFX rendering) uses subpixel spacing, so if you look at other T's (like the T in Everson Typography) you'll see that the bitmap pattern is different. In the XP version which snaps to whole pixels every T is the same.

The trade-off here is better spacing vs consistent bitmap pattern.

Cheers, Si

PS Nick - Irish Bloggers alert - http://www.sluggerotoole.com/archives/2005/03/fontastic_paper.php ;-)

>Irish Bloggers alert

So that's why they chose the typeface. Maybe I should go after the Korean market as well!

Some informal observations:

The biggest difference I can see in the way CT renders text compared to OSX (Panther) is in the way horizontal elements are rendered.

On both CT and OSX with sub-pixel smoothing on ("Best for LCD"), vertical elements are rendered using sub-pixel rendering.

With CT, horizontal elements are pixel-aligned, just like non-antialiased type. This leads to a bit of jaggy-ness in strokes that are close to horizontal, like the middle of the lowercase s in the examples John provided. The most noticeable difference in the Longhorn example is the addition of sub-pixel spacing, as Si noted.

With OSX, horizontal elements are antialised the conventional way, using varying shades of gray. This leads to smoother glyph shapes, even in slightly horizontal elements, but also a certain degree of fuzziness with some fonts at some sizes. OSX also appears to use sub-pixel spacing because the glyph patterns vary in the same way as the Longhorn CT example.

To my eye, CT renders text more crisply with some distortion of the glyph shapes, while OSX renders text with more even color and less distortion at the expense of crispness.

They both look good to me, but for different reasons. I don't know which would be best for reading.

The Longhorn renderer can apply y-direction antialiasing to smooth horizontals, but this would normally only be turned on as the type size increases (this is design specific: MS have defined a new GASP table format so that font developers can set the sizes at which y-direction antialiasing kicks in). Y-direction antialiasing makes display type look excellent, but as in the OSX display, it reduces contrast and crispness at small sizes. One thing that pretty much all legibility/readability research agrees on is that contrast is key to reading.

Interesting. This is one of the things that bugs me about the current version of CT I've used on a Windows XP set up I have. With CT on, display sizes look more jaggy than with simple antialiasing. It's good that it's being addressed.

Left, Verdana. Right, Corbel. Total graphic width on my screen = 162mm. Longhorn CT rendering.

Cambria specimen. Total graphic width on my screen = 75mm. Longhorn CT rendering.

Candara specimen. Total graphic width on my screen = 75mm. Longhorn CT rendering.

Calibri specimen. Total graphic width on my screen = 75mm. Longhorn CT rendering.

Corbel specimen. Total graphic width on my screen = 75mm. Longhorn CT rendering.

Thanks, John, for these marvellous specimens. I am just blown over by how much more readable Verdana is than anything else you have shown. You have really helped me deepen my appreciation of Verdana. In the past, I believe I have muttered some words to the effect of 'oh, Verdana, nothing special, anyone who has ever seen a 4-1/2 point news sans could have done it.' And I have also said that Verdana was the world's ugliest sans serif, even if it was also the world's most functional. Well, I take that all back. Verdana is a miracle of legibility, and in massed type, as you have set it here, I must to my utter astonishment admit that I actually find it almost heartbreakingly beautiful. Now those are words I thought I would never utter!

Again, with Georgia. You really help me see why it is so much more readable than the comparison fonts. Carter is really much better than even I thought he was.

Great job!

Constantia specimen. Total graphic width on my screen = 75mm. Longhorn CT rendering.

Yes, Verdana is really excellent: difficult to improve on, and it adapts well to ClearType. It will be interesting to see what people think of the Meiryo Latin, which is based on Verdana. I was about to post specimens of that type, including comparison with Verdana, but now Bill has shown up in this thread, I'm not going to post any more.

I think Georgia suffers under ClearType: the stem weights, geared to producing the best b/w pixel arrangements, make the colour of text very uneven under ClearType. It would be great if MS hired Matthew to revise Georgia for ClearType: it is a splendidly readable face (barring the default linespacing, which is too tight, but I doubt if that is Matthew's fault).

Anyway, goodbye, again.

This is a fascinating display. It is immediately clear that Verdana, in either rendering system, looks better than the other sans serifs. It is less obvious why Georgia, again in either rendering system, is superior. Very generally I would suggest that it is a matter of the proportions - - the proportions and weight of each letter individually, and each in relation to its neighbour, the perfect judgment of the serifs, which really do aid readability, and the well-judged x-height. What I find particularly remarkable is that when magnified even a little, one sees that the rendering of the newer fonts is better in some respects than for Georgia. Yet there is something so sturdy and functional about the shapes of Georgia that even at its worst, it easily emerges as the most readable. How on earth did he do it?

One thing I am curious about is that so many of the new fonts seem to be somewhat compressed. Traditional wisdom has it that this impedes readability, and I am with traditional wisdom here. But was there a practical reason to make the new fonts slightly compressed?

I notice also with interest that few of the numbers in the new fonts seem happy. Some are too small, others have too much ascender and descender play, few seem just right in relation to text, although they might look fine on their own. Were they tested in text?

Do please post some more. It would be nice to see Meiryo Latin, and it would be lovely to see Georgia in one of the narrower columns of continuous text such as the one directly above.

it would also be great to see some larger text samples and perhaps a few fitting tests.

Oh, and a question: you write, "I suspect that most of you will have a lower resolution screen than mine, so the graphics will appear larger on your screens than on mine."

Could you tell us what your resolution is? Is it some special kind of monitor that can't be generally bought? And what are your preferred CT settings? What I'm getting at is, if you're going to say that the particular equipment you use when viewing these pages is important, than you ought to specify the equipment. That way some of us might be able to view the material under optimum conditions. I would expect that everyone reading this has some kind of reasonable minimum as regards equipment, but perhaps I am naive?

Bill, see my response in the 'CT and the future' thread re. Verdana and Georgia.

[I'm sitting here -- unable to sleep as it happens -- wondering if it is possible that we might actually have a civil discussion. I would like that.]

I think the new CT fonts might seem compressed because they're not what we've become used to in 'screen fonts'. With the exception of Calibri, which I think of primarily as a subhead type, they're actually not compressed in terms of the norms of print typography. Candara is very generously proportioned, and Corbel and Constantia are both quite typical in width for sans and serif text types. ClearType reduces the gap between screen and print, and as it and monitor resolutions improve the gap will be further reduced. Unlike some of the people at MS, I'm not convinced that reading on screen will ever be as comfortable as reading print (presuming similar levels of relative quality -- there is some printing that is decidedly worse than text on screen!), but I do think that ClearType brings the conventions of screen and print typography closer together. Yes, after using Georgia and Verdana in every screen reading situation for so long because they were the only option, most other fonts are going to seem compressed at first.

Yes, numerals were tested in text. I have a theory about numbers on screen, based on the fact that we don't actually read numbers in the same sense that we read text (except perhaps commonly recurring numbers such as dates from recent years). I suspect that we notice the compromises of low resolution more consciously in numerals than in letters forming words.

By the way, the new CT fonts all contain at least four different numeral styles (not counting superscript and subscript): proportional oldstyle, proportional lining, tabular oldstyle and tabular lining, and some of the families also include smallcap lining numerals. MS decided that the proportional oldstyle figures would be the default for all the families except Cambria (the latter is intended things like for business documents, in which tabular lining is likely to be expected as default).

I will try to find some time to post more graphics in the next few days. If you let me know what kind of things you would like to see in a fitting text, I can see what I can do. Making the graphics with my test tools is time consuming, because I can't show multiple sizes or styles at the same time, so have to cobble the graphics together.

Regarding my monitor, as I noted in the caption for the first graphic, my display is 146ppi. This is a 15" widescreen Dell laptop display, running 1900 x 1200 pixels. So I have a pixel pitch of about 0.1725mm. As far as I know, it is the best display this side of the IBM 200ppi monitor.

I have another laptop that I use when I'm travelling because it is very small and light, but it only has about a 105ppi screen and it is very difficult to go from reading ClearType on my Dell to reading it on the lower resolution screen. For some tasks, I turn ClearType off on the smaller machine and use b/w rendering instead. ClearType really comes into its own around 130ppi: that's the 'sweet spot' at which, even at text sizes, design details like serif form start to become distinguishable and the individual identity of typefaces start to assert themselves.

>John, a footnote: When you say "One thing that pretty much all legibility/readability research agrees on is that contrast is key to reading." Do you mean contrast in the sense of foreground/background separation, or stroke contrast? I don't see that legibility/readability research addresses the second at all, and when it does (indirectly, through the serif / sans serif conundrum), the test--as Ole Lund's PhD thesis shows--are riddled with cognitive scientific problems. Arguments for the importance of contrast have another kind of foundation.

here is Calibri 72 point spacing of the "u"

Calibri spacing Calibri72point.bmp (433.9 k)

This is great, thanks very much. At first glance I see nothing alarmingly irregular in this (the Calibri spacing), although I could (and in fact do) quarrel with the underlying parameters, but that is another issue. This makes me suspect that there really _is_ something fantastically wrong with the Poynter showing. I should have the book soon, anyway. But ... if the Poynter setting is an accurate representation of the book (and I see no reason to suppose that it is not) and if the book, therefore, has some really awful spacing, the question arises, how on earth did it get there? Is it even remotely possible that the book was set with InDesign, with the optical kerning settings on? And could they have caused these problems? Just last week, finally, after all these years, I actually tested InDesign CS for an article. (Surely, John, you will be glad that I have done this, as you have been asking me to do it since at least the end of 1999!) I discovered some rather alarming things about the optical kerning - far worse than anything even I ever suspected. Could that be the problem? Or, could some designer have done some hand kerning that was hopelessly misguided? Such things, after all, do happen! Needless to say, it's MS that ought to be finding out the answers to these questions, not me! But let me get on with my measurements. Too bad I don't have the fonts! I should hate to have to go abegging to the ACT for them!

Peter, I mean contrast of black and white. When I mean the other kind of contrast I'm always careful to specify stroke contrast.

That is a useful distinction in the vocabulary.

Is it even remotely possible that the book was set with InDesign, with the optical kerning settings on?

The character set showings in the book use InDesign's optical kerning along with tracking, since the goal in this is simply to ensure that all letters are reasonably evenly spaced, including conbinations that occur in such showings but not in text so might not be kerned in the font. But none of the other showings of the fonts in the book should be optically kerned. If they were, it would be an error, and I'm sure I would have heard about it from Luc(as), because he is very particular about his spacing, as anyone who has seen his 'Kernologica' presentation knows. I don't always agree with the way he spaces and kerns, but I do know that he is very thorough in his approach.

Ah. It _could_ explain everything, if the text in the Poynter specimen was, indeed, kerned by InDesign.

Of course, one couldn't be sure without testing the font in InDesign, but I have a very good idea of what might have happened.

Consider the word ' jumps' set at 12 point in Adobe Garamond OT Pro. No kerning for any of the letter combinations. Everything just sets perfectly, it has been perfectly spaced by not any longer so Slimbach. Now apply 'optical kerning'. InDesign subtracts 23 units of space between j and u, adds 14 units between u and m, subtracts 16 units between m and p, and -1 units between p and s.

The u is now much too close to j, and there is a coruscating river of space between u and m. A well-made font is now useless -- it looks like shareware.

Something very like this could have happened to the text shown on Poynter, and tracking would only have magnified the unfortunate artefact.

For many years I suspected that what I have just described would happen with InDesign optical kerning. But because of my friendship with John, and my relationship with

If you are referring to the image of Calibri on <a href="http://www.poynter.org/column.asp?id=47&aid=78683">this page</a> then that is not the character set showings to which I referred, but part of the text samples which should not be optically kerned and, obviously, are not tracked. To confirm what we're talking about can you grab and post the precise image from the Poynter pages to which you are referring, and let me know what combination you think has spacing problems. Then I can check it against the book in print, and also check the original InDesign file.

I'm intrigued to know how the Poynter people produced the images. The website says the images are scanned from the book, but the total lack of any paper texture indicates that the scans have been touched up in Photoshop.


Good Heavens! John! Do you realize that's how our famous quarrel started?

Really? I never understood how it all started. If that was it, it hardly seems worth all -- or any -- of the years of bitterness, insults, accusations, etc.


Anyway, here is the Verdana / Meiryo Latin comparison, which I think is very interesting. The differences in the designs on screen are much more evident than they are in print, especially at the larger size shown here (nominal 11pt on my setup).

The very tight linespacing is generally acknowledged as the sole obvious fault of Verdana (and also Georgia). The linespacing of Meiryo is particularly huge in comparison, but bear in mind that the linespacing in Meiryo is primarily intended for Japanese text, of which I'll post an image shortly.

Left, Verdana. Right, Meiryo. Total graphic width on my screen = 152mm. Longhorn CT rendering.

In the Now read this book there is only a very small, three line comparison of Verdana and Meiryo, but the most obvious thing about it is that Meiryo is more space-efficient than Verdana. I was wondering how much more efficient, so prepared this illustration. The strings of letters in this image represent the relative frequency of lowercase letters occuring in an English text of 34,643 words; I've compared this to frequencies in other, somewhat smaller samples (c. 10,000 words) to confirm that it is similar or identical to what you could expect in other texts. The only distortion in the frequencies is the q, which was rounded up to 1 from .3 in order to ensure that at least one instance of every letter appeared in the sequence.

To assist comparison, I have reduced the linespacing of Meiryo in this graphic to match that of Verdana.

Top, Verdana. Bottom, Meiryo. Total graphic width on my screen = 149mm. Longhorn CT rendering.

Last image today:

Here is a specimen of Meiryo Japanese text at the same sizes as the samples of the Latin specimens of Cambria et al, above. I know almost nothing about Japanese typesetting rules, so have just cobbled together random sentences, and I'm pretty sure that I've broken at least one rule about what characters can occur at the beginning of a line. But this gives you some idea of just how phenomenally good the ClearType rendering of kanji is.

I've underlined, in red, an instance where the number of horizontal strokes has been reduced with hinting at the smallest size (compare to the same kanji glyph in the medium size); there may be others. Eichi Kono discusses this technique in Now read this, and explains the cultural precedence for this in particular calligraphy styles.

Meiryo specimen. Total graphic width on my screen = 112mm. Longhorn CT rendering.

Most interesting images. I hate to be so perverse, but I don't think I like Meiryo as much as Verdana. However, I've only looked at these images, briefly, on the Mac, on two monitors, 1600x1200, colour and greyscale (which remains I must say sharper than anything else I've ever seen). I'll look on the PC later. Speaking of which, I have exactly the same Dell screen you specified. Does it make any difference what browser is used? How should ClearType be set up? And are there any other desired parameters? I wouldn't want to make some silly mistake in some setting that would invalidate the view.

But before anything else, and just based on the quickest glance, I am coming to wonder the following: I am not at all sure I like the new proportions of Meiryo (considered not at all as a companion to Japanese, but simply as an alternate Latin to Verdana - that is of course a wrong consideration, but let it stand for the sake of argument), and I know I don't like many of the proportions (x, desc, asc, cap height) of the new CTs. So: is there a technical reason for these proportions? Or was there some research which suggested these proportions as desirable and which was incorporated into the design briefs? I suspect I'm on the wrong track here, but I do wonder.

Speaking of the design briefs, that is helpful background, and there is much to respond to in that and other areas when time allows.

Finally, I find myself uncomfortable with the new default line spacings, which I find too generous. I wonder the following: are the new line spacings calculated to improve comprehension of every word? If so, is it possible that they impede rapid skimming? My point being that I don't actually want, on screen, in many cases, to read every word of text before me as I haven't the time. I want to be able to glance at the entire paragraph and take away as much information as I can in a couple of seconds. For this purpose, I would suggest that tighter default line spacing might be more helpful.

Okay, one last post, and then I'm going to sleep.

Since you're looking at screenshots rather than live rendering, it doesn't matter how you have your ClearType tuned. What you are seeing is how I have my ClearType tuned. You should be able to make a direct comparison of live Georgia rendering on your own screen to the first image above if you do want to compare tuning. Be sure to compare to the WinXP rendering in the image, not the Longhorn. The background colour of the above images is R255,G255,B220.

The proportions of the new types was not technically determined, no. The horizontal proportions were left to the discretion of the designers. Since the new CT renderer allows glyph boundaries to fall on subpixels, there isn't a technical need to conform to something approximating full pixel widths as there was for Georgia and Verdana. As I wrote previously, I think part of your reaction to the proportions of the new types might be that you are so used to the wideness of Verdana and Georgia in on screen reading. I'd be interested to know, in a couple of years, after you've had a chance to live with the new types for a few months, whether you still dislike their proportions.

Regarding the linespacing, the idea was that the default linespacing should be similar to that which one would use e.g. in a book for a text column of typical length according to the canons of traditional typography. I don't think one can make assumptions about how people want to read text on screen. Yes, there is probably more skimming on screen than in print, simply because of the quantity of information presented, but I think it would be a mistake to set default vertical metrics in a font based on this. And, as I say, this is only the default spacing, employed where there is no human intervention in the linespacing. You can always increase or reduce the linespacing as appropriate to particular text using CSS.

Goodnight.

Hmmm. Sleep well!

When you rise, I'm curious: The specimen that shows the lines of letters repeated... In the old mak days, there was an option for switching from sub-pixel to whole pixel positioning. (it was not called that , but that''s what it did). In those days, you could tell right off by putting your finger on the i or l key, and holding it for while, then, if you looked closely at the picket fence, you could see some spaces larger than others in the sub-pixel positioned vs. pixel positioned specimen, where the picket was/is perfect.

In the specimen, (Post Number: 1857), both the images of the glyphs and the spacing between the glyphs varies. So, when I say color has become more important than rythym, you know what I mean.

Agreed or halucination?

(Also, if I were in charge, I'd make the dot on every i in this specimen, every one, twice as heavy (taller in this case). Then I'd spend the rest of the day on the j dot.

Otherwise, I think these are some of the finest looking screen shots of what must be the finest screen type that can be made with XP's CT.

Agreed or halucination?

In the specimen, (Post Number: 1857), both the images of the glyphs and the spacing between the glyphs varies. So, when I say color has become more important than rythym, you know what I mean.

Can you perhaps grab the image and edit it to indicate specific places where you think the spacing varies? I'd like to confirm that we are talking about the same thing.

If the spacing did vary, it could only vary by a subpixel, but I think there is a spacing illusion caused by the variation in the glyph rendering that may account for what you are seeing. If you look at the close-up from the Meiryo in post number 1857, below, you will see that the spacing of touched sub-pixels is actually very regular indeed, but the colour density of those touched pixels varies, making some glyphs look lighter on one side or the other. I think this can create the illusion of greater space between some glyphs than others.



The question of rhythm is an interesting one in regard to screen typography, because although it is possible to create very regular intervals of verticals in b/w bitmap rendering -- and this was a major goal of most of the hinting that has been done in the past decade and a bit -- this is done at the expense of that interplay of rhythm and form that we prize in good text faces for print. So, for example, the subtle reduction in the widths of the bowls of a lowercase m relative to that of the n is abandoned in b/w screen rendering in favour of total regularity in the spacing of verticals. Is this really typographic rhythm? Or is it a mechanically determined simulacrum? The rhythm of a well hinted font in a b/w display, compared to the rhythm of even the same typeface in print (Verdana is a good example, despite being designed around bitmaps: Matthew has put into the design as much subtlety of rhythm as he could, which is only evident in print), is a bit like the rhythm of a drum machine compared to Buddy Rich.

The promise of sub-pixel rendering and positioning -- only just beginning to be delivered on the higher resolution screens, in my opinion -- is that the subtleties of typographic rhythm, i.e. rhythm as it relates to form and not merely to regularity, can be deployed on screen as it is in print. (although obviously not to the same degree of refinement).

I think at the moment, we're in an awkward position where the technology of screen displays is in transition, and we can no longer optimise type for a presumed level or type of display, as we could when MS commissioned Georgia and Verdana. ClearType is in part an anticipatory technology. This is also true to some degree of the new CT fonts; there are details in Constantia, for example, which probably won't be evident on screen, except at very large sizes, for several years, but they are based on the direction of development of both the CT renderer and screen resolution.

Counting subpixels in the enlargement above, I note that there seems to be a single subpixel increase in space between the touched subpixels of the 2nd and 3rd i and between the 5th and 6th. But this is because the stems of the 3rd and 6th i appear to be one subpixel narrower than the others. This confirms my suspicion that the actual spacing is regular -- in terms of the advanced widths of the glyphs falling on regular subpixel boundaries --, but there is an illusion of spacing variation based on the variation in stem rendering.

At this point in the conversation, it would probably be helpful to hear from Mike Duggan or Beat Stamm, since we're reaching the limits of my own technical knowledge of ClearType. Once we start getting into the maths of supersampling and colour filtering, I'm less confident in my ability to explain, or even understand.

OK. I know I'm going to get jumped on for all-surpassing ignorance in this area, but why is it necessary for the same consecutive letter to be rendered differently? Is there some good documentation anywhere that would explain this? (And if you tell me it's all in that book I haven't read yet, I shall offer myself as a sacrifice to -- well, something with a typographic theme.) That can't, intuitively, be desirable. Yet it doesn't seem to make much practical difference either, does it? Is this one of those cases where you get say 70% advantage and 30% disadvantage, so you go with it since the benefits outweigh the losses? Also, now that we are getting a little technical, how do competing systems work? What parts do patents protect? (I suppose I should just look them up instead of asking. -- uspto.gov, I think, for anyone interested, or just google patent server.)

Bill: if you want to control the space between letters with subpixel accuracy, then the relative offset of the glyph with respect to the pixel grid will have different subpixel values for different renderings. Otherwise, you're forced to quantize to whole-pixel space values, which can be ok if the spacing of the font is carefully tuned to screen resolutions, but can also be a serious problem for WYSIWYG rendering.

The basic antialiasing algorithm is to render the glyph at a higher resolution in monochrome (say 400 dpi for a 100 dpi rendering), then count the number of "on" pixels within each 4x4 cell. If you offset the pixel by 1/4 increments, that corresponds to offsetting the high-resolution rendering by 1 pixel increments before doing the subsampling, which is all pretty straightforward. (Adobe uses 4x4 subsamping and 1/4 pixel subpixel positioning accuracy in their font renderer that ships with Acrobat; Apple does something a little different which results in even higher precision).

Doing grayscale subpixel offset to a stem aligned to the pixel grid (as the result of hinting, perhaps) reduces contrast. If the high res rendering is ####____, and you subsample by 4, then you get #_, which is quite crisp. Move two pixels to the right, so that the high res is __####__, and the subsampled version has two pixels of 1/2 intensity.

With ClearType, subpixel positioning has much less of an effect on contrast, because of the subpixel LCD addressing. A stem which is rendered as RGB___ and shifted 1/3 pixel to the right turns into _GBR__, which visually looks almost the same. (I believe Adobe uses 1/3 pixel subpixel positining increments in their CoolType implementation). Even if you shift by 1/2 pixel, you get _gBRg_, which is a lot better than the grayscale case.

As far as I know, there is no patent protection on basic subpixel positioning in antialiased text (I know AA text itself goes back well over 20 years). MS, of course, has a sheaf of patents on their ClearType implementation, but it is unclear to me whether those cover the basic concept of LCD subpixel rendering, or just their specific implementation. If the latter, then Apple (with their "font smoothing") and Adobe (with CoolType) don't have anything to fear in the patent department.

As for other competing systems, you might be interested in my FontFocus, which does not use LCD subpixelling, but does a much better job of preserving spacing than techniques based on quantizing to pixel coordinates. I am finding, incidentally, very little interest in the technology, so it's not clear I'm going to keep working on it (even though I do have a few more ideas on how to refine it still further).

In simpler terms, Bill, the rendering will vary depending on whether a stem edge falls on an RG GB or BR subpixel boundary. That will determine across which subpixels the whole stem will fall, and hence what colours need to be dealt with. See page 13 of Now read this, which explains the principle and purpose of subpixel positioning.

Note that this is particular to the new, Longhorn version of the CT renderer. The version you have on WinXP always renders each glyph the same way because the left and right sidebearings always fall on a full pixel boundary.

John: thanks for explaining in simpler terms. In the case of ClearType, it's clear that it's not just a case of offsetting by 1/3 pixel (i.e. whole color stripe) increments, as you can see from the following example (taken from your Meiryo, but with the stripe pattern faded in so you can more clearly see the pattern in the underlying subpixel image).



Note how the first two i's have a "darkest pixel" width of two pixels, while the third's is a single pixel wide. If I had to guess, I'd say that the subpixel positioning is calculated to 1/6 pixel accuracy. So the subpixel offset does have an effect on contrast, even aside from the question of which color stripes get addressed.

In CT rendering, outlines are 'supersampled' across 16 pixels, and for colour filtering purposes these are weighted to account for greater human sensitivity to some colours than to others: Green 8, Red 5, Blue 3, if I recall correctly. As I understand it, it is impossible to employ subpixels positioning without some effect on contrast, because of the difficulty of obtaining identical shades of 'grey' in red, green and blue subpixels. Nice graphic, by the way.

I've asked Greg Hitchcock at the MS ART group about the stem rendering and resulting illusion of spacing variation. Hopefully either he will chime in here or provide me with some information that I can post.

Thanks to you both. I'm afraid I'm going to have to be sacrificed but may I please visit Ralph's site first?

Greg Hitchcock reports:

I didn't get a chance to track down what is happening here. I can see
the issue. The "dot" of the "i" does seem to fill out the whole 7
subpixels when the main stem is only 6.

I'll look at it further, but it will probably be in the middle of next
week.

So, as they say, 'watch this space'.

Raph, it would be a pity if you abandoned FontFocus. It looks really, really good.

I have no intention of sounding typographically blasphemous, but which was exactly the reason which pushed in the direction of having anti-aliased text display in the new Operating Systems?

It may be I'm negatively naive or it may be my total ignorance on the matter (and on advanced technical issues), but I keep looking at anti-aliased type and still miss in which way it could/should be better than aliased one.

At the beginning I thought the idea and the direction taken by the research on onscreen readability might have come from the need of having a better rendering of Japanese or graphically complex glyph systems (even Arabic, which I assume is harder to read in small sizes compared to Latin), but in the end I came home after work, and as I switch on my Mac with Os9, I can't help but being delighted by Georgia and even Arial in their aliased form.

Is there a main reason for which people felt the need of an anti-aliased screen typography?

You're on to something Claudio.

It's not that there is anything wrong with the technical goal of sharper looking type, but the question should focus more on the relationship between design and technology.

In other words, how are these new CT types "true to the medium"?

The virtue of Base and Verdana is that their design was informed by the constraints of the screen medium. The same is self-evident with pixel fonts such as Unibody.

In what way do the designs of the new CT faces address the unique qualities of ClearType, or is the program a pastiche of print typography?

I find anti-aliased text easier to read than aliased, at least on LCDs (and I don't use anything but LCDs these days). When I was running OS9 on a Powerbook, I always found the text to be too crisp, having an effect similar to the "dazzling" which affects print faces when there is too much contrast between the thicks and thins.

On an LCD, you get sharpness similar to print, but at a much lower resolution. Try reading aliased type in print and you can see the problem. Some sort of anti-aliasing is called for, but clearly there are differing opinions about the best approach.

CRTs are a different story. I would agree that aliased text looks better. But I think this is because of the inherent softness of the CRT. Anti-aliasing just makes it look even softer and blurrier.

This debate will probably continue inconclusively until reasonably-high-resolution displays become the norm and the point becomes moot (i.e., indefinitely).

Mark wrote:
This debate will probably continue inconclusively until reasonably-high-resolution displays become the norm and the point becomes moot (i.e., indefinitely).

In the meantime, I'll have a sober reading with aliased type. Luckily I hardly use my old portable iBook.

In what way do the designs of the new CT faces address the unique qualities of ClearType, or is the program a pastiche of print typography?

Nick, do you have a copy of Now read this? If not, I can arrange to have one sent to you.

In the book, all the designers discuss the specific design decisions that relate to the ClearType rendering technology. At the kick-off meeting in Redmond, we began with a technical briefing on how ClearType works, and the development progressed from that understanding. The initial phase of development for the designs involved sending test glyphs for hinting, to test rendering under ClearType. My own design went through two rounds of weight testing, with five variant stem weights. Decisions about details were informed by the kind of forms that ClearType renders particularly well, and avoiding as much as possible the kind of shallow horizontal diagonals that look jagged under ClearType. I also conducted a series of tests to determine the best italic angle for Constantia's forms under ClearType.

The designs do address the unique qualities of ClearType. What they do not do, unlike the Base types for example, is to fetishise the technology stylistically.

Refer also to my explanation of the design briefs in the other thread: the types need to look good in print as well as being readable on screen, because they are intended for areas of use that include both print and screen.

Not to toot my own horn too much, but this paper presented at Electronic Imaging '04 might help with questions about how changes in display resolution affect text rendering contrast. See also Avi Naiman's thesis (link to PS file) for an older but thorough study, especially of AA vs. bilevel rendering. Unfortunately, most of his work was done with CRT displays and doesn't generalize well to LCD's, for the reasons that Mark pointed out.

>Nick, do you have a copy of Now read this?

Simon has sent me one, but it hasn't arrived yet.

>The designs do address the unique qualities of ClearType.

Good. But it appears that I can't comment further on this till I have,
as instructed, read the book and seen the fonts in print.

>What they do not do, unlike the Base types for example,
is to fetishise the technology stylistically.

Base -- a deviant typeface, eh? Must be why I like it.

Posted on Tuesday, March 29, 2005 - 4:47 am:

So you're saying that it's kind of ironical to discuss the qualities of a series of type designs, because the designer doesn't have control over the design anyway, because the new rasterizer for which the types have been developed will render them in dozens of different ways on the screen? Is that the basic point?

Well, so it is. That raises the question, at what point does this make any perceptible difference in the real world? Isn't that part of the science of CT, that artefacts will not be perceptible? I suppose there is the possibility that the studies are all wrong, and the artefacts will actually result in subliminal viewing fatigue, but that seems far-fetched.

A problem, I now see, with John's images, is that they are, in fact, images, and can thus be enlarged and substantial defects revealed.

But so what? If you enlarge a photo too much, and put your nose right to the print, the grain will be too large and you won't see anything. But if the normal viewing distance is respected, everything looks sharp and artefact-free.

The same applies here. When we are looking at real CT onscreen, will we really be able to tell that characters are non-uniformly rasterized and that spacing is subliminally irregular? And if we enlarge the type onscreen, won't such artefacts decrease, not increase? I mean, the type is scalable, so artefacts should be less and less obvious as size increases. (The exact opposite happens with John's images, since they are images of type, not actual type. I hope this is clear to everyone - it is the nature of the bitmapped image versus the nature of scalable type.) I say that I expect CT artefacts to decrease as size increases, and I do, but will that really happen?

Re increasing the density of the dot of the i, may I play devils advocate and ask if there isn't a point, in a high x-height font, where the i could then be confused with the l? The possible advantage of a weak dot on the i is that you might minimize this possibility.

Re customers with old equipment, MS must be calculating that those who buy Longhorn will primarily have LCDs with reasonable resolution (at least 1280x960 I would guess) and must have reasonable grounds to make this calculation.

How will the CTC types render on older screens using older versions of Windows?

And should we be worrying about that? Isn't the whole point of these types that they are not recommended for webpages?

What interests me more is the hint of a closed, proprietary nature of the type system - a holy grail that the type industry has been searching for ever since the unfortunate invention of PS and TT.

How does CT-style rendering contrast with Cool Type and Quartz when used with older CRTs?

Mais je divague. I don't want to lose track of my question: we have established that CT will render 7 consecutive i's in 7 different ways. But we also know that we will never normally have to view a text stream of 7 consecutive i's. However, the point is has been made: strokes are rendered differently, according to their position on the screen vis-a-vis subpixel positioning. What do we gain from this? What do we lose?

And another question: what is the processing overhead for CT? We know it is turned off by default in XP, but why? Turning off a great new feature by default is always a mistake, as GX showed. So what's the problem with CT that mandated this counterintuitive situation?

here is an example showing Natural Widhts ClearType v Subpixelpositioned CT. For reading I much prefer the more accurate spacing and placement of glyphs in the SPP case. There is a slight loss of contrast but the gain is more accuracy in letterfit or letterspacing, (see the word "subtleties" in both cases. I find the SPP to be much better spacing, find your own favorites :-) and a much more even tone or typographic color to the page. I dont notice the different stems when set in normal text, of course its more obvious when a line of "i's" are set together.

NWvSPP.bmp NWvSPP.bmp (1579.1 k)