Epson 4870 resolution setting vs true resolution vs print DPI (ATTN Epson 4870 Resolution Test contr

[false_dmitrii]

At some point I picked up 600dpi as the "practical" limit for scans of
most printed media, including photos. This was supposedly due to
limitations of print processes.

Today I experimented with a 50-year-old B&W photo print (a "normal"
photo, on photo paper--don't know the term to distinguish it from
something in a magazine) and my neglected Epson 4870 Photo. I found
that although a scan at EpsonScan's 600dpi was far smoother than one
at 300dpi, the 1200dpi setting offered further visible improvement.
In other words, aliasing dropped away and distinct details continued
to develop without signs of grain or other media-related detail
limits.

I went ahead and made a 500MB(!) 4800dpi scan. There was further
improvement, but it was very subtle and suggested that I was around
the practical resolution limit. Overall, then, I'd say the 1200dpi
scan was fine for most uses but did not fully capture *everything*.

This would not be a factor for enlarging the original photo to page
width or so, but it might affect an attempt to isolate and enlarge a
small portion (i.e. one person from group) to a modest final size.
It's also a consideration for keeping the scan true to the original
for long-term storage and future use.

An earlier 600dpi scan of a printed color manual revealed every color
dot when I zoomed in and required Moire pattern removal, so I was
surprised at how much today's photo continued to improve as resolution
shot up.

My question, for all you experts: What is the relation between the
dpi numbers of the media and scanner settings I've given?

Is the 600dpi print limit wrong, or at least taken out of context?
Is there something different about the 50-year-old B&W media?
Or, is the Epson's "1200dpi" or even "4800dpi" true resolution in fact
closer to the 600dpi print limit?
If the latter, are there steps I might take to boost detail resolution
and smoothness without creating 30-100MB files (100-and-up in color
)? Software tweaks? Physically assisting the scanner somehow?

My underlying question is "what's the best resolution for scanning
printed photos with my 4870," but since that's context-dependent, I'm
trying to improve my understanding of resolution limits. If possible,
I want to produce acceptably "definitive" long-term photo scans right
now; otherwise I'll stick to far more practical current-use sizes.

Thanks for all help!
false_dmitrii

 

[Kennedy McEwen]

At some point I picked up 600dpi as the "practical" limit for scans of
most printed media, including photos. This was supposedly due to
limitations of print processes.

Today I experimented with a 50-year-old B&W photo print (a "normal"
photo, on photo paper--don't know the term to distinguish it from
something in a magazine) and my neglected Epson 4870 Photo. I found
that although a scan at EpsonScan's 600dpi was far smoother than one
at 300dpi, the 1200dpi setting offered further visible improvement.
In other words, aliasing dropped away and distinct details continued
to develop without signs of grain or other media-related detail
limits.

I went ahead and made a 500MB(!) 4800dpi scan. There was further
improvement, but it was very subtle and suggested that I was around
the practical resolution limit. Overall, then, I'd say the 1200dpi
scan was fine for most uses but did not fully capture *everything*.

This would not be a factor for enlarging the original photo to page
width or so, but it might affect an attempt to isolate and enlarge a
small portion (i.e. one person from group) to a modest final size.
It's also a consideration for keeping the scan true to the original
for long-term storage and future use.

An earlier 600dpi scan of a printed color manual revealed every color
dot when I zoomed in and required Moire pattern removal, so I was
surprised at how much today's photo continued to improve as resolution
shot up.

My question, for all you experts: What is the relation between the
dpi numbers of the media and scanner settings I've given?

Is the 600dpi print limit wrong, or at least taken out of context?
Is there something different about the 50-year-old B&W media?
Or, is the Epson's "1200dpi" or even "4800dpi" true resolution in fact
closer to the 600dpi print limit?
If the latter, are there steps I might take to boost detail resolution
and smoothness without creating 30-100MB files (100-and-up in color
)? Software tweaks? Physically assisting the scanner somehow?

My underlying question is "what's the best resolution for scanning
printed photos with my 4870," but since that's context-dependent, I'm
trying to improve my understanding of resolution limits. If possible,
I want to produce acceptably "definitive" long-term photo scans right
now; otherwise I'll stick to far more practical current-use sizes.

Hi,
to avoid confusion try to use the terms dpi (dots per inch) when
referring to the printed page, either newspaper, magazine , inkjet or
laser and ppi (pixels per inch) when referring to images, such as
scanners etc. Sometimes the latter needs further discrimination as
well, to avoid confusion between pixels in the final image and scanner
samples which may not always be the same thing.

The 600ppi limit for photos is a general rule of thumb for photographic
prints, but it us not a universal rule. Wayne Fulton recommends 300ppi
as the limit for scanning colour photographic prints on his scantips
website at http://www.scantips.com/resolut.html but again, this is a
general rule and there are exceptions, depending on the emulsion of the
original print. Even so, going from these limits to 1200ppi or even
4800ppi with your 50 year old print is a major step so you need to ask
yourself if you really are seeing improved resolution, rather than just
improved image quality which can be achieved just as readily with the
lower resolution scan.

You mentioned that aliasing dropped away when you went up to 1200ppi
from 600ppi on your photographic print, but are you sure it is
additional resolution that you are seeing and not just better
visualisation of fine detail because they are scaled up in size?

Somewhere near the middle of the web page I referenced above, Wayne
compares the image of a 200ppi scan with one from a 400ppi scan and
concludes that they have the same resolution. Whilst it may be
debatable whether that is exactly true, there is certainly very little
resolution difference between them.

So, before you rush off to scan all of your old B&W photos at 4800ppi,
make sure you are comparing the images at the same size and using a good
interpolation technique to implement that scaling - eg. Bicubic as a
minimum.

Having said that, it is certainly possible that your 50-year old photo
print does have sufficient resolution to require 1200ppi or more,
especially if it was printed from a large format negative with little or
no enlargement. After all, only 10 years before that print was made an
awful lot of similarly printed aerial photographs were pored over under
high magnification on a daily basis - the analysts would not have used
magnification if the emulsion did not contain the resolution to justify
it. ;-)

 

[Robert Feinman]

My underlying question is "what's the best resolution for scanning
printed photos with my 4870," but since that's context-dependent, I'm
trying to improve my understanding of resolution limits. If possible,
I want to produce acceptably "definitive" long-term photo scans right
now; otherwise I'll stick to far more practical current-use sizes.

Thanks for all help!
false_dmitrii

Conventional b&w enlargement on silver halide paper is extremely
fine grained and has very high resolving power. The limitations on
rescanning have always been as a result of the enlargement process from
the film itself. Most "sharp" images probably end up with about 8 lp/mm.
A contact print from a carefully made negative could be as much as 40
lp/mm.
For printed material from books, etc. the halftone screen is the limit
and ranges from a low of about 75 dpi to about 150 dpi. Scanning at
twice the screen frequency should be enough.
Another limiting factor is the texture on the paper you are scanning.

 

[Wayne Fulton]

The reason that my site comments on it is because people do often want to
scan a wallet size print and blow it up 4x to 8x10 or A4 size. This simply
doesnt work out well, because the first photo print just doesnt have it to
give, and I try to explain why.

It is not due to the scanning - anyone that has ever tried photographing the
wallet print on a film copy stand knows that the new film copy wont enlarge
any better, because the original print just doesnt have the detail to give.
The print simply was not designed to be enlarged. We should instead start
over from the original film, which is designed to be enlarged as its only
purpose, and can easily do it. A print is a relatively poor second
generation copy of the film original.

We can indeed scan color photos at 600 dpi and print 2x size at 300 dpi,
which is the correct theory of enlargment, number-wise. And 2x may be
acceptable, but it is not as good as hoped. Results printing 3x size is
worse, quite soft. Prints simply have a limit of what they can give (but the
original film is designed with vastly higher capability).

My opinion is that 300 dpi is a practical limit for prints, based on the
typical 4x color print from the photofinisher (35 mm film). Others such as
Roger Clark at http://www.clarkvision.com disagee a bit about degree, but his
standard was a professional 2x print from 4x5 inch film, which of course is
not the typical situation for most of us. Even so, it doesnt disagree much.

But historic B&W prints were often/usually contact prints at original film
size, no enlargment at all. These are often a little better regarding
detail. 600 dpi is a reasonable try for those film-size B&W contact prints.
Assuming of course that the purpose needs that much image size.

 

[false_dmitrii]

Hi,
to avoid confusion try to use the terms dpi (dots per inch) when
referring to the printed page, either newspaper, magazine , inkjet or
laser and ppi (pixels per inch) when referring to images, such as
scanners etc. Sometimes the latter needs further discrimination as
well, to avoid confusion between pixels in the final image and scanner
samples which may not always be the same thing.

I realize there's a difference. My understanding of it is partial and
probably wrong in spots, so it seemed safer to stick with the broader
use of "dpi" than to introduce more potential for inaccuracy. But
I'll pay more attention for future posts.

The 600ppi limit for photos is a general rule of thumb for photographic
prints, but it us not a universal rule. Wayne Fulton recommends 300ppi
as the limit for scanning colour photographic prints on his scantips
website at http://www.scantips.com/resolut.html but again, this is a
general rule and there are exceptions, depending on the emulsion of the
original print. Even so, going from these limits to 1200ppi or even

This limit does seem accurate for most of the color photo prints I've
scanned.

4800ppi with your 50 year old print is a major step so you need to ask
yourself if you really are seeing improved resolution, rather than just
improved image quality which can be achieved just as readily with the
lower resolution scan.

I don't have a web page, but if anyone wants a small example I can
e-mail or post to your preferred public site. I have no photography
background and am not the best to interpret my own results. However,
even at 4800ppi the image didn't look grainy, halftoned, or surface
pocked to my eyes.

You mentioned that aliasing dropped away when you went up to 1200ppi
from 600ppi on your photographic print, but are you sure it is
additional resolution that you are seeing and not just better
visualisation of fine detail because they are scaled up in size?

It's possible my terminology is off. I compared (separately) a gleam
on the lens of a pair of glasses and a single nostril at the same size
onscreen (i.e., both 1" across at different zoom %). I zoomed in a
few hundred % to compare the pixel detail. Both became smoother and
rounder as sampling resolution increased. The difference was too
pronounced to be caused by slight differences in resize %, and I could
zoom in farther on the higher resolution image before aliasing
appeared. Neither resolution's image showed significant signs of
media-related degradation before pixels became a factor.

Having said that, it is certainly possible that your 50-year old photo
print does have sufficient resolution to require 1200ppi or more,
especially if it was printed from a large format negative with little or
no enlargement. After all, only 10 years before that print was made an
awful lot of similarly printed aerial photographs were pored over under
high magnification on a daily basis - the analysts would not have used
magnification if the emulsion did not contain the resolution to justify
it. ;-)

Most of the hi-res images I've tested are "posed" family pictures,
probably taken by a professional. It will be interesting to see how
relatives' home-developed photos from long before 1950 compare
detail-wise.

Knowing that 1200+ ppi can be required, I can make my resolution
decisions based on each photo's particular condition. Out of
curiosity, was there a particular period when prints dropped under
600ppi en masse, such as the spread of cheap consumer cameras & photo
labs, the introduction of new color film formats, etc.? Or were the
majority always 600ppi or less, at least when intended for home use?

Regards,
false_dmitrii

 

[false_dmitrii]

But historic B&W prints were often/usually contact prints at original film
size, no enlargment at all. These are often a little better regarding
detail. 600 dpi is a reasonable try for those film-size B&W contact prints.
Assuming of course that the purpose needs that much image size.

Some of the oldest prints I'm looking at are badly faded (>100 years
old) and need to be transferred while there's still enough detail for
restoration. I'd prefer to restrict myself to my modest "expected
use" limits, but I also want to capture as much as possible for
improved future technology, other people's use of the images, etc. I
might very well end up running the scanner at 1200dpi or up. Time to
load up on CDRs!

I'd be interested in anybody's experience with different resolution
settings when dealing with 100-year-old prints for restoration and
long-term family "archiving" (possibly too strong a word). Have you
scanned at 600dpi only to find it insufficient sometime later? Many
of these are Alaskan landscape prints, so I suspect getting every
detail might matter more than with a group portrait. OTOH, they may
turn out to be too faded for such high resolutions to matter. I'll
have to try some today.

Can anyone comment on the distinction between the Epson's stated
resolution and the original photo's ppi? I'm still curious about
whether its "1200dpi" setting is capable of resolving 600ppi or better
of the original media. Still wonder also if there's a way to improve
detail at any given resolution setting. I thought I read something in
the NG about masking the empty scanning area with a nonreflective
black background to improve focus.

Wayne, your website is extremely useful. Does your book offer
additional advice related to the current discussion?

Thanks, Wayne et al., for your helpful information.
false_dmitrii

 

[Wayne Fulton]

Some of the oldest prints I'm looking at are badly faded (>100 years
old) and need to be transferred while there's still enough detail for
restoration. I'd prefer to restrict myself to my modest "expected
use" limits, but I also want to capture as much as possible for
improved future technology, other people's use of the images, etc. I
might very well end up running the scanner at 1200dpi or up. Time to
load up on CDRs!

1200 dpi still seems the wrong idea to me Dmitrii, but it also wont hurt
anything other then memory/file size and time and effort expended. Results
should be just as good.

The purpose of resolution is for enlargement purposes. If one scans at 1200
dpi, then one can enlarge to 4x size if printed at 300 dpi (1200/300 = 4x).

This works well for scanning film. If you indeed are planning to enlarge it
4x, then 1200 dpi is the way to try, that is what the numbers say, and there
is no other way to get there. However, my point was that photo prints are
limited well under that, and this is the reason that 4x results will be very
disappointing.. 3x will also be disappointing from a print. 2x may be
acceptable, in the same way that scanning at 300 dpi and printing at 150 dpi
may be acceptable. The obvious reason is that this is about all the color
print has to give. 4x enlargement is trivial for film, but its a Very Big
Deal from prints.

I dont mean to word it as "photo prints have 300 dpi of detail". The print is
not even digital. I only mean that scanning typical 4x photofinisher color
snapshots at more than 300 dpi simply wont do much more for us, it doesnt get
better as we might hope. Historic B&W contact prints can do a bit more. Film
can do a lot more, and if you have the way to scan the original film for a
comparison, the difference becomes very obvious.

Regardless of what the print may have to offer, if your final goal might be to
copy the prints at original size at 300 dpi, then 300 dpi is all you need for
an original size copy (no enlargement). Higher resolution cannot substitute
for detail that is faded away. Resolution is about tiny detail, not about
faded and missing detail. Tiny detail would seem to be the least problem in a
badly faded case.

But, it cant hurt either, other than in effort.

 

[Kennedy McEwen]

It's possible my terminology is off. I compared (separately) a gleam
on the lens of a pair of glasses and a single nostril at the same size
onscreen (i.e., both 1" across at different zoom %). I zoomed in a
few hundred % to compare the pixel detail. Both became smoother and
rounder as sampling resolution increased. The difference was too
pronounced to be caused by slight differences in resize %, and I could
zoom in farther on the higher resolution image before aliasing
appeared. Neither resolution's image showed significant signs of
media-related degradation before pixels became a factor.

Here is where I think you might be being misled by the image
manipulation package and confusion that with what you have heard/read.

When you "zoom" an image on something like Photoshop or PaintShop Pro
(or most other packages, these are just probably the two most popular
examples) what happens is each pixel on the image is represented as a
square block of pixels on the display. So on a 200% zoomed image you
see each pixel as a block of 2x2 display pixels on your screen. This
makes the image appear larger, but it also makes the pixels themselves
more visible. Pixel visibility is *NOT* aliasing.

When you start with a higher sampling density (more real sampled ppi in
the scanner) then obviously you will need to zoom up greater to see the
pixels at the same scale. However, because the higher sampling density
is exceeding the resolution capacity of the original photograph the
pixels will be smoother and more continuous. A similar result can
obtained by upscaling the lower ppi image to the same size using a
resampling algorithm like bilinear, bicubic or Naczos interpolation.
Only by comparing this image with the higher resolution scan will you be
able to tell if the original low resolution scan was actually too low to
capture all of the finest detail in the original image. Zooming up the
image won't show you that because most of the information you will see
in the zoomed version is just pixel artefacts which are invisible to the
naked eye at normal size - and impossible for many printers to even
reproduce.

If you read through Wayne's pages (or buy his book) you will find quite
detailed explanations of the difference between zooming for view and
scaling by different interpolation methods, particularly the more common
types available in most image processing packages.

Aliasing, as opposed to pixel visibility, is the corruption of fine
detail or repeated patterns so that they appear much larger on the final
image than they actually are in the original. A typical example is a TV
presenter's checker pattern jacket, being aliased by the colour
subcarrier and producing much larger, very visible patterns in each
colour, resulting in clear colour fringing. Another example is viewing
two picket fences, one through the other. As you change the distance
between you and the two fences, you will see a large pattern of light
and dark areas which change size with your distance and move to either
side very rapidly as you change your point of view. That large pattern,
generated by sampling the image of the second picket fence through the
apertures in the first, is aliasing.

On film and prints the grain can alias but, since the grain is a random
pattern, what results is a coarser random pattern. However, in order to
produce aliasing, the scanner must be capable of reproducing contrast in
finer detail than its sampling density can support. The Epson scanner
(and most flatbeds these days) is designed to have minimum contrast
reproduction of detail finer than the sampling density of the scanner,
so it is actually designed to avoid aliasing. Hence, even though you
increase the resolution of the scan, unless you scan sufficiently fine
to resolve the individual grains, you will not see any grain aliasing
(or any other type of aliasing) in the image.

 

[false_dmitrii]

Regardless of what the print may have to offer, if your final goal might be to
copy the prints at original size at 300 dpi, then 300 dpi is all you need for
an original size copy (no enlargement). Higher resolution cannot substitute
for detail that is faded away. Resolution is about tiny detail, not about
faded and missing detail. Tiny detail would seem to be the least problem in a
badly faded case.

But, it cant hurt either, other than in effort.

My main problem is trying to account for the unforseen uses I or
others might have for the photos in the future. 20 years from now
someone might want to pull the same image up to fill an 8-foot-wide
computer screen. Who knows? If I'm going to be the one preserving
the images, I'd like the original capture to be the best it can be for
practical purposes--preserving all of the meaningful data, and perhaps
a bit more, without obsessing over every grain or ink blob. But I
need to follow up on Kennedy McEwen's advice first.

With computers, I usually have no trouble focusing on "today's uses".
With these pictures, though, I don't want to handicap future use if I
can scan them properly right now.

false_dmitrii

 

[Hecate]

I'd be interested in anybody's experience with different resolution
settings when dealing with 100-year-old prints for restoration and
long-term family "archiving" (possibly too strong a word). Have you
scanned at 600dpi only to find it insufficient sometime later? Many
of these are Alaskan landscape prints, so I suspect getting every
detail might matter more than with a group portrait. OTOH, they may
turn out to be too faded for such high resolutions to matter. I'll
have to try some today.

May I make the suggestion that if you have Photoshop you purchase a
copy of Karen Eismann's book on Photo Restoration and work on them
first before doing any printing?

 

[false_dmitrii]

Here is where I think you might be being misled by the image
manipulation package and confusion that with what you have heard/read.

I'm afraid I haven't read or heard much of anything about resizing
theory. Or imaging physics, etc.

When you "zoom" an image on something like Photoshop or PaintShop Pro
(or most other packages, these are just probably the two most popular
examples) what happens is each pixel on the image is represented as a
square block of pixels on the display. So on a 200% zoomed image you
see each pixel as a block of 2x2 display pixels on your screen. This
makes the image appear larger, but it also makes the pixels themselves
more visible. Pixel visibility is *NOT* aliasing.

<snipped explanation>
You know, I almost went down this direction in my earlier post, but I
decided it wasn't close to your meaning. It's a bit confusing to
distinguish the original image's "detail potential" from the original
media's "grain limit". Is a good example the difference between the
level at which grain appears on a negative and the level at which even
slight camera shaking can blur out details that could have been
captured? Thus, while lines may continue to get smoother when
enlarged, they don't contribute anything new.

If you read through Wayne's pages (or buy his book) you will find quite
detailed explanations of the difference between zooming for view and
scaling by different interpolation methods, particularly the more common
types available in most image processing packages.

I think I should re-read his site in a month or so once my experience
is solid enough for me to absorb whatever I missed the first time. I
might even get the book just for the added convenience.

<snipped aliasing explanation>
I still can't apply the term accurately, but your descriptions were
clear and helpful. I'm used to it as a display driver "anti" setting.

On film and prints the grain can alias but, since the grain is a random
pattern, what results is a coarser random pattern. However, in order to
produce aliasing, the scanner must be capable of reproducing contrast in
finer detail than its sampling density can support. The Epson scanner
(and most flatbeds these days) is designed to have minimum contrast
reproduction of detail finer than the sampling density of the scanner,
so it is actually designed to avoid aliasing. Hence, even though you
increase the resolution of the scan, unless you scan sufficiently fine
to resolve the individual grains, you will not see any grain aliasing
(or any other type of aliasing) in the image.

Regarding grain, as I reexamined and resized the test scans that
triggered this post, I picked out slight blotchy contrast differences
that must be "grain" or its B&W print equivalent. (Is there a
different term?) However...

I resized a small selection from my 300,600,720,and 1200ppi scans. I
took the selection from the exact same relative position on each
image. I then bicubicly resized the selections to 4800ppi in Paint
Shop Pro 8 and Photoshop Elements 2(better at this) and compared them
to my native 4800ppi scan. Each selection resized to 960x960 pixels.
The 600 and 720 became more acceptable when resized, but there were
still pixelly gradations along edges vs. the 4800 with both at 100%
zoom. The 1200 was nearly indistinguishable and, oddly, slightly
smoother than the 4800 in the PSE2 resize.

(PSP8 and PSE2 are my "current need" concessions. Pricey PSCS might
be better in some ways, but I don't need it to get good full-res scans
or produce good images for current use.)

Is there superior resizing software that might make up the 600 to
1200ppi difference? Is Shortcut PhotoZoom any good? If I can
"restore" the quality when needed, I don't mind doing the original
scan at less than full potential.

I'd really like to get your own opinion on the comparison if you can
spare the time. Is there any way to provide you with the files,
either by e-mail or posted to your preferred website? There are 5
originals, ranging from 4k to 400k as PSP8 LZW-compressed 8bit
grayscale TIFFs. It's much easier to provide the originals and trust
you to resample them properly. I'll change format if you prefer
another.

Thanks for all advice so far.
false_dmitrii

 

[Bart van der Wolf]

SNIP

Is a good example the difference between the
level at which grain appears on a negative and the level at which
even slight camera shaking can blur out details that could have
been captured? Thus, while lines may continue to get smoother
when enlarged, they don't contribute anything new.

You can obviously not extract more detail than the original offers, but you
can extract less. There is nothing wrong with using the maximum
non-interpolated resolution your particular scanner can offer, except for
the amount of memory it requires. A benefit (so far not mentioned?) of doing
so is that it makes retouching a lot more accurate, because you probably
will downsize for output (although however little additional resolution you
get, your inkjet printer wiil use it).

Bart

 

[Bart van der Wolf]

SNIP

Is there superior resizing software that might make up the 600 to
1200ppi difference? Is Shortcut PhotoZoom any good? If I can
"restore" the quality when needed, I don't mind doing the original
scan at less than full potential.

You cannot restore what's not there to begin with but, since most resizing
is done because of printing requirements, IMHO nothing beats "Qimage" (not
for the price anyway). The Shortcut software (I tried the demo of S-Spline
Pro) is not bad, but for continuous tone images it offers nothing that
Qimage can't do better. If you're interested, follow this link
http://forums.dpreview.com/forums/read.asp?forum=1003&message=7642362 for a
discussion and a poll on preferred results.

Bart

 

[false_dmitrii]

May I make the suggestion that if you have Photoshop you purchase a
copy of Karen Eismann's book on Photo Restoration and work on them
first before doing any printing?

ITYM Katrin Eismann. Thanks for the recommendation, it looks like
something that would improve my work across the board. However, it
doesn't cover my resolution questions. Do you also recommend her
"Real World Digital Photography", which seems more relevant here?
Also, she is noted for explaining the theory behind the use of a
particular approach. Would her advice then be easy to translate to
image editors other than Photoshop? I haven't had much trouble moving
between PSP8 and PSE2 at sort of a self-taught (with much online
assistance) early intermediate level.

false_dmitrii

 

[false_dmitrii]

You can obviously not extract more detail than the original offers, but you
can extract less. There is nothing wrong with using the maximum
non-interpolated resolution your particular scanner can offer, except for
the amount of memory it requires. A benefit (so far not mentioned?) of doing
so is that it makes retouching a lot more accurate, because you probably
will downsize for output (although however little additional resolution you
get, your inkjet printer wiil use it).

If I had unlimited storage at two or more locations, I'd be finished
scanning by now. The big issue right now is the benefits of 1200
and 4800 dpi print scanning, as these settings can easily produce
files 100 to 500MB and up. I'd rather not have 1 image per CD if
possible. It helps considerably that these are B&W images--16bit
instead of 48bit. Or 8 vs 24 if enough people shout that 16bits per
channel is a waste of storage.

My current impression is that for my test photo, the 1200ppi image can
be scaled to 4800 without any meaningful difference, but any lower
resolution will suffer slightly, at least with my available resizing
tools. I'd love to get a second opinion on my image samples.

false_dmitrii

 

[Kennedy McEwen]

<snipped explanation>
Is a good example the difference between the
level at which grain appears on a negative and the level at which even
slight camera shaking can blur out details that could have been
captured? Thus, while lines may continue to get smoother when
enlarged, they don't contribute anything new.

Very similar indeed.

Regarding grain, as I reexamined and resized the test scans that
triggered this post, I picked out slight blotchy contrast differences
that must be "grain" or its B&W print equivalent. (Is there a
different term?)

Not really - grain is the correct term for B&W emulsions. In colour
terms it is usually "dye clouds" rather than true grain, but most people
still use the same term.

However...

I resized a small selection from my 300,600,720,and 1200ppi scans. I
took the selection from the exact same relative position on each
image. I then bicubicly resized the selections to 4800ppi in Paint
Shop Pro 8 and Photoshop Elements 2(better at this) and compared them
to my native 4800ppi scan. Each selection resized to 960x960 pixels.
The 600 and 720 became more acceptable when resized, but there were
still pixelly gradations along edges vs. the 4800 with both at 100%
zoom.

It is difficult to say whether this is due to slightly more information
being there in the 4800ppi scan or whether it was just the limitations
of the resizing algorithm that you were seeing. Given the following
though it is probably a bit of both.

The 1200 was nearly indistinguishable and, oddly, slightly
smoother than the 4800 in the PSE2 resize.

(PSP8 and PSE2 are my "current need" concessions. Pricey PSCS might
be better in some ways, but I don't need it to get good full-res scans
or produce good images for current use.)

Is there superior resizing software that might make up the 600 to
1200ppi difference?

There are a lot of resizing algorithms available and a whole field of
imaging science related to the development of improved algorithms each
improving the image quality over previous versions or achieving the same
quality with less computing power. Like a lot of things though, you
have to compromise somewhere - bicubic is just one compromise between
the ideal algorithm and realistic processing time today - in ten years
time it probably will be considered a relatively poor trade-off between
the two, given the anticipated increase in processing power available to
you by then.

Is Shortcut PhotoZoom any good? If I can
"restore" the quality when needed, I don't mind doing the original
scan at less than full potential.

You might like to try fractal resizing. This actually encodes the image
in a different, size independent, manner and then just resizes it to fit
on your screen or print to the size you want.

I'd really like to get your own opinion on the comparison if you can
spare the time. Is there any way to provide you with the files,
either by e-mail or posted to your preferred website? There are 5
originals, ranging from 4k to 400k as PSP8 LZW-compressed 8bit
grayscale TIFFs. It's much easier to provide the originals and trust
you to resample them properly. I'll change format if you prefer
another.

You can email them directly to me following the instructions below.
However, at the moment I am in the process of making some major
modifications to my computer, so I can't guarantee to pick up email for
some time.

 

[Hecate]

ITYM Katrin Eismann. Thanks for the recommendation, it looks like
something that would improve my work across the board. However, it
doesn't cover my resolution questions. Do you also recommend her
"Real World Digital Photography", which seems more relevant here?
Also, she is noted for explaining the theory behind the use of a
particular approach. Would her advice then be easy to translate to
image editors other than Photoshop? I haven't had much trouble moving
between PSP8 and PSE2 at sort of a self-taught (with much online
assistance) early intermediate level.

I'd recommend anything by Katrin.

(It was late at night, sorry <g>).

Yes, that's a good book too. The techniques she uses will all relate
to Photoshop, but a lot of her techniques are just as valid in PE2.
I'd don't know about PSP8, but would assume that they're adaptable.
With the exception of Martin Evening's Photoshop for Photographers,
I've rarely seen a book as clearly written, explained and helpful as
Katrin's.

 

[false_dmitrii]

Sorry for the delay, this was a busy week. Thanks for the additional
explanations. This all seems to be getting clearer.

There are a lot of resizing algorithms available and a whole field of
imaging science related to the development of improved algorithms each
improving the image quality over previous versions or achieving the same
quality with less computing power. Like a lot of things though, you
have to compromise somewhere - bicubic is just one compromise between
the ideal algorithm and realistic processing time today - in ten years
time it probably will be considered a relatively poor trade-off between
the two, given the anticipated increase in processing power available to
you by then.

If my original scan output can preserve all meaningful detail and be
"restored" to 4800ppi smoothness or better if it ever becomes
necessary, I'll be content.

You might like to try fractal resizing. This actually encodes the image
in a different, size independent, manner and then just resizes it to fit
on your screen or print to the size you want.

I remember encountering this option 10 years ago for an unrelated
project. I got the impression that in practice, it would seem much
like resizing standard vector graphics--lossless, and equal quality at
almost any size. Does the fractal format cause any difficulties when
working with popular photo editors? Can it be identical to the
original at the original ppi?

You can email them directly to me following the instructions below.
However, at the moment I am in the process of making some major
modifications to my computer, so I can't guarantee to pick up email for
some time.

Sent from alternate "megs" account prior to this post, about 500k
total. Your input is much appreciated.

false_dmitrii

 

[Kennedy McEwen]

Sorry for the delay, this was a busy week.

Same here - its amazing how much time an intermittent fault on an Intel
P4 CPU can take to identify! :-(

I remember encountering this option 10 years ago for an unrelated
project. I got the impression that in practice, it would seem much
like resizing standard vector graphics--lossless, and equal quality at
almost any size. Does the fractal format cause any difficulties when
working with popular photo editors?

The software I looked at was never really integrated with image editors,
just provided a means of storing the original image in a fractal format
which could then be opened, scaled to the required image size and stored
in conventional format for import into an image editor. Hardly
seamless, so I never went beyond evaluation.

Can it be identical to the
original at the original ppi?

The intention is that it isn't identical, but that the differences are
imperceptible.

Sent from alternate "megs" account prior to this post, about 500k
total. Your input is much appreciated.

For anyone else following this, Dmitrii sent me 5 images of the same
5x5mm area of the image scanned at 300, 600, 720, 1200 and 4800ppi in
8-bit grey scale.

Looking at the images it is clear that the 300ppi image lacks resolution
and image information that is present in the others, however the 4800ppi
scan has captured, in addition to the photographic image, a lot of
emulsion grain, defects (such as dust and scratches) and paper texture
which, I assume, is not what you are interested in preserving - although
it may be, in which case disregard the following.

By 600ppi all of the image content in the original photograph seems to
have been captured. I scaled this image in stages, first to 1800ppi
using bicubic, then applied a slight gaussian blur (0.7 pixel diameter)
to remove residual pixel visibility, then up to 4800ppi and finally a
slight unsharp masking (150% @ 2pixels) to compensate for the previous
blur. The resulting image looked very similar to the 4800ppi scan,
differing only in the visibility of grain and paper texture and defects
(scratches of the emulsion and dust etc.). Indeed the grain and paper
texture looks a little more contrasty in the scaled image than in the
4800ppi comparison, indicating I might have been a little too severe
with the final unsharp mask.

Whilst it is certainly possible that there is a very small imperceptible
loss of information from the original image at 600ppi my view is that
higher resolution scans do not add anything apart from noise and defects
to this and consequently can't be justified in terms of image content.

 

[false_dmitrii]

I owe you a response.

Thanks once more for your outside opinion. It helped me revise my
appraisal of detail levels. I now agree with your opinion of the
600ppi image (more below). However, to my uninformed eye, there were
subtle hints of detail in the 4800 that the 600ppi image could not
fully reproduce--for example, some faint lines of the eyeglass frame.
I don't think the test image needs to go past 600ppi, but it's enough
of a borderline that I'll continue to watch for exceptionally sharp
prints that might benefit. Nice to know the extra resolution might
still come in handy for non-transparent media.


<snip>

(re. fractal resizing)

The software I looked at was never really integrated with image editors,
just provided a means of storing the original image in a fractal format
which could then be opened, scaled to the required image size and stored
in conventional format for import into an image editor. Hardly
seamless, so I never went beyond evaluation.

I also recall claims of exceptional compression, but that was ages
ago. I can't remember the ratio. It's probably changed since then,
anyway.

<snip>

(re. my test images)

By 600ppi all of the image content in the original photograph seems to
have been captured. I scaled this image in stages, first to 1800ppi
using bicubic, then applied a slight gaussian blur (0.7 pixel diameter)
to remove residual pixel visibility, then up to 4800ppi and finally a
slight unsharp masking (150% @ 2pixels) to compensate for the previous
blur. The resulting image looked very similar to the 4800ppi scan,
differing only in the visibility of grain and paper texture and defects
(scratches of the emulsion and dust etc.). Indeed the grain and paper
texture looks a little more contrasty in the scaled image than in the
4800ppi comparison, indicating I might have been a little too severe
with the final unsharp mask.

The blur made the difference. Resizing directly to 4800 in PSE2 left
subtle but visible pixellation along curved edges. With your method,
it was indeed possible to nearly match the appearance of the 4800ppi
original starting from the 600ppi scan. I tried the USM at 100% and
was satisfied (not that I really know how to apply it properly...

And, of course, this is independent of dedicated scaling software or
future improvements in the PS/PSP algorithms. With all important
detail present at 600ppi, I'm comfortable setting aside the additional
resolution until it proves useful.

Any tips on independently acquiring knowledge such as when to resize
twice with a blur in the middle? I'd never have figured that out on
my own.

Whilst it is certainly possible that there is a very small imperceptible
loss of information from the original image at 600ppi my view is that
higher resolution scans do not add anything apart from noise and defects
to this and consequently can't be justified in terms of image content.

Out of curiosity, might some future algorithm do a better job at
resizing or sharpening the image when starting from a nice, clear set
of grains? It doesn't strike me as a forseeable concern, just the
only case in which the higher resolution would matter for my scans.

Regards,
false_dmitrii