Re: filmscanners: What causes this and is there any easy solution ?

["Steve Greenbank" <steve@gccl.fsbusiness.co.uk>]

> >So the projection effectively helps mask the grain what a happy
coincidence
>
> While there maybe some merit to your comments about dust in the air
masking
> flaws in the slide being projected, I had the actual surface texture of
the
> projection screen in mind as well as the actual viewing distance
independent
> of any dust.  The further away from the screen you view the image the less
> likely you are to see things like grain in that like a Surrat painting
your
> eyes tend to blend the individual particles of grain into a single
> continuous tone structure even though under a loupe or standing up close
you
> will still see theindividual grains.

I thought I had covered this with some sort of statement like "even when
viewed quite close-up", but I must have rephrased this and removed it before
I posted the message. Anyway I just tried it 40x60 inch projected onto plain
white paper. With Velvia  (circa 1990) (I used the slide from which the
original sample of blue sky was made). I have to get within 16 inches to see
it at all and even then it is so faint you might miss it if you weren't
looking for it. Even from 3-4 inches it is minor. I then tried some early
Fujichrome 400 (circa 1985) and you can see the grain easily from 15 feet on
some slides. I can't wait to try scanning some of these!

>  As for screen texture, most screens
> have a pebbled or/and rectilinear surface intended to gather and
concentrate
> light so as to make them brighter (they are not smooth surfaces without
any
> texture); this surface texture also tends to break up individual noise and
> grain patterns so as to mask the grain structure of what is being
projected
> unless it is really very graining so as to have the appearance of an old
> newspaper 65 line screen halftone.

I hadn't considered this and nor did I fetch my screen when I tried the
slides tonight. But I can see that how
this would work.

> >Lower sampling rates lead to higher noise to signal ratios.
>
> I think there is probably a point at which there is NO PERCEIVABLE
decrease
> in the signal to noise rations and further increased optical resolutions
are
> of little practical point except to permit increases in output sizes while
> still maintaining a reasonably high quality non-interpolated resolution or
> to permit cropping and enlarging of small portions of the original while
> maintaining reasonably high quality non-interpolated resolutions.  Most
> monitors cannot use resolutions over 100 dpi and most printers cannot use
> resolutions over 300 dpi.  Since the less noise you have the more apparent
> the display of grain will be, it may be a good thing to compromise and
allow
> some noise to be introduced in order to tone down the sharp appearance of
> grain structure.

To some extent a little noise may help. Indeed some noise is sometimes added
deliberately in some signal processing techniques.
My sketchy understanding of digital signal processing tells me that you
require 2x (a few experts insist 4x is better[just], but for the rest of
this post I'm going to use 2x) the final output sampling rate to achieve an
almost totally accurate output. Hence CD's sample at 44KHz to achieve
accurate sound up to 22KHz.  I think the 300dpi used in the best printers
comes from the human eye being unable to see more than 150dpi so you need
2x150 or 300dpi to achieve the desired result. So for a 12x18 you need
3600x5400 which is just short of 4000dpi. I have seen Velvia printed well at
20x30 so I believe a scan of at least 6000x9000  (6000dpi) would be better
still. In the case of the Fujichrome 400 you are probably right that 4000dpi
and possibly 2000dpi is a waste of time. Something to try on a rainy day and
there's plenty of them in the UK :-)

>
> >Whilst resampling down from 4000dpi will reduce noise to signal ratios.
> >I am pretty certain that it is always best to scan at best optical and
then
> >resample down if you require a lower resolution.
>
> Although resampling down from 4000 dpi may or may not reduce the
appearance
> of noise but not the actual existence of noise, b it also will result in
the
> loss of informational data that cannot be gained back later and the
possible
> production of other troublesome artifacts.  The reduction in resolution
that
> does reduce signal to noise rations is not via the use of resampling but
via
> the actual reduction in optical resolutions being used from 4000 dpi to
some
> optical resolution under that if your scanner has an optical resolution of
> 4000 dpi.  If it has a maximum optical resolution of less than 4000 dpi
than
> any scan over that is an interpolated scan that has been upsampled by the
> scanner software and not an optical resolution, while any scan less than
the
> maximum optical scan resolution is an optical resolution.

Up sampling should generally be avoided if at all possible as it will always
lead to some nasty artefacts. I tried it in the hope the artefacts
introduced would be less noticeable than the noise removed in the down
sampling.

>
> While it is true that the conventional wisdom suggests that one scan at
the
> highest optical resolution you scanner will permit; I do not believe it
> recommends downsampling as an the suggested practice for acquiring a lower
> resolution as contrasted to rescanning the original at the required lower
> optical resolution.  Resampling upward or downward in the scanner or
> elsewhere is a last resort option and not a recommended standard practice.
>

Scanning at best optical resolution will always produce the best possible
result. When you scan an image you are digitising it and to get an accurate
result you need 2x the dpi to have an accurate representation. So if you
require a 1000dpi scan  then doing a 2000 dpi scan and resampling it to
1000dpi will give you a better result. You also have to consider the image
being scanned - if it is 300dpi then anything > 600dpi won't make any
difference. Also the ability of components in the scanner could mean that
whilst it says 4000dpi the lens (say)  actually prevents it getting near it.

> What I was suggesting was that if you compromise and reduce your scan's
> optical resolution by 500 to 2000 dpi you might be able to reach that
sweet
> spot for that particular image and film combination where the noise is not
> all that perceivable nor is the grain structure of the original.  Remember
> different films have different grain structures and different images
display
> grain differently.  An optical resolution that is optimum for one type of
> film and image combination may not be optimum for another.  In short some
> grain structures scan better than others and allow for higher optical
> resolution scans than others.

This is true for poor quality original images that will only print 7x10. But
you can only achieve the same or better result by resampling down. Scanning
at a point where you see no grain means you are invariably giving away image
detail too and the grain will still be there but the human eye won't pick it
up because the pattern has gone. Experience (when I eventually try it) might
tell me that the best I can get from Fujichrome 400 is 1000dpi and so
scanning at 2000 dpi will be no worse than 4000dpi.

>
> >I am hoping to archive the pictures in a form that will allow any one to
be
> >selected at random to be output at any size that I may require at that
> time.
> >Perhaps I'm being a bit over ambitious, but I don't see a lot of point in
> >archiving them digitally if I can still get better prints from the fading
> >original.
>
> A little over ambitious and over optimistic with expectations that might
be
> a bit too high. Technically, you will always get better prints from film
> than from digital as of the digital hardware available today, although
that
> is changing and may not be true in the near future.  Besides, I thought we
> were talking about transparencies (i.e., slides) that were being projected
> at some point and not negatives; hence, I do not understand your remark:
> "...I don't see a lot of point in archiving them digitally if I can still
> get better prints from the fading original."  Are we talking about color
> negative film of slide film?

People generally prefer prints, I prefer the quality of slides. I did for
quite a while carry two cameras one loaded with print film and one loaded
with slide film. I generally use slide for landscape and print for anything
else. These days I take most of my shots on my digicam. The quality is not
as good (particularly landscape) but I can just go mad at near zero cost
(190 shots today at Leonardslee Gardens, Horsham, Sussex, UK - highly
recommended late April and May if your in the area) and the immediate
results are emensely pleasing. Prints A4 easily and even A3 is not bad.  I
realised that the quality of the Epson 1270 digicam prints were much better
than the vast majority of the prints I had done in Labs. Even some of the so
called hand prints. So I figured 3Mp(bodged 8bit) v 20Mp (12bit) , no
problem. I also saw the output from a digital Fuji Pictrography 4000 - hell
these are significantly better than the 1270 and better than any print I can
remember seeing.

> There is a difference and that difference
> entails some differences in how they scan comparatively.  Transparency
film
> tends to be a lot more contrasty than negative film which can result in
> emphasizing the grain structure more than is the case for negative films;
> also negative films are more within the dynamic range of consumer film
> scanners than transparency films which means that the latter tend to block
> up in the highlights and shadow areas more when scanned.

Yes. But I think slide film has traditionally had less grain. Don't know why
or maybe I've just been using the wrong brands.

>
> >I did try this by resampling a 4000dpi to 2000dpi and 1333dpi and then
> >resizing back (without GF), but you have to reduce the pixel count too
much
> >and you are better off blurring the original. GF would have produced
> >marginally better results, but in my experiene GF is slightly better in
the
> >2x-3x range not a miracle worker so I still think a slight blurring would
> be
> >better.
>
> My suggest was not to do the resampling via Photoshop or some other image
> editing application but specifically with GF in both directions.  By
> resizing back to 4000 dpi using Photoshop or some other image editing
> program  and not GF, you defeated the purpose of the exercise which was to
> let the GF technology work for you in smoothing out the noise and grain
> structure.  The other forms or modes of resampling use different
> technologies which will not result in any smoothing but may produce the
> opposite.  As for GF being slightly better in the 2x-3x range, I am not
sure
> what you are referring to; I have found that it works just as well if not
> better with respect to both resizing the image as well as re-resolution
the
> image at much higher ranges than that (e.g., the 10x-15x range).  I have
> been very successful resizing a portion 2 1/2 x 3 1/2 @ 600 dpi image to a
> 16 x 20 @ 600 dpi image which printed out without any artifacts, unwanted
> noise, or unexpected grain.  GF does have a sweet spot in terms of the
file
> sizes that it works best on (which are in the 20-40 MB range); but it does
> work well with smaller and larger file sizes - although larger file sizes
> take forever to encode and decode.

I think we are sort of in agreement here.
The grain was not totally removed by the resize and so scaling up whether
with GF or PS the output will still contain parts of the grain.
GF works takes ages but works much better than PS with 3x+ magnification
below this the benefits become more and more marginal so certainly below 2x
I would argue are not worth the effort (or the money) and felt it wouldn't
have made a significant difference in my test. Once I have downsized below
2000 dpi I am starting to throw away detail and I don't want to do that
either. GF cannot recover detail that has been lost so I would prefer to
filter the grain out. If the image is very grainy then I realise that I will
have to compromise the final potential size - as you would if you printed
the image normally.


>
> However, you failed to say if you tried the GF demo which limits the size
> files you can work with to small to moderate size files no larger than
> around 40MB or the full retail version of GF which has no such limitation;
> you also neglected to mention if you used lossless or virtually  lossless
> modes of encoding in GF.
>

I have a demo but it will scale up well beyond 40MB. Possibly it doesn't
accept input bigger than this but I have only used it with digicam pictures.

>
> By the way, GF comes closest to filling you desire for an archiving
solution
> that "... the pictures in a form that will allow any one to be selected at
> random to be output at any size that I may require at that time."  The
only
> drawback is that you would have to save and keep the files in GF's .stn
> format, which would require GF to open and resave the files - at least for
> the present.  This means that we do not know if in the future GF and .stn
> format will disappear or if it will become an established standard format
> like JPEG and TIFF which all image editing programs will open and close,
> encode to and decode from.

Wouldn't use GF STN has it may have no future as you point out.

Currently they are stacking up 72Gb of hard drive in compressed TIFF. I
haven't decided what to do with them yet, but I think I will be using a
combination of TIFF and JPEG depending on the subject and the quality. I may
abandon TIFF for high quality JPEG 2000 if and when it becomes available.

Steve