Digitization Quality

raph's picture

I've been looking at digital revivals fairly closely, and am quite struck by the variation in quality. For example, here are three versions of the capital E from Centaur:

E from Centaur, digitization 1E from Centaur, digitization 2E from Centaur, digitization 3

I'm not going to say which is which just yet, because I think it's likely to color people's perceptions. It shouldn't be too hard to find signs of autotracing.

Please, typophiles, tell me what you think of these three digitizations, and don't hold back. I'll post the answers tomorrow.

raph's picture

Admins, could you change the title to "Digitization quality" please? I've been bitten by my browsers's auto-complete "feature" again. Thanks in advance.

heidi's picture

Bad

hrant's picture

The third one is crap, the first one is fine for offset in terms of reproducing the [relative] softness of letterpress, the second one is nice and crisp, which makes it better for smaller sizes (although the color is too light for that) and to me at least more attractive large as well.

hhp

heidi's picture

BTW I see only 2 samples

raph's picture

The answers:

1. My effort, working from a very high quality scan of the title page alphabet of "American Proprietary Typefaces".

2. Agfa Monotype's CentaurMTStd, OpenType edition.

3. Lanston Type Company's Metropolitan Roman No. 369.

I'm sure the soft corners are a matter of personal taste; I rather like them myself. But, just for fun, I'll make an alternate sharp version.

Giampa's picture

The criteria in judging the curation of historical typefaces is the inherent qualities preserved in letter shapes and their forms.

Unknown is the intended point size or "the source"? Centaur is a revival so one must choose accuracy over mechanics. Digitizing type face revivals is a not always a restorational process, not when the masters are available. I must say everyone has their own idea as to what is right and what is wrong. As long as they are agree with me then they are right too.

Lucky them.

In other words what appears to be the worst may actually be the best. I am sure of John Downer would have some comments worthy of this discussion.

Unknown is the intended point size or for that matter "the "source"? Also, since this face is a revival one must consider philosophy. It is a curational process. A process greatly misunderstood by the digital type community.

Interpretation, or "smoothing" is not curation, sorry.

Also there is this very strange habit by all to show illustrations at a greatly enlarged pt. size the biggest curational mistake made by other font developers when making comparisons to historical typefaces. Also you have not shown any comparison to even a printed example from "the original hot metal Centaur". Would that not seem important to you? Well, if not, revivals are not the business for you. They should be shown at the intended type size. In other words if it is based on 6 point you should pick 6 point in your menu to make "fair" comparison. Did any of you ever think about that? Most of you don't seem think at all, so I take that as a no.

Metropolitan E

Now I have no friggin idea how accurate JPEG's are, but as you can see, not very. This abysmal digital technology we have adopted has "cheapening" qualities unlike the wonders of ink on paper which when letterpressed is enriched. That said, the E on the right hand side of this JPEG is closest to intended point size. It would be best to compare output from page layout software in high resolution rather than these TV sets which are more appropriate to Gerry Springer episodes, or sometimes, these forum discussions.

But whatever you are showing could be compared with Lanston's Metropolitan shown here. With the exception of the top serif which ours is decidedly smoother, "your third example" appears to be the most historically accurate if it is intended for a text size. So who ever made that is at least working from a text based size. Centaur is a book face so that is a

raph's picture

As promised, a sharpened version of the E. While I prefer the softer rendition, there's an argument to be made for making the masters sharp. It's always possible to go from sharp to soft mechanically (with precise control over the degree of the effect), but the reverse direction is trickier.

My tracing of Centaur E, sharp corners

Giampa's picture

Are you suggesting "multiple master

hrant's picture

> Centaur is a revival so one must choose accuracy over mechanics.

But the question is, accuracy to what? Is it smart to reproduce the potentially unintentional "errors", that are furthermore restricted to one size in metal but not restricted in terms of size in digital?

Although I don't want to nitpick about what "revival" means exactly, my feeling is that an intelligent revival tries to get inside the brain so to speak of the original, going beyond the limitations of the tools and circumstances.

As for sharpness, it is indeed largely a matter of taste, but there's still the issue of "information loss", which goes only one way.

hhp

raph's picture

Gerald,

Yes, I would like to debate you on most of the points you bring up. I will start by posting my reference images.

Here's a 60 point E from American Proprietary Typefaces, beautifully letterpress printed, scanned at 2400 dpi then scaled to 744 dpi. The colorspace is calibrated to sRGB, but this is a grayscale-only image.

60-point Monotype Centaur E, scaled to 744 dpi

And here's an E from the text of the same book, 13 points, 2400 dpi scan. I've sharpened it a bit, but otherwise it's a pure scan.

13 point Monotype Centaur E, 2400 dpi scan

Giampa's picture

Raph Levien,

Your Lanston font appears to have been generated by one of our publishers, or very old software, or is a custom version of our font made by a dealer, or a customer.

A former dealer comes to mind, one that sells backdated font software. Where did you purchase it and when?

I presume you did purchase it.

clearly our Metropolitan E differs from your example.

Metropolitan E

Hrant,

> Centaur is a revival so one must choose accuracy over mechanics.

But the question is, accuracy to what? Is it smart to reproduce the potentially unintentional "errors", that are furthermore restricted to one size in metal but not restricted in terms of size in digital?

How do you know they are errors?

Your second part "furthermore restricted to one size in metal but not restricted in terms of size in digital?"

Interesting but not surprising you would choose 60 pt as a model for a text face? Hopefully no one else agrees with you other than Ralph.

At some point, if you finish your work you will have your optical scaling software completed becoming the standard. At that point would you wish for all font developers to redo their fonts to meet the coming new standards? I don't suspect you would be so mean spirited.

Choosing a 12 pt model would seem natural for many book faces so I am most surprised you would defend the novelty of 60 pt as the model for a text face. Well maybe not. Mistakenly I may have assumed you read books.

However, as you can see by both scans, which is what the designer was intending, to be printed on paper with ink, that the two examples in 60 pit and also the 13pt?? (good reason to doubt, Centaur was not cut in 13 point) are both rounded. So if modern methods were to accurately reproduce these faces should not these qualities be included?

Before you answer though. That is a loaded question.

raph's picture

Gerald,

To answer some of your questions: the GIF images I posted were rendered using Ghostscript to bitmap, then downsampled by a factor of 8, and a gamma of 1.8 applied. The final gamma step is to minimize the subtle jaggie-like effects that remain even when doing an anti-aliased rendering.

I believe the font technology is quite modern in all cases; the Agfa sample is in OpenType, while yours is from the PDF file posted on your website.

In any case, I seriously doubt that font technology has a significant effect. Yes, converting from one spline format to another can result in some loss, but among high-quality formats such as TT (quadratic Beziers), PS (cubic Beziers), and Ikarus (I've seen this described as a Hermite spline, but unfortunately no longer have my Karow book to check) the loss should be minimal. In particular, the most noticeable effect of quantizing to a 1000-unit grid should be slight deviations from tangent continuity at knot points.

BTW, I know my spline math pretty well. Lately I've been experimenting with Cornu spirals as a potentially better primitive for font design than Bezier splines, and have been cranking through quite a bit of math involving first- and second-order continuity, error metrics for fitting splines to complex analytical curves such as the Cornu spline, and so on. Just a warning in case you were thinking of getting into a mathematical pissing match with me :)

I absolutely believe in optical scaling, and absolutely agree that it's a mistake to trace patterns designed for large sizes and expect them to work when mathematically scaled down. In fact, I've been doing a rather careful study of the ATF books to attempt to reverse engineer the techniques the Bentons used to achieve optical scaling, and intend to write up my findings soon.

That said, I do not believe your methodology of digitizing pattern plates, even those designed for text sizes, is sound. I believe, if one is aiming for accuracy, the goal should be to make the end product match the original as closely as possible. The intermediate steps may shed light, but ultimately what matters is the patterns made by pigment particles on a piece of paper.

In particular, your beloved pattern plate goes through two important steps before becoming a printed page. First, it is engraved into a matrix. The cutting tool used in this process has a non-zero radius, so the resulting matrix does not have exactly the same shape as the pattern plate. The Bentons were absolute masters at controlling these parameters to achieve good scaling, rather than just treating the engraving process as a degradation of the Platonic ideal represented in the pattern plate.

The second important step, of course, is transferring ink from the type surface to the paper. Understanding this process down to the micron level is hugely complex, involving as it does the physics of the viscosity of the ink base, capillary action of the fibers of the paper, and so on. Then, the appearance is further affected by issues such as light scattering below the surface of the paper. These more subtle effects are a large part of the reason why offset printing, no matter how fine, cannot reproduce exactly the look and feel of letterpress, and why letterpress still has a niche in book arts. Again, I've studied these effects fairly carefully as part of my research into halftone screening, just in case you were wanting to get into a pissing match about that.

That said, it's not that hard to make an approximate model of letterpress printing. The main effect is gain, and beyond that there's a subtle softening effect, as well as of course an edge roughening as the ink interacts with the texture of the paper at the roughly 100-micron level.

Metal type designers know all this and take it into account. If gain were the enemy, they'd cut matrices with thinner and thinner strokes as the size went down, to compensate.

To turn your questions back at you: is it historically accurate to print text using a modern, extremely high resolution offset press or laser printer, using outlines digitized from pattern plates? To me, the answer is clearly no.

Lastly, I believe you have poor quality digitizations of these masters. One of my main metrics for evaluating this quality is the variation in curvature; high curvature variations are characterstic of "bad splines", especially generated from autotracing. A particularly striking example is the terminal of the middle stroke; while the lower half is relatively smooth, the upper half has a long flat section bracketed by two tight turns. I see no evidence that this is intended by the original designer, borne out in any printed samples of metal Centaur, or that it's an aesthetically defensible improvement over the original design.

To return to the issue of scaling, we now live in an era of postmodern typography. In the days of metal, the scaling adjustments were locked into the metal when the matrix of a particular size was cut. These days, while sophisticated designers choose digital fonts with optical scaling appropriate to the size, it's far more common for one size to fit all. Even when optical sizes are available, I'm sure it's not unheard of to use a "text" scale in a larger size to create a softening effect, or vice versa. When you use an outline in a large size, the curves must be smooth. Considering that the original metal rendition of Centaur was for the Metropolitan Museum for use in titling, I'm sure you can appreciate this.

When reproducing in small scales, the details of curves are less important. What becomes more important is the robustness of the thin/thick ratio, openness of counters, etc. In your "Metropolitan", I see little or no evidence of such adjustments; in particular, your stroke contrast and serif design appear no more robust to me than either the Agfa version or my tracing based on 60 point metal originals.

Yes, I believe in adjustable parameters for fonts, including gain and sharpness in emulation of letterpress printing. I'm not convinced that Adobe's multiple master technology is the best way to accomplish this parameterization, for detailed technical reasons. And no, I do not work for Adobe.

Lastly, your pretending that the name "Metropolitan" is the true name for Centaur bugs me. Show me evidence that Bruce Rogers ever referred to the font under this name. Otherwise, explain to me how your choice to release your font under this name differs from macXware's release of a Hobo knockoff under the name "Bobo".

raph's picture

The 13 point estimate was based on fitting text size to a modern PostScript typeface, so you're almost certainly right that I've got it wrong. Does anyone with Pankow's book know the true size? Yet another hazard for attempts at authentic reproduction.

Gerald, the "E" I posted is reproduced from your specimen sheet PDF. I believe this counts as fair use, but if you disagree I will send you a check even though I have no intention of ever using your font. I disagree that your "Metropolitan" E differs from the example I posted; rather, I believe it's more likely that my rendering process provides a clearer view into the subtleties (and thus defects) of the outline. Yours is only a 4x supersample without any compensation for display gamma; it's not surprising that the antialiasing artifacts drown out subtle variations in curvature.

hrant's picture

> How do you know they are errors?

You don't. Neither do you know they're not.
You use your brain to decide. Instead of blind, literal faith.

> you would choose 60 pt as a model for a text face?

Here we go again.
OK, let me try my hand at this free-form revisionism:
"You say 60 point? That's pretty ironic, coming from a person who grew up on Jupiter."

--

Ralph, I agree strongly with almost everything you wrote, except:

> If gain were the enemy, they'd cut matrices with thinner and thinner
> strokes as the size went down, to compensate.

I fell into this trap myself a few years ago. I thought that because gain is absolute, smaller sizes needed to be thinner. This is not untrue, but the optical necessity for the opposite is much stronger.

hhp

dan_reynolds's picture

Ralph, comparing Gerald Giampa and his Lanston operation to macXware isn't fair.

Even if his digitalization is as poor as you say it is, it isn't piracy* against Rodger's Centaur, or against AGFA Monotype (I'm presuming that AGFA owns all of the applicable copyrights on Centaur

Giampa's picture

Ralf,

Hrant says your are wrong, do I have to tell you also?

hrant's picture

Gerald, which of your fonts feature optical scaling?
I mean instances of the same typeface for different masters, not merely size-aware single fonts.

BTW, Raph, you should know that Gerald is a poet - rationality doesn't go far with him.

hhp

raph's picture

Hrant: I get the feeling we're both trying to say the same thing here. Modern high-resolution printing has, for all practical purposes, zero gain. It would be possible to compensate for gain, reproducing the effect of zero-gain printing, by making strokes of smaller sizes even thinner (in ratio to the unit em) than larger sizes. But optical scaling means making strokes thicker (relative to em, and thus lower thick/thin contrast) in smaller sizes. That the Bentons chose to do the latter rather than the former is strong testimony on which way is "right". That said, probably 95% of the typographic work done today is done with no regard for optical scaling.

Dan: I did not intend to imply that Lanston is pirating fonts, while it is pretty clear that macXware is. I apologize if I have given anyone that impression.

John Hudson's picture

An aside regarding Ikarus:

While it is indeed the most accurate method of digitising from analogue artwork, there are two big caveats:

1. Ikarus is only as good as the operator, both in terms of fidelity to the original artwork and, crucially, in producing outlines that can be cleanly converted to other outline formats.

2. While Ikarus itself was an excellent archival format for very high resolution outline digitisation and storage, the ability to generate from that data good quality outlines in other formats depended on the quality of the conversion and, and as noted above, the way in which the Ikarus digitisation was done. A good Ikarus operator know how to use the tool to produce results that could be cleanly converted to bezier outlines for PS fonts, even taking into account limitations of specific conversion routines.

Over the years, I've been given source bezier outlines for various projects, from large and respected foundries and, in many cases, converted from Ikarus data made by URW themselves. Very often the outline quality is very bad, and much work needs to be done to clean up the outlines. It is often difficult to tell how much of the problems are due to human sloppiness and how much to bad outline conversion. Even if the original Ikarus data accurately represented the analogue artwork, that is only part of the process of getting from that artwork to a clean font in another format.

Below is an example of the kind of thing I saw quite often. I can assure you that Adrian Frutiger's drawings didn't look like that.
Poor Ikarus to PS conversion

raph's picture

Gerald: I respect your work as a printer; your website and specimen pages show off your work to beautiful effect. I'm certainly not trying to imply that you're doing anything shady. It's not my intent to get into a big flamewar with you; with luck we can keep this conversation civil and hopefully learn a little something from each other.

Is that good enough?

"Considering that the original metal rendition of Centaur was for the Metropolitan Museum for use in titling, I'm sure you can appreciate this."

Can you say that again. I am missing the punch line. What is your point?

According to Herbert Johnson's article in American Proprietary Typefaces, the font known as Centaur was first cast for production by Western Type Foundry in Chicago, in 12- through 60-point sizes. These fonts were titling capitals only, and licensed to the Metropolitan Museum of Art for their exclusive use. The Museum printed a number of posters using this font, but copies seem to be rather rare today.

The punchline is that the original font was designed to work well in sizes up to at least 60 points. At text sizes, subtle variations in curvature are not visible, but at display sizes they are. I am sorry to criticize your work (and perhaps I do not have the bragging rights to do so), but I believe that your digitization is inferior, for the purpose of display text, to the design that BR designed and seems to be faithfully reproduced in my printed sample.

What would you call a private face for the "Metropolitan Museum of Art".

According to Johnson's article, the name given at the time was "Museum Caps." Johnson also quotes a letter from BR musing on the choice of name, and noting that William Morris often named his fonts after the books they were designed for.

I'm still very curious to know exactly where the name "Metropolitan" came from. When was it first used? You seem to be in a unique position to shed more light on these questions.

My scans are of the book "American Proprietary Typefaces", set in Monotype Centaur and printed in 1998 by the Stinehour Press in Chicago. The text is letterpress, as is obvious from the bite into the paper. It is not a digital source.

I want to be very, very clear about my sources and processes, and absolutely do not want to step on anyone's property. My main purpose is to learn the art of type design, not to make money. Of course I have looked at the Agfa version and yours. That is the best way type designers learn - by looking carefully and critically at the works of others.

The outlines I have drawn are traced solely from the printed sample cited above. All knots and control points have been placed by my own hand. In the first version I posted, the corners were drawn round to carefully match the original scan. In the "sharpened" version, I relied on my eye to square the corners.

I'm sorry we have this confusion about 60pt vs text sizes. Since I'm tracing from a 60pt source, I obviously make no claim about its suitability in text. You seem to be saying that your Metropolitan is based on a 12-point pattern, and may not be suitable for display use. I am most emphatically not trying to say that reproducing the 60pt outlines in text sizes is a good idea (even though that's what most people actually do with "revival" fonts). It's entirely possible that there's no dispute here at all, despite all the voice-raising.

Since you seem so intent on dueling, may I propose a challenge? Choose a letter or fleuron from a source which has not yet been digitized, ideally from the ATF 1912 or 1923 books (since I have both and because of the masterful printing), and then let's both post our outlines. The Typophile community can then judge the respective quality of our work, based on faithfulness to the original, smoothness of curve, and sharpness of detail. I'd consider it especially fun to draw an optically scaled series - I'd be happy to produce that in Multiple Master format. If you win, I promise to acknowledge Metropolitan as the definitive digitization of Centaur, and not to release any version of Centaur for the next 10 years. If I win, you have to buy me a beer.

Deal?

raph's picture

<font class="dontLookLikeCrap">John: thanks for the additional notes on Ikarus. I had the 1987 Karow book shortly after it came out, and I do not know what happened to it (probably given to the library when I moved to California). Now, of course, I could kick myself.

Anyway, I had thought it was impossible to find information about Ikarus online, but I was able to track down Karow's 1991 paper, Digital Punch Cutting. Clearly, Ikarus was good stuff, and equally clearly, it is possible to get into trouble when converting between outline formats. PostScript (Type1 on 1000 unit em) works best when knots are placed with economy.

But I stand by my basic assertion that conversion between font formats is a quantitative problem, and that when done right, the mathematical error is below any reasonable tolerance for visible artifacts.</font>

John Hudson's picture

But I stand by my basic assertion that conversion between font formats is a quantitative problem, and that when done right, the mathematical error is below any reasonable tolerance for visible artifacts.

Cubic bezier to quadratic conversion remains a practical problem in a number of ways, even if the conversion is accurate. The only way to get really close curve fidelity seems to be to use a large number of off-curve points in the quadratic, which is what FontLab does. Even then, I sometimes find the curves off in ways that I find unacceptable and which require manual editing. But there is also a problem with having too many off-curve points on a short quadratic curve in a TT font that one then wants to delta hint. The multiple points frequently cause spikes during hinting, and one ends up having to fiddle about delta'ing lots of points or trying to apply middle-deltas in situations where a final-delta is appropriate.

If we're hinting for ClearType, this tends to be less of an issue. But when we were doing more black and white hinting we used to spend many, many hours manually cleaning up converted outlines. And, of course, the number of points affects the file size.

raph's picture

Some more thoughts.

I agree that it's relatively easy to convert IK data into high quality TrueType. It's certainly possible to convert it into high quality Type1, but unfortunately if done in the most straightforward way, you'll see problems like Figure 19 in the Digital Punch Cutting paper.

I think I understand the issues in generating good quality Type1 from high resolution curves. It's not hard to detect when you're introducing tangent continuity problems by quantizing control points to the unit grid. To fix these problems, there are two basic choices. First, you can use the Type1 "div" operator to specify the control points at higher resolution than the em grid (in fact, by representing the coordinates as rational numbers, it's possible to guarantee exact tangent continuity). Second, you can start taking knots out. Roughly speaking, if you double the number of knots used to represent a high resolution curve, you're also doubling the tangent continuity errors. But how do you know which knots you can take out without distorting the curve itself? It's a tricky numerical problem. Unfortunately, it's one you have to deal with if you want the fonts to be compactly encoded - the div trick certainly has the potential to bloat up the font file sizes.

Earlier I wrote:

"That said, I do not believe your methodology of digitizing pattern plates, even those designed for text sizes, is sound. I believe, if one is aiming for accuracy, the goal should be to make the end product match the original as closely as possible."

I'd like to clarify this statement a bit. One of the most important things that must be done when optically scaling to small sizes is adding gain. So what I question is the idea that it's a good idea to reproduce the original pattern plates without adding gain somewhere in the process. Unfortunately, that's probably what most users of the font are going to do.

That said, I'm sure that if you were to, say, make a photopolymer plate from text in 12pt Metropolitan, then print it on nice soft paper, the results would be absolutely gorgeous, probably directly comparable to letterpress set from Monotype Centaur. But how many people are really going to do that?

Adobe's philosophy of optical scaling, by contrast, builds the gain into the outlines of the fonts. That way, when you print the text on a low-gain, high resolution device, you do at least get the gain. I think, for the majority of users, that's preferable.

So I think that digitizing the pattern plates is an excellent idea, as long as you have the gain built in somewhere in the process.

Thomas Phinney's picture

Raph wrote: "The text is letterpress, as is obvious from the bite into the paper. It is not a digital source."

Although one would like that to be conclusive, it is possible to make photopolymer plates from digital work, and print it letterpress. It's not quite the same, but close enough to fool most folks.

Cheers,

T

Giampa's picture

Thomas,

Alas, we agree.

Steinhour press, and our forum friend Gerald Lange are both famous for using digital output to polymer for letterpress printing. That is the stated production preference of Steinhour Press. I am working with them on several projects. Besides the likely hood of 13 pt Centaur hot metal is remote if not impossible.

If you want, I can ask Steinhour Press. Any wagers? My bets go on digital Centaur.

"American Proprietary Typefaces"

By the way Thomas, this may be a stupid question, why would Centaur be considered an American Proprietary Typeface?

2. Exclusively owned; private: a proprietary hospital.

3. Owned by a private individual or corporation under a trademark or patent: a proprietary drug.

Shouldn't they have used Metropolitan? After all, Centaur is English.

Thomas Phinney's picture

"why would Centaur be considered an American Proprietary Typeface?"

I don't know. It would be more accurate to say that Centaur is descended from a proprietary American typeface. But then, I don't have that book, so I can't say what their criteria were....

Cheers,

T

hrant's picture

Gerald, I guess the answer to my question is: none.

But that still doesn't explain why you constantly state that Goudy was a buffoon hack.

> may I propose a challenge?

That can't work, because one of the most imortant elements of design, practicality (eg doing something quickly enough for it to be cost-effective) would be missing. It becomes a challenge of digitization skills, which is a mere subset of the more important design judgement.

> One of the most important things that must
> be done when optically scaling to small
> sizes is adding gain.

That doesn't sound right. One should:
1) Add weight (for optical reasons).
2) Trap (to counteract filling-in due to gain).

BTW, I'm a firm believer that the tools do indeed affect the results. Maybe not during digitization, but certainly during design. And I also believe that a good designer can greatly overcome the tools (be it the bezier or the broad pen), because the purpose of type (especially text type) is the help the reader, and the tool is circumstantial to him.

----

BTW, Typophile is seriously messing up on email delivery; I never got about half the posts in this thread.

hhp

raph's picture

The book uses word "proprietary" to mean specific to one printer (the proprietor), rather than the more modern meaning of protected by intellectual property law. Pankow gives the criteria for inclusion as:

* Designed by or under the direction of an American
* Produced after 1892
* Intended for composition in metal
* Originally limited to the used of an individual or press, though such a press need not have been private

I do not agree that a trademark is always a true name.

I read the Digital Punch Cutting paper more carefully, and found the interesting statistic that, in URW's work with Ikarus, 10% of hand-digitized points were found to be outside error tolerance, and had to be fixed up. I'm very curious what the corresponding statistic was for Giampa's curational efforts in Metropolitan.

raph's picture

Hrant:

I should have written "add weight" rather than "add gain". Adding weight is necessary for optical scaling, and simulating letterpress gain is but one way to do it.

As for the trapping, I'm less convinced. Depending on the way you add the gain, you might not get significant filling-in. In fact, if you convolve with a circular pen (which is basically what the Benton matrix engraver does), you leave inner corners completely sharp.

I can demonstrate:

demo of gain, image 1
demo of gain, image 2

The E outlined in blue is the original outline. To its left, I've applied a bit of negative gain. The three E's on the next line have increasing amounts of gain, all applied using convolution with a circular pen. Note that I did not do any nonuniform scaling here, so the bolder versions appear visually a bit condensed. The Bentons, masters that they were, used a combination of gain and stretch to scale designs to smaller matrices.

You can also add softness independently to outer and inner corners. Here's an increasing softness radius applied to outer corners:

softness demo, outer corners

And the corresponding effect applied to inner corners:

softness demo, inner corners

All these tweaks were done in the bitmap domain, but I'm coding up the effects in outline space (much harder, but I'm feeling up for a challenge).

hrant's picture

> simulating letterpress gain is but one way to do it.

No, ink gain is not in the same dimension as the weight gain that's optically necessary.

> if you convolve with a circular pen (which is basically what the Benton
> matrix engraver does), you leave inner corners completely sharp.

No, because the cutter is a rotating thing, it can never make corners as sharp as the hand graver. Jim Rimmer has attested to this on Typophile. Gerald will fume that 1/1000 of an inch is plenty fine enough, but anybody with a grasp of the extreme acuity of human vision and the difference between what a layman thinks he sees versus what he actually processes visually (not to mention an abstract desire for fidelity/quality on the part of the type designer) knows that's not true.

> To its left, I've applied a bit of negative gain.

That can't be negative gain, because the inside corners are less sharp - look at the join of the middle arm to the stem. If you do find a robust algorithm for negative gain however, do let me know! That would be valuable.

--

In your samples, you're exhibiting rounding on the outside corners but sharpness on the inside ones. This is basically because you're actually working with a "negative" image. What I'm getting at is that you can indeed choose whether to preserve sharpness for the inside or the outside corners (depending on whether you're working with a positive or negative image), but not both. The pantograph/rotating_cutter cannot match the graver in sharpness - period. Also, when there's a big difference of sharpness between insides and outsides, you're losing design integrity (unless you happen to want that effect for some reason).

hhp

raph's picture

Hrant: are you confusing inner and outer corners? By 1912, ATF was using the Benton matrix engraver to cut matrices directly (previous incarnations of the pantographic engraver were used to cut punches). Thus, the larger the engraving tool, the more area cut out, which means more area at type height when the type is cast.

So if the engraver drills a single perfectly round hole, the resulting cast type will be a round dot. If the path of the tool is a perfect right angle, the resulting outer edge will be a rounded corner, with the radius of curvature on the rounded part equal to half the diameter of the cutting tool.

But if the engraving tool traces a plus shape, the raised shape should have perfect sharpness at the intersection. You might get some softening, of course, when you ink it and apply it to paper, but in any case you expect inner corners to be sharper than outer corners. If you were engraving the punch rather than the matrix, you'd expect it to be the other way around.

That's the theory, anyway. Let's see how it looks for real:

4800 dpi scan detail, p75, atf 1912

That's a 4800 dpi scan from page 75 of the ATF 1912 book, showing a detail of the "M" from 18 point Bodoni Book. I am awestruck at the sharpness of the inner corner here - by eyeball, I'd estimate that the effective radius of curvature is around 5 microns (about a quarter mil).

Applying negtive gain is equivalent to drawing the white areas with a circular pen. Thus, you expect outer corners of the white areas to gain rounding. Indeed, that's what you see in the first E I posted above.

I agree that a big difference of sharpness between inside and outside corners is a design flaw. My study of the Benton optical scaling strongly suggests that they added a tiny amount of deliberate softening to minimize this effect. In all cases I've analyzed so far, the pattern plate was designed for half the maximum size intended to be cut, which is theoretically very near the optimum compromise point. If you scale up much beyond twice the pattern plate's design size (as was done to some extent in later specimen books), then you get an inner roundness not matched by the outer sharpness.

I'm using a circular pen here, but it's possible to apply "miter joins" (in PostScript lingo) to corners that are identified as such in the outline. That results in both sharp inner and outer corners. The main tradeoff is the extra care needed to correctly classify corners - if a sharp corner is mistakenly drawn as a very tight curve instead, the inner corner would be soft instead of sharp. I'm not much of a fan of hyper-sharp scaling, though, and am perfectly happy to accept some softening, especially if it buys me stability with respect to subtle variations in the outline. Of course, that's a matter of taste too.

hrant's picture

> I am awestruck at the sharpness of the inner corner

But the outer ones are crap. That's because it's working "negatively" - it's cutting from inside the black, so to speak. If you cut from inside the white instead (by cutting a punch and not a matrix for example), the polarity of the quality will be flipped. Considering that gain builds up more on the inside corners, I'd say that cutting inside the black is less bad. But the point is you can't have both with the pantograph. That's why quality-minded people hand-sharpened the output from the pantograph. I once posted a scan of a photo of a bank of people doing just this at the Tetterode foundry. But the "spirit" of the pantograph eventually got rid of such refinements (including optical scaling, mind you, which the Benton boys tried -but failed- to convince their users to implement).

> happy to accept some softening

More than that already provided by gain?
I mean in letterpress - in offset it's a moot point since the fonts are necessarily digital.

But yes, I guess it is a matter of taste - I'm hoping though that sludge went out with the 70s.

hhp

John Hudson's picture

One of the most important things that must be done when optically scaling to small sizes is adding gain. So what I question is the idea that it's a good idea to reproduce the original pattern plates without adding gain somewhere in the process.

One thing to bear in mind is that the preference of individual letterpress printers ranges from a 'kiss' of the type on hard papers, with the minimum of type necessary to create a black image, to deeply embedded and heavily inked type on soft rag. So creating a faithful revival that in some way compensates for gain is likely to be influenced by a) the preference of the printer of the source material (especially if only one source is employed) and b) the preference of the digitiser in selecting printed sources.

While the examples I have of Gerald's letterpress printing have a fairly deep impression, he uses ink sparingly. When I posted some photographs of the 1924 Caslon specimen, he criticised them as being over-inked. He may consider this an objective criticism, but I'm inclined to think it is a subjective one, since the printer of the 1924 specimen was George W Jones, who presumably knew what he was about.

hrant's picture

Deep impressions are generally provincial and unprofessional.

hhp

Giampa's picture

Hrant,

No, because the cutter is a rotating thing, it can never make corners as sharp as the hand graver. Jim Rimmer has attested to this on Typophile. Gerald will fume that 1/1000 of an inch is plenty fine enough, but anybody with a grasp of the extreme acuity of human vision and the difference between what a layman thinks he sees versus what he actually processes visually (not to mention an abstract desire for fidelity/quality on the part of the type designer) knows that's not true.

Jim says what Jims knows. What he does not know is Bentons, Jim does not have one, never worked on one. Kent Lew will attest to the fact that a Benton will cut razor sharp corners and has made that very statement in the Typophiles. The Benton has a totally different system. The lower (work is upside down) the cutter the smaller the radius of the cutter surface. In other words the cutter shaft is not the issue, the point is. Razor sharp.

This conversation will be lost on you. You know nothing of this technology and will never allow yourself to learn. You are a waste of air space in that area of type expertise.

Otherwise you give very good recommendations on tea, and I thank you for that.

Giampa's picture

Raph Levien,

I agree that a big difference of sharpness between inside and outside corners is a design flaw. My study of the Benton optical scaling strongly suggests that they added a tiny amount of deliberate softening to minimize this effect. In all cases I've analyzed so far, the pattern plate was designed for half the maximum size intended to be cut, which is theoretically very near the optimum compromise point. If you scale up much beyond twice the pattern plate's design size (as was done to some extent in later specimen books), then you get an inner roundness not matched by the outer sharpness.

Can you be kind enough to explain this in plain English. For instance, can you indicate it in points and picas or in measurements so I can follow your point?

Giampa's picture

John,

Deep impressions are generally provincial and unprofessional.

hhp

Is that what you are saying John?

raph's picture

Hrant wrote:

More than that already provided by gain?
I mean in letterpress - in offset it's a moot point since the fonts are necessarily digital.


I should have probably stated up front that my main interest is in digital fonts for high resolution, effectively zero-gain reproduction. In fact, what really excites me is computer displays in the 200dpi and above range. At that resolution, you finally have enough pixels to properly reproduce serifs and the like, but of course it's all very different than printing. With LCD and OLED technology, gain is exactly zero.

I'm not saying that my goal is to fool people into thinking that they're looking at a letterpress printed page, but at the same time I think it's possible to capture some of that quality. If you just take digitized pattern plates and reproduce those with zero gain, text will look extremely thin and spidery compared to real printing made from the same designs.

The issues between 200 dpi displays and offset printing are similar but of course not identical; in particular, high resolution offset has very little gain. On the other hand, offset probably is much more sensitive to things like rounded corners.

But yes, I guess it is a matter of taste - I'm hoping though that sludge went out with the 70s.

I am not talking about bringing back the days of American Typewriter or Benguiat Gothic. The ATF sample I posted above has an outer rounding radius of around 1 mil. It only looks like "crap" because I posted a 4800 dpi scan; depending on your monitor, that's something like a 50x blowup. Trust me, it looks plenty sharp at actual size.

John: I definitely appreciate what you're saying about variations in printing style, paper, and so on. The ATF 1923 book helpfully includes bound-in signatures printed on various luscious papers, in addition to the rather hard enamel surface of the main text. In many cases, it's actually possible to do a direct head-to-head comparison with the same printer, same font, just different paper.

That said, there's something else notable about the ATF 1912 specimen book: the gain from the letterpress printing process is quite small, astonishingly so considering the modern conceptions of printing from that era. There's a specimen of a hairline font (Light Litho Gothic, if memory serves) which definitively establishes that the press gain was less than three quarters of a mil, and almost certainly less than that.

Thus a rather startling conclusion: most of the weight adjustment for optical scaling was cut into the metal. Only a fraction of it was a consequence of press gain. To me, it doesn't really matter how much the relative contribution of each is. It's the final printed result that matters.

Even so, under the assumption that the printing in the ATF books was an amazingly faithful reproduction of the metal type, it makes sense to trace outlines from high-quality scans made from the books. Indeed, based on my experiments so far, it seems quite satisfactory.

Of course, for Centaur, it may well be that none of the samples available have the technical near-perfection of the ATF 1912 book, and that going back to the metal type or the original pattern plates would reveal qualities not immediately visible on the printed surface. I've got a couple more definitive sources on Centaur on order through abebooks, so I should be able to get a better handle on these questions.

Oh, and if the American Proprietary Typefaces book is based on digital sources, how would one explain the stylistic differences between the 60pt and text samples? Maybe Stinehour Press has access to a secret digital version with optical scaling, while the rest of us have only the single-scale version. Hmmmm...

Thomas Phinney's picture

There's a specimen of a hairline font (Light Litho Gothic, if memory serves) which definitively establishes that the press gain was less than three quarters of a mil, and almost certainly less than that.

Hmmm. I would have guessed that press gain might be less for a really thin hairline than for a thicker stroke. Anybody care to enlighten me on that question?

T

Giampa's picture

Ralf,

I know this is like bringing a coal to Newcastle.

hrant's picture

> Razor sharp.

And I'm Nefertiti.

> Lead type is far more flexible than
> polymer because you can makeready
> under any individual lead character.

If you believe that lead typography is more flexible than
digital setting, then that explains why you hate Goudy.

BTW, I thought you were a "master printer"? A few weeks ago I made the mistake of wanting to makeready under the type, but Gerald Lange (who is a master printer) explained that it's much smarter to increase the makeready on the drum itself.

And based on what you just wrote, I can tell you know next to nothing about the mechanics of photopolymer printing. You just hate change, because you no longer have the energy to learn. You're stuck celebrating the past. The leaden, inflexible, rusting past.

hhp

Giampa's picture

Hrant,

BTW, I thought you were a "master printer"? A few weeks ago I made the mistake of wanting to makeready under the type, but Gerald Lange (who is a master printer) explained that it's much smarter to increase the makeready on the drum itself.

So if Gerald Lange explained it to you, try explaining it to me?

I am all ears. Ears have drums, Ringo Star has drums, cars have drums, presses do "not" have drums. I am not following you.

Please tell me where to look for the drum on a press. I hope it is not important because I never got to use it in my 40 years of experience.

Giampa's picture

Hrant,

The reason I hate Goudy is because he is better than you anybody else. It pisses me off big time.

Giampa's picture

Hrant,

Gerald, which of your fonts feature optical scaling?

For the last time.

http://www.lanstontype.com/Lanston-Trademarks.html

hrant's picture

Goudy is also very DEAD. Get over it, more forward.

BTW, if you're ever in LA (or in the vicinity of a press with me there) I'll show you exactly what I mean. It's really pretty simple. No need to cut tiny rectangles of tissue (1 mil) and place them exactly under each sort. Sheesh.

> http://www.lanstontype.com/Lanston-Trademarks.html

I don't feel like deciphering that.
Just tell me which of those designs has multiple fonts for different sizes.

hhp

Thomas Phinney's picture

Gerald,

That page certainly doesn't answer the question. There is no information there about optical size variants.

Also, I am rather surprised to see Lanston claiming trademarks on a lot of old ATF typeface names, such as Globe Gothic, Wedding Text, and Cloister Black. How is it that you would own these names?

T

Giampa's picture

Thomas,

"That page certainly doesn't answer the question. There is no information there about Thomas, optical size variants.

Also, I am rather surprised to see Lanston claiming trademarks on a lot of old ATF typeface names, such as Globe Gothic, Wedding Text, and Cloister Black. How is it that you would own these names?"


I think the size of the list may indicate I have better things to do with my time. Maybe some of you don't.

On the other matter of your concern, read the fine print.

Giampa's picture

Something is screwy with the copy paste function.

Sorry Thomas.

Hrant,

Goudy is also very DEAD. Get over it, more forward.

Life is short, art is long.

On the other matter of optical scaling. I advise that you draft an air tight contract with Adobe before you waste your time on software that will not be "application friendly", or made redundant at a whim.

First of all, if it is a good idea, which means in Adobes terms, a market for it, they don't need you, they really don't need you. Mind you that never deterred you before. In fact with the remarks you have made to our good friend Thomas, they really don't need you. Ever heard of manners?

Then, and only then will I help you. Otherwise it is just a waste of my time. Bin there, done that.

Does Gerald Lange know that the outside tracks of a music record travels faster than tracks closer to the axis?

John Hudson's picture

Hrant: Deep impressions are generally provincial and unprofessional.

Gerald: Is that what you are saying John?

No, what I thought I was saying very clearly is that I've seen too much variance in depth of impression, amount of ink, density of paper, etc. among the work of good printers clearly in control of what they were doing, and producing attractive results, to ascribe to anything other than preference or fashion. There very clearly is such a thing as underinked type, and such a thing as overinked type, that most people in the trade will recognise as such. What there is not is agreement about what constitutes perfectly inked type.

Regarding depth of impression, particularly, much seems to depend on the paper. If the paper is thick with a deeply textured surface, you'd better have a deep impression because otherwise you run the risk of letters breaking up on the surface. That's just common sense. Of course, selecting paper is also a matter of preference and fashion. [Unless, of course, the client comes along with giant sheets of recycled grey blotting paper with bits of used bandaids and broken staples in it. Remember that stuff Gerald?]

Giampa's picture

John,

Remember it too well. I didn't relish my smashed fome when I printed on some workers lunch that had been disgarded on the dandy roll.

Syndicate content Syndicate content