NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/comp/text/desktop/512

Internet Message Format | 1992-12-26 | 6.9 KB

Xref: sparky comp.text.desktop:512 rec.photo:22573
Path: sparky!uunet!pipex!bnr.co.uk!uknet!mcsun!julienas!ircam!francis
From: [e-mail] (Joseph Francis)
Newsgroups: comp.text.desktop,rec.photo
Subject: Re: Photo Scanning
Message-ID: <[e-mail]>
Date: 26 Dec 92 22:25:52 GMT
References: <[e-mail]> <[e-mail]> <[e-mail]>
Organization: IRCAM, Paris (France)
Lines: 120

Newsgroups: comp.text.desktop,rec.photo
Subject: Halftoning (Re: Photo Scanning)
Summary:
Expires:
References: <[e-mail]> <[e-mail]> <[e-mail]>
Sender:
Followup-To:
Distribution:
Organization: IRCAM, Paris (France)
Keywords:

In article <[e-mail]> [e-mail] writes:
>In article <[e-mail]>, [e-mail] (Joseph Francis) writes:
>> In article <[e-mail]> [e-mail] writes:
>>>In article <[e-mail]>, [e-mail] (Joseph Francis) writes:
>...
>> Why 300dpi? Everything I've read says that the eye, in normal
>> conditions 29.7 cm from a 21x21cm square (the correct distance for
>> looking at photographs, w.r.t. the image size for least optical
>> distortion), can't resolve more than 300dpi @ 256 grey levels per
>> pixel of the image. I think it has to do with the angle subtended by
>> the image on the retina, and the density of receptors. Color is
>> similar in treatment, 300dpi per color layer, but this model doesn't
>> account for discrepancies in color sensitivity (green, vs. red).
>>
>I've seen this bull too. Need I say more? The problem is that a simplistic
>'resolution required' test fails in many subtle important cases, where the
>resolution errors compound in horrid ways. That is why good typesetters run
>~7k dots per inch or more.

That's good one. I looked at my typesetter price sheets and they only
go up to 2540 dpi (8bits/pixel @300dpi, 158.75lpi gives 256
grey-levels per dot). Many printers won't (can't) really handle more
than ~150lpi halftones anyway, since at drop sizes below that you get
into interesting physical problems with ink. 4800dpi gives you 300lpi
@8bits/pixel (256 grey-levels) and 450dpi @8bits/dot, that's 450lpi
halftoning (3x the max printers will normally handle) puts you at 7200
dpi.  Don't you think that's well, a bit much? You're in the regime
where random ink spreading will wipe out any fine differences, unless
you are printing on glass in a vacuum at 0G. I might suggest that
7000dpi is, well, exaggerating a little. I understand completely about
compounded errors; I don't think that's the issue here. Just
perceptible resolution. I don't think the eye can see any difference
at all, according to my geometry, between 1200dpi and 600dpi printing
anyway, much less 2450dpi to 1200dpi. Except, of course, for
disappearing hairlines, which are recommended against for
high-resolution typesetting anyway (the 'disappearing' part).

---

I thought more about this problem, and the ongoing question I have
about duotoning. If you scan a photo with low contrast, and assign
that area to 50% grey in a print, then scan the same photo for
high-contrast regions, and assign that to a 100% black for print, you
get several great things happening. (I first imagined that you would
merely scan for a specific band of brightness, and combine layers, but
this presented so many problems I had to reconsider. You have to see
where you want information: and I thought for photos, 'shoot for
detail in highlights, print for detail in black' so you should do
precisely that for scanning; and then add information in the grey
region for smooth transitions.)

1. There will be more information concentrated in the highlights and
shadows, what I see now is a problem. Detail loss in the black is
something I've seen referred to in all scanning texts. Scanning also
makes pictures slightly more contrasty.
2. There will be equally as much information in the middle greys. The
low-contrast region. So, you retain nice smooth transitions, without
banding artifacts.
3. You will have effective 300lpi in the image as printed. 150lpi for
black, 150lpi for the grey, black-on-grey. There is overlap, certainly
but what is happening is that you are filling in information in detail
in dark regions.
4. I don't see how you can preview the stuff easily with DTP. (I have
been working with a photo today, and I'm going to get PBM to overlay
halftoned and color-assigned TIFFs in 24-bit space, and see how that
looks. Hopefully this will be an accurate preview.)
5. You can actually assign any /color/ to the 50% and 100% regions and
get wonderful tinted printing.
6. You can extend the method. When scanning a photo, make not just one
contrast separation, but two, or three, assigning each to a different
grey level. When overprinted, you should get more and more information
in the image, and therefore better overall resolution. This also means
you can use transparent colors for printing in the different contrast
bands and get very interesting effects, but that is secondary. Most
effects I've seen now assign colors merely to different intensity levels.

I bought a heterosexual friend of mine a copy of Madonna's, uh, Magnum
Opus "Sex" (with, I note, a cheery CD single of 'Erotica', Dita
notwithstanding: at that price, why not the whole album? (I won't tell
you what the damn thing cost in Paris)). I immediately started looking
at it closely with a loupe.  Almost all the B&W photos are printed as
I detailed above, Black-on-Grey, with grey being the low-contrast
layer. Some of the photo sets were black-on-four-color, and I really
couldn't tell any major difference except for misregistration. None of
the two-tone prints were just two tone,they were invariably
four-color. It was all around 160lpi screening, printed on uncoated
paper. It is considered a medium-quality coffeetable artbook.
Unfortunately, I can see already that the stamped aluminum cover is a
problem (fingerprints: why not matte black?) and the uncoated paper is
warping. The spiral bind is a disaster waiting to happen. I'm going to
get the new Mapplethorp super-compendum and see how those photos are
printed, which are my #1 favorite B&W photo printings. Really luxurous
rich greys and detailed blacks.

>Knuth, in one of the TeX/Metafont books/srticle gives a sobering example of
>round-off error in the 13 odd decimal place. Can't find the ref at the mo...
>In the article on device drivers I think.

Yes, but unless you are printing on precision paper with precision ink,
I think you don't need to worry about more than the lsb for 8-big scanning.

>The Kodak leaflets on screening have a lot of good stuff on this ( or at least
>they used to, I haven't looked at onne for 20 odd years... )

Ahh! But that will be analog screening...

--
| Le Jojo: Fresh 'n' Clean, speaking out to the way you want to live
| today; American - All American; doing, a bit so, and even more so.