old slides and scanning

I agree totally, rafe, i just wanted to point out the (potentially
fatal) flaws in the OP. This guy's website suggests he is offering
these services to ARCHIVE people's images. If I wanted to archive my
collection of slides, there is no way I would let it be done on
anything less than a 2700 dpi flim scanner. In reality, I would want
it done on a 4000 dpi scanner, and yes, I've used both types of scanner
and seen how they differ. So unless you are just archiving images
taken on a $50 point and shoot, or they are only for web-viewing, I
think 2400 is selling most images short....(and especially when he is
referring to a flatbed scanner that probably only achieves about
1800-2000 ppi anyway)...

And yes, I agree you probably won't spot the difference between a 24
and 48 bit image *if* it is just the initial scan... but if you are
going to be playing with levels, and doing a fair amount of
post-processing, working with 48 bits has quite noticable advantages.
In particular, posterisation will become an issue with some operations,
and you can end up with something looking like a GIF file.. Once
bitten, twice shy!

If all this was about was creating a DVD for family viewing on the
telly... maybe. But if you were planning to process and print those
images to say 12x8 with minimal loss of quality....

By the way, it probably needs to be said that the only reason the OP
can't see much difference between 1200 and 2400 is that his scanner is
only capable of about 1800 anyway - so that is more of a commentary on
the scanner!

On Wed, 16 Mar 2005 18:40:24 -0700, "Roger N. Clark (change username
to rnclark)" wrote:


Do these samples not show that 4000 ppi drum scans is getting
most of the info, and the comparison of ~2400 versus 4000 ppi
is quite a big difference? Then consider the OP says not much
difference between 2400 and 1200?


I guess I'd take a stand somewhere between chrlz and the OP.
One exaggerates toward the low end, the other towards the
high end.

I've only seen a handful of real images, from any scanner,
that show significant detail beyond 4000 dpi. And I say
this after a brief period of owning a fine old 5000 dpi
drum scanner.

Even so, I believe you're now scanning your 4x5 with
an Epson 4870? And surely you're not going to claim
that its real resolution is anywhere near 4800 dpi?

Plus there's something very strange about your scan-
resolution comparisons. Where's the grain? Even the
best slide films will have lots of grain showing
at these resolutions. (The 2nd, 3rd and 4th 35mm
images in scanres-samp2.jpg)


I am a total convert to 16-bit processing. My first
step in working on an 8-bit/pixel jpeg from my digital
camera is convert to 16-bit. That way I minimize
posterization and can get more out if the image.


I'm not quite convinced. And neither is Dan Margulis.
But with memory and drive space as cheap as they are,
one can have it both ways. I confess to doing more
of my scanning at 16 bit these days, though I'm not
at all convinced of the need or benefit.

We agree, I hope, that most scanners don't deliver
much more than 11 or 12 bits (at best) of real data,
regardless of their claimed bit depth. Anything
beyond that is usually noise.

16-bit processing simply allows you to defer your
major color moves further down the image-processing
chain before irretrievable loss of tonality occurs.

If you scan carefully -- ie., use all the range
available in all color channels -- the benefit
is minimal. Unless of course you plan to do
really radical color moves in Photoshop.

If you use half the range, you are in effect
throwing away one bit worth of image data.

Given a perfect output device, how many folks
could distinguish every step of a 256-step
grayscale wedge?

If you went to 128 steps, how much more distinct
would the steps be?

Now add noise to the picture (literally, since
most images have noise in abundance) and the
edges between the steps gets even hazier.

I think if there weren't a histogram for people
to look at, 16 bit processing wouldn't be anywhere
near as popular is it is. Somewhere along the
line someone decided that gapped histograms
were a Very Bad Thing... regardless of whether
there's an observable effect of that, on-screen
or in print.

Again, if the image in question involves subtle
gradients and large "near-monochrome" areas, those
extra bits might come in handy. But I've seen few
examples of this in practice.


rafe b.
http://www.terrapinphoto.com

rafe bustin wrote:

Even so, I believe you're now scanning your 4x5 with
an Epson 4870? And surely you're not going to claim
that its real resolution is anywhere near 4800 dpi?

That is correct. I'm scanning my 4x5 velvia at 3200 dpi.
The scanner resolution is probably about that. The scans
are a little softer than then drum scans I have had done.
The problem with the drum scanner I use (Linotype-Hell; I need
to get exact model number) is that while it does
up to 11,0000 dpi and did 16 bits/pixel output,
the files written had to be 8-bit. I need to see if that
has changed in the last year. Anyway, what I found with the
epson is that while it was a little less spatial resolution
than the drum scan, the 16-bit output made better images (!!)
because the intensity detail was so much better.

Plus there's something very strange about your scan-
resolution comparisons. Where's the grain? Even the
best slide films will have lots of grain showing
at these resolutions. (The 2nd, 3rd and 4th 35mm
images in scanres-samp2.jpg)

There is lots of grain in the images. Probably the best example
is the image at:
http://www.clarkvision.com/imagedeta...ml#digicamres2
See the image in this section, just above results/conclusions.
The image labeled "4x5 original 1.03x" shows the grain well.
Remember these images are fujichrome velvia, which shows much
less grain in scanned images than provia 100 which is what I see
most people use in scan comparisons.

I'm not quite convinced. And neither is Dan Margulis.
But with memory and drive space as cheap as they are,
one can have it both ways. I confess to doing more
of my scanning at 16 bit these days, though I'm not
at all convinced of the need or benefit.

I also use a program called ImagesPlus which works in true
16-bit math as opposed to Photoshop's 15-bit. I'm seeing
posterization in Canon 1D mark II images processed in photoshop
in "16-bit" that I don't see in ImagesPlus.
For examples of an intensity profile of 8-bit versus 16-bit,
see Figure 9a,b,c,d at:
http://clarkvision.com/imagedetail/dynamicrange2

We agree, I hope, that most scanners don't deliver
much more than 11 or 12 bits (at best) of real data,
regardless of their claimed bit depth. Anything
beyond that is usually noise.

I agree. Most have 12-bit A-to-Ds anyway, or are
photon noise limited. But even with 12-bit DSLR data,
I see artifacts from processing in photoshop's 15-bit math.
I guess I need to put together some web pages showing
that effect. I'll add it to my list.

If you scan carefully -- ie., use all the range
available in all color channels -- the benefit
is minimal. Unless of course you plan to do
really radical color moves in Photoshop.

I agree, except when you need to compress dynamic
range, like bring detail out of clouds, it can't be done
in 8-bit (and I'm not talking color shifting).
Similarly when you want to bring detail
out of the shadows. 16-bit scans give you a better
chance of getting the detail, just like RAW files
on a digital camera give you more detail.

Given a perfect output device, how many folks
could distinguish every step of a 256-step
grayscale wedge?

This is irrelevant. The whole point of image processing
is to bring the dynamic range into what can be output
into a medium for human viewing. It is a matter of
capturing the huge scene dynamic range, which is often
over 10, 11, or more stops, and compressing it for
viewing. Often the output medium is the limiting step,
not the human eye.

I think if there weren't a histogram for people
to look at, 16 bit processing wouldn't be anywhere
near as popular is it is. Somewhere along the
line someone decided that gapped histograms
were a Very Bad Thing... regardless of whether
there's an observable effect of that, on-screen
or in print.

I strongly disagree. I've been doing image processing
for about 30 years. Sometimes I'm working with
data with a signal-to-noise tens of thousands and dynamic
ranges of millions (scientific applications). The ability
to capture a scene and bring out detail that could be seen
with the eye is pretty amazing--things that could not be done
with a traditional enlarger without special masks and
extreme skill. And that goes for film and 16-bit scans as
well as digital cameras and RAW files.

Again, if the image in question involves subtle
gradients and large "near-monochrome" areas, those
extra bits might come in handy. But I've seen few
examples of this in practice.

Perhaps it depends on what images you like to do.
I would say every landscape photo that includes the sky
and clouds qualifies.

Roger

rafe bustin wrote:

Even so, I believe you're now scanning your 4x5 with
an Epson 4870? And surely you're not going to claim
that its real resolution is anywhere near 4800 dpi?

That is correct. I'm scanning my 4x5 velvia at 3200 dpi.
The scanner resolution is probably about that. The scans
are a little softer than then drum scans I have had done.
The problem with the drum scanner I use (Linotype-Hell; I need
to get exact model number) is that while it does
up to 11,0000 dpi and did 16 bits/pixel output,
the files written had to be 8-bit. I need to see if that
has changed in the last year. Anyway, what I found with the
epson is that while it was a little less spatial resolution
than the drum scan, the 16-bit output made better images (!!)
because the intensity detail was so much better.

Plus there's something very strange about your scan-
resolution comparisons. Where's the grain? Even the
best slide films will have lots of grain showing
at these resolutions. (The 2nd, 3rd and 4th 35mm
images in scanres-samp2.jpg)

There is lots of grain in the images. Probably the best example
is the image at:
http://www.clarkvision.com/imagedeta...ml#digicamres2
See the image in this section, just above results/conclusions.
The image labeled "4x5 original 1.03x" shows the grain well.
Remember these images are fujichrome velvia, which shows much
less grain in scanned images than provia 100 which is what I see
most people use in scan comparisons.

I'm not quite convinced. And neither is Dan Margulis.
But with memory and drive space as cheap as they are,
one can have it both ways. I confess to doing more
of my scanning at 16 bit these days, though I'm not
at all convinced of the need or benefit.

I also use a program called ImagesPlus which works in true
16-bit math as opposed to Photoshop's 15-bit. I'm seeing
posterization in Canon 1D mark II images processed in photoshop
in "16-bit" that I don't see in ImagesPlus.
For examples of an intensity profile of 8-bit versus 16-bit,
see Figure 9a,b,c,d at:
http://clarkvision.com/imagedetail/dynamicrange2

We agree, I hope, that most scanners don't deliver
much more than 11 or 12 bits (at best) of real data,
regardless of their claimed bit depth. Anything
beyond that is usually noise.

I agree. Most have 12-bit A-to-Ds anyway, or are
photon noise limited. But even with 12-bit DSLR data,
I see artifacts from processing in photoshop's 15-bit math.
I guess I need to put together some web pages showing
that effect. I'll add it to my list.

If you scan carefully -- ie., use all the range
available in all color channels -- the benefit
is minimal. Unless of course you plan to do
really radical color moves in Photoshop.

I agree, except when you need to compress dynamic
range, like bring detail out of clouds, it can't be done
in 8-bit (and I'm not talking color shifting).
Similarly when you want to bring detail
out of the shadows. 16-bit scans give you a better
chance of getting the detail, just like RAW files
on a digital camera give you more detail.

Given a perfect output device, how many folks
could distinguish every step of a 256-step
grayscale wedge?

This is irrelevant. The whole point of image processing
is to bring the dynamic range into what can be output
into a medium for human viewing. It is a matter of
capturing the huge scene dynamic range, which is often
over 10, 11, or more stops, and compressing it for
viewing. Often the output medium is the limiting step,
not the human eye.

I think if there weren't a histogram for people
to look at, 16 bit processing wouldn't be anywhere
near as popular is it is. Somewhere along the
line someone decided that gapped histograms
were a Very Bad Thing... regardless of whether
there's an observable effect of that, on-screen
or in print.

I strongly disagree. I've been doing image processing
for about 30 years. Sometimes I'm working with
data with a signal-to-noise tens of thousands and dynamic
ranges of millions (scientific applications). The ability
to capture a scene and bring out detail that could be seen
with the eye is pretty amazing--things that could not be done
with a traditional enlarger without special masks and
extreme skill. And that goes for film and 16-bit scans as
well as digital cameras and RAW files.

Again, if the image in question involves subtle
gradients and large "near-monochrome" areas, those
extra bits might come in handy. But I've seen few
examples of this in practice.

Perhaps it depends on what images you like to do.
I would say every landscape photo that includes the sky
and clouds qualifies.

Roger

On Wed, 16 Mar 2005 21:40:01 -0700, "Roger N. Clark (change username
to rnclark)" wrote:


I strongly disagree. I've been doing image processing
for about 30 years. Sometimes I'm working with
data with a signal-to-noise tens of thousands and dynamic
ranges of millions (scientific applications). The ability
to capture a scene and bring out detail that could be seen
with the eye is pretty amazing--things that could not be done
with a traditional enlarger without special masks and
extreme skill. And that goes for film and 16-bit scans as
well as digital cameras and RAW files.


So it goes. This topic's been beat to death on
this and other fora for years, and even the
experts disagree. I respect and admire your
work and articles in this topic, but I think
there's still room for debate.

I don't usually resort to math in my approach to
"digital darkroom" but I know what works and what
doesn't. Posterization, even in clear blue skies,
is not an issue with my files or prints. Pure
luck? I doubt it. See if you can spot posterization
on any images on my website or the scan snippets
site. Clear blue sky? How about the 2nd "Perfect
Scan" near the top of the scan-snippets site?

On "my" side of the argument is a very clever fellow
named Dan Margulis, whom you may have heard of. In
fact you'll see a thread of this same discussion
somewhere on Dan's web site. A fellow named
Jeff Schewe is his main opponent in that debate.

I'm not about to discourage anyone from using 16
bit image files, but I've heard lots of ridiculous
claims for the benefits of doing so. For digicams
I certainly encourage the use of RAW files, but
that's a slightly different issue.

As for image resolution... again, I'm eager to
receive and post full-res scan samples -- any
film, any lens, any camera -- that can beat
the scans already posted. I really am
interested in the question, "Just how much
detail can be extracted from 0.25" x 0.25"
of film." Only real photos please, not
test targets. If you've got 6000 dpi scans,
I'll be happy to post a 1500 x 1500 pixel,
full-res snippet, cherry-picked for the
sharpest detail you can find, and JPG'd at
best quality/minimimum compression. Please
send a small "overview" JPG of the frame
from which the snippet is derived.


rafe b.
scan snippets:
http://www.terrapinphoto.com/jmdavis

"Gerrit 't Hart" wrote:

OK now that all of us have read (and tried to understand) the arguments
can
someone please advise which flatbed to buy?

Epson 4870 (under US$500) or Epson 4990 (US$600). (Prices are just guesses.)

I have about 4000 colour slides and lots of black and white family photos
going back to the early 1940s whioch I would like to digitise for the
whole
family. We are not wealthy and were wondering whether an Epson of about
2800dpi is enough or 3600 dpi or if in fact we need to spend $900 for one
with 4000 or even 4800 dpi? As you are aware the price does rocket upwards
the higher the resolution.

Both these provide digital ICE, which automagically hides dust and scratches
when scanning color slides. Not having ICE is not nice. The 4990 has a
larger transparency adapter, so can scan more slides at the same time and
might be worth the extra money in speeding up the process.

The 4800 dpi claimed for these is a tad bogus. You can scan at 2400 dpi with
them and get nearly the same results (but better results than with "2400
dpi" scanners).

David J. Littleboy
Tokyo, Japan

rafe bustin wrote:
On Wed, 16 Mar 2005 18:40:24 -0700, "Roger N. Clark (change username
to rnclark)" wrote:

I am a total convert to 16-bit processing. My first
step in working on an 8-bit/pixel jpeg from my digital
camera is convert to 16-bit. That way I minimize
posterization and can get more out if the image.

I'm not quite convinced.

It depends on your working space. 8 bits might be enough for sRGB,
but if you're working in a wide gamut space -- and that is a good idea
when you're scanning transparenices -- then 16 bits is a Good Thing.

Andrew.

In article ,
wrote:
rafe bustin wrote:
On Wed, 16 Mar 2005 18:40:24 -0700, "Roger N. Clark (change username
to rnclark)" wrote:

I am a total convert to 16-bit processing. My first
step in working on an 8-bit/pixel jpeg from my digital
camera is convert to 16-bit. That way I minimize
posterization and can get more out if the image.

I'm not quite convinced.

It depends on your working space. 8 bits might be enough for sRGB,
but if you're working in a wide gamut space -- and that is a good idea
when you're scanning transparenices -- then 16 bits is a Good Thing.

Indeed. If the bits that make up a pixel value can represent lots of
different colours, then it stands to reason that they can represent fewer
sahdes of those colours - something has to give. 8 bit per channel wide
gamut images would seem to be at a greater risk of posterisation than
narrower gamut images.

This implies that when converting to 8 bits, sRGB for web use, one should
convert to sRGB *first*, then convert to 8 bits, and not th eother way
around.

On Thu, 17 Mar 2005 20:03:05 -0700, "Roger N. Clark (change username
to rnclark)" wrote:


Cloud highlights blown:
http://www.terrapinphoto.com/chengs_tree.html
http://www.terrapinphoto.com/marcydam2.html
http://www.terrapinphoto.com/mt_dix_grass.html
http://www.terrapinphoto.com/chim_po...eflection.html
http://www.terrapinphoto.com/bridlep...solarized.html
http://www.terrapinphoto.com/old_lobster_trap.html
http://www.terrapinphoto.com/morocco_cityscape.html
http://www.terrapinphoto.com/duomo_from_uffizi.html
http://www.terrapinphoto.com/log_fen...ayah_bald.html
http://www.terrapinphoto.com/kinsman_cairn2.html
http://www.terrapinphoto.com/nahmakanta_lake.html
http://www.terrapinphoto.com/cadillac_flowers.html
http://www.terrapinphoto.com/acadia.html
http://www.terrapinphoto.com/acadia_sunset2.html
http://www.terrapinphoto.com/tide_pool.html
http://www.terrapinphoto.com/red_cliff.html


I think we must disagree on the definition of
"blown" highlights. I prefer to use the term
clipped, if you don't mind G.

"Clipped" to me, means a signficant count of
pixels in one or more color channels at 0 or 255,
ie. a histogram piled up against either wall.

I found three images from the above list with
some degree of clipping. On that basis, I'm
happy to dismiss those three as lousy scans.
(But nothing nearly as bad as your first
sanjuan image.)

I plucked several of these images back off the
site and selected a 30x30 pixel region from the
brightest portion of the sky. The histograms
of those regions were perfectly reasonable
(to my thinking) in every color channel. I'm
curious what you'd have expected to see
differently had these been 16-bit scans.

I use as much tonal range as I dare. If I
allocate more tones for the highlights, or
for the shadows, I've got to steal them from
somewhere else. Surely you know all that.

You can't look at a given image and presume
that a given region should have a given a
specific median or standard deviation of
tones. As the "artist," I get to choose
how the tones are distributed.

The only way I know to get around this is
with selective tonal control or contrast
masking, which I do use from time to time.

In my experience, scanning and editing in
16-bit isn't going to change a thing. But
I'll tell you what... I'm ready to challenge
my own assumptions and see if I can make a
few more rigorous tests. All these negatives
are still around, and the scanner's ready to
roll.

Thanks for the challenge. I'll report back
when I've learned a bit more.


rafe b.
http://www.terrapinphoto.com