Re: Sample inkjet printouts

[Rebecca and Rowland <rebecca@astrid.u-net.com>]

At 1:03 pm +0100 30/3/99, Berthold Horn wrote:
>Hi:
>
>At 20:28 1999-03-23 +0000, Rebecca and Rowland wrote:
>
>>[snip]
>
>>btw Berthold - have you received my test prints yet?
>
> Thanks for sending the sample printouts.  I must say that at first I
> thought you might have mislabelled them, since to my eyes they do not
> prove your point. In fact, to me the PK version of CMR12 looks "choppier,"
> with not as  even grey as the Type 1 version of CMR12.

It's easier to read to my eye, and looks more like the definitive versions
printed in Knuth's book.  You're quite right about the evenness of the
grey, but what's that got to do with anything?  I'm comparing the printouts
not with a subjective aesthetic standard, but with a definitive printed
standard in Knuth's books.

>  Part of the
> difference may be that  the Type 1 version is suppressing overshoots at
> this size, while the  PK version seems to render rounded glyphs larger
> than the flat-topped glyphs.

Hmm...  my main point is that the Metafont version looks more like the
Computer Modern founts as printed in Knuth's books than the PS Type 1
version.   One can argue aesthetics until the cows come home and get
nowhere, so I've ignore the aesthetic argument entirely: compare the
printouts with the definitive Computer Modern Roman in Knuth's books.  You
should see what I mean immediately.

> Since we obviously have some biases here, the only sensible thing I
> can see to do is to scan the pages in and put them up on a web page
> somewhere for people to look at.  It would be nice then to use e.g.
> cmr10 at 10pt rather than cmr12 at 12pt, since the latter introduces
> a confounding difference in round-off that has little to do with
> "quality" (see below).  The obvious difference in x-height rounding
> in CMR12 at 300dpi obscures any other differences there might be.

But this is *exactly* the sort of quality point I've been talking about!
Of *course* you'd expect ATM and Metafont to produce similar results in
limited cases.  When you're working at medium to low resolution, such
rounding problems are inevitable, and are a very serious quality issue.  If
ATM is such a good rendering engine, it ought to be able to handle all the
borderline cases at lower resolutions excellently.  Since it can't, it's
not as good as some people claim.

[snip]

>> Things to note: the large sizes are identical no matter what does the
>> rendering, which implies there's nothing wrong with the founts.
>
> or more accurately: "which implies there's nothing wrong with the OUTLINES"

Hmm...  Okay.

>> The page of CM/PS at 12pt obviously has an excessive ex height.
>
> Let's instead say that the apparent x-height of the rasterized Type 1
> fonts is larger than that of the bitmaps derived from the Metafont
> code. This doesn't mean that one of them is "wrong".  It's a difference
> in how they are rounding off to the discrete grid.
>
> Note that 12pt at 300dpi is 59.775.. pixels.  So a x-height of
> 0.430555.. em corresponds to 21.448 pixels. Apparently the alignment
> zone at the x-height is causing the rasterizer to round this level
> upward in the Type 1 font while Metafont rounds it downward. CM at
> 12pt is where this difference shows up most noticably.  You don't see
> this with CM at 10pt.  This has been noted before, and I suspect is
>why you chose that particular size and resolution.

I chose that particular size (not resolution - that's fixed) because it
shows up more problems than other sizes on my printer.  Of *course* I
showed you the worst case.

>I find the Type 1 version more "open" and more  readable.  It is only
>"wrong" if you assume that it should round off  to the underlying grid
>in exactly  the same way as the Metafont version does, which is not
>possible given the very different way the grid-fitting happens.

Your aesthetic judgement isn't relevant: what matters is how well the two
printed versions match the definitive printed versions of the Computer
Modern founts in Knuth's books.  Do that comparions.  What do you think?

>> The problem is clear when you look at the test page containing the word
>> `French' repeated at various sizes:  The `Fr' combination runs
>> together at several sizes of thge CM/PS founts, while there's a clear
>> gap between the two letters in the Metafont versions.
>
> Well first of all, notice the extreme kerning (KPX F r -81.595) which
> brings the serif on the left of the r right underneath the descending
> serif on the tip of the F.  Then, the bottom of the serif on the F
> comes down to 458/1000 em, while the top of the r rises to 441/1000 em.
> So the gap is 17/1000 of an em or 0.204pt at a font size of 12pt.
> You don't have to make the font size very small before that is a pixel
> or less.  It is not a good idea to design a font with such close
> approaches, since it dramatically limits what adjustements are allowed
> in grid-fitting.

Erm...  Surely that means it's only a bad idea if you don't pay enough
attention to detail when writing the code to produce the fount?

> Apparently Knuth arranged for the grid fitting in
> the Metafont code to take into account the Fr kern pair.  The Type 1
>code for F does not have any adjustments related to the Type 1 code
>for r.

In other words, it's a good idea to do what Knuth did if you want a fount
like that and can design the code to deal with the matter.

> As a matter of interest, I do find some collisions in the cover letter
>itself.
> In particular the A in LaTeX overprints the stem of the A, and the X
> touches the E.

Yes, that's right.  I did flag this particular problem: you didn't have to
look for it yourself, did you?  That problem is mainly caused by the
incredibly stupid code to print the LaTeX logo.

> Same in "OzTeX", and also the acronym "ATM".
> But with the ragged printing of the ink jet its hard to say much more.

You've spotted yet more problems with a PS Type 1 fount rendered by ATM,
haven't you?  That was Monotype Joanna.

>Anyway, I don't think we can settle this via words alone.  Maybe you can
>find someone with a scanner and put up the comparison for others to see.
>Of course, as I said, it would make more sense to show CMR10 at 10pt
>than CMR12 at 12pt.

Well, no, it makes more sense to show cmr12 at 12pt, because that
demonstrates the problems better.

>By the way, I couldn't find information on the Deskwriter 520 on the
>HP web site, even under "discontinued products"  - so I don't know
>much about its properties.

300dpi monochrome inkjet, able to print 600dpi along one axis; but OzTeX
doesn't allow you to use a non-square resolution.

Rowland.