Apache-Talk @lexa.ru 

Inet-Admins @info.east.ru 

Filmscanners @halftone.co.uk 

Security-alerts @yandex-team.ru 

nginx-ru @sysoev.ru 




      :: Filmscanners
Filmscanners mailing list archive (filmscanners@halftone.co.uk)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[filmscanners] Re: Compression

On Thu, 10 Oct 2002 20:11:05 -0500   (snsok@swbell.net) wrote:

> Can you elaborate on your reservations about his film to digital
> comparison??
> Certainly the images he presents on the first page are compelling.

Posted: Oct. 07 2002,06:02


A couple of points which hopefully add a little on aliasing/tonal
smoothness, and dynamic range. My biggest concern with the review and
conclusions is how the #### I am ever going to afford a 1DS...


Of course there will now be a chorus of those who say, "Ya, but a drum scan
would have really shown a bigger
difference in favour of film." Humm. Maybe. But here are my thoughts on
this recurring topic. I have had drum scans made from my 35mm and medium
format film on several occasions. Yes, an 8000 ppi scan is impressive, and
can make bigger prints. But, I'm also convinced that while they give me
more pixels, I don't get a whole lot more real data. There simply isn't
that much more information on film than about 4,000 PPI. Above that we get
bigger files, but not much more information. Maybe, 20% more than the 3200
PPI scans that my Imacon Flextight Photo scanner is capable of, but not 2
or 3 times as some inexperienced people presume from the numbers.


I agree with Michael's 4,000ppi  'diminishing returns' assessment as far as
image detail is concerned, but it isn't enough to reach conclusions
regarding tonal smoothness.

Film grain is irregular both in size, distribution and topology, and all of
these attributes interact with the fixed geometry of pixel size and
distribution. What many people believe is grain in their scanned images is
often aliasing, which produces an exaggerated grain-like structure
comprising false colour/false luminance pixels.

Avoiding this in low-pixel-count scanners is only possible by using a
low-pass optical system (soft lens or antialiasing filter) which hurts
image detail as well.

Grain aliasing can be so extreme as to render a scan unusable, especially
with grainy, fast B&W silver-based films, but generally it is acceptable
with scanners  >=4,000ppi.

When it happens, the result is far more 'texture' than the film image
exhibits when viewed or printed by analogue techniques.

Having seen comparative drum scans of an ISO100 tranny done at 4,000,
8,000, and 12,000 ppi, there isn't much gain in resolution of image detail
above 4,000ppi, which accords with what Michael says in his review. By
8,000 ppi the scan has virtually all of it. Yet 12,000ppi shows a
significant increase in grain detail over 8,000ppi, with better rendition
of individual grain shapes. A lot of people say this doesn't matter, grain
detail is not image information we want...

However, if we don't capture precise grain information, what we get through
the conversion to pixels is aliasing, and a characteristically false and
'noisy' rendition of the film image. An Imacon scan at 3,200ppi necessarily
adds an unknown  extra dimension of aliasing noise.

How much, and how destructive depends on the anti-aliasing filter and CCD
properties, even the lightsource, but it is important to recognise that any
comparison with scanned film is not a comparison with film itself.

Pedantically, I think it would be worth doing a comparison involving a
12,000ppi drum scan, to get a more absolute measure of the differences
between the film images and the EOS1DS. Not for the sake of resolution per
sebut to judge relative tonal smoothness more accurately.

Having said all that, I am sure digital will still win, but the existing
methodology probably makes film look rather worse than it is in this
respect (and JPEG'ing of an image full of aliasing products compounds the
issue, to film's disadvantage).

Michael's observation of moire and the artefacting of the red-shed boarding
are more of the same. Aliasing is inherent with pixel-based systems, though
the Foveon sensor will avoid  colour aliasing and only alias luminance.

For me, the only obvious remaining advantage of film (apart from avoiding
divorce and bankruptcy) is dynamic range.

My main interest and use of scanning is so I can use colour negative and
B&W chromagenic films, exactly because they have a mad dynamic range : what
you can see with adaptive eyes, you can photograph, pretty much.

I have tested and found 12+ stops with these materials. 10 stops of this is
completely usable, there is severe compression in the rest, but it is still
there. This is absolutely liberating compared to tranny, and gives me the
control I only had in the wet darkroom with B&W, and more.

Alongside this, the D1S is, as reported, distinctly poor at 6 stops.

What's more, it is almost certainly an engineered inhibition. CCD sensors
can now achieve 14 stops range, and do so completely linearly. I don't know
any reason why CMOS should be worse, so it sounds like Canon have knobbled
the range.

I hope this will prove to be something that applies only to JPEGS, and the
RAW files will contain far more range.

There is a very good reason why Canon may have done so : wide range images
look absolutely awful on-screen, impossibly flat, low contrast. You have to
work with them, as raw material, and you wouldn't want to have to cope with
the consequences of JPEG, it has to be clean, raw, high-bit data.

So the JPEG format is probably assumed to be for rapid onward transmission
as a production file, with levels set, and only 8bit/ch available. To do
anything else would frighten the desk editor who receives the file, and
involve them in making judgements which are best left to photographers...

Fingers crossed, anyhow.

Tony Sleep (sometime filmscanner reviewer - suddenly retired
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe 
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or 


Copyright © Lexa Software, 1996-2009.