How much memory for high res scans?

[Druid]

I have an Epson 2400 scanner, when doing scans from negs or slides at
high resolution my PC seems to 'grind to a halt' , can anyone advise
if it would be worth doubling my ram?

ATM my system is -: Ath Xp2600, Win Xp home, 512Mb PC2700 DDR ram.

I was considering adding another 512Mb, has anyone tried adding ram
for the above reason and does it really help?

John.

 

[J. A. Mc.]

I have an Epson 2400 scanner, when doing scans from negs or slides at
high resolution my PC seems to 'grind to a halt' , can anyone advise
if it would be worth doubling my ram?

ATM my system is -: Ath Xp2600, Win Xp home, 512Mb PC2700 DDR ram.

I was considering adding another 512Mb, has anyone tried adding ram
for the above reason and does it really help?

Is that system using a USB 1.1 or true 2.0 bus for connection of the
scanner?

I've much the same configuration, just Toshiba branded and am running the
3200. I don't notice any slow down even when at 3200 on a 6cm negative.

 

[Tim]

And the size of the file you are trying to scan is? (This will be
indicated at the bottom of the preview window in the scanner
interface.)

 

[Druid]

Is that system using a USB 1.1 or true 2.0 bus for connection of the
scanner?

USB 2.0

 

[Druid]

And the size of the file you are trying to scan is?

Not sure, TBH I have not had the scanner very long, did a few photo
scans and thought I might try a neg or slide, assuming that I would
need hi res I tried to scan at the max res but it took so long that in
the end I got fed up waiting and aborted!

I have done neg and slide scans at lower settings ok, but most
articles on scanning suggest that you need a high res for negs/slides?

I assumed that I must need more memory than 512 for hi res scans?

John.

 

[CSM1]

Not sure, TBH I have not had the scanner very long, did a few photo
scans and thought I might try a neg or slide, assuming that I would
need hi res I tried to scan at the max res but it took so long that in
the end I got fed up waiting and aborted!

I have done neg and slide scans at lower settings ok, but most
articles on scanning suggest that you need a high res for negs/slides?

I assumed that I must need more memory than 512 for hi res scans?

John.

Just make sure that you are scanning at the max optical resolution and not
an interpolated one.

The optical resolution is the first number in the spec. such as 2400 X 4800.
This means that the CCD can resolve 2400 pixels per inch of the scanner bed
horizontally and the stepper motor can move the carriage 4800 pixels per
inch down the scanner bed.

Interpolation takes a long time because the software is creating pixels
where there were none.

 

[Kennedy McEwen]

Just make sure that you are scanning at the max optical resolution and not
an interpolated one.

The optical resolution is the first number in the spec. such as 2400 X 4800.
This means that the CCD can resolve 2400 pixels per inch of the scanner bed
horizontally and the stepper motor can move the carriage 4800 pixels per
inch down the scanner bed.

Interpolation takes a long time because the software is creating pixels
where there were none.

The second number (4800 in your example) is NOT interpolated and those
pixels are not being created by software, they are real. New, real
pixels are captured at every stepper motor position. Clearly that takes
twice as long to gather when the motor steps 1/4800in each time than
when it steps 1/2400in at a time.

The only interpolation is in the axis of the CCD, and some scanner
drivers do not interpolate at all in that axis, simply creating
rectangular pixels by replicating samples in two square pixels. Even
when they do interpolate the intermediate pixels, adding two numbers and
left shifting one place is one of the fastest operations that a
microprocessor can implement, so the time added for pixel interpolation
along the CCD axis is negligible compared to the step and exposure time.

 

[Mac McDougald]

The second number (4800 in your example) is NOT interpolated and those
pixels are not being created by software, they are real. New, real
pixels are captured at every stepper motor position. Clearly that takes
twice as long to gather when the motor steps 1/4800in each time than
when it steps 1/2400in at a time.

The only interpolation is in the axis of the CCD, and some scanner
drivers do not interpolate at all in that axis, simply creating
rectangular pixels by replicating samples in two square pixels. Even
when they do interpolate the intermediate pixels, adding two numbers and
left shifting one place is one of the fastest operations that a
microprocessor can implement, so the time added for pixel interpolation
along the CCD axis is negligible compared to the step and exposure time.

========
Oh, come on Kennedy...

If there were no interpolation happening to create the final square
pixels at 4800ppi, the scanner would have to be 4800x4800 rather than
2400x4800. Not to claim that 4800 scan on such a scanner often does
resolve more information than 2400.

I realize your high end knowledge of the science of optics and sampling,
both in theory and practice; but in more pedestrian terms, unless we
wind up with non-square pixels with different ppi rating for the two axes
OF those pixels, there is interpolation going on here after the hardware
does its thing.

Mac

 

[Mac McDougald]

.... Not to claim that 4800 scan on such a scanner often does
resolve more information than 2400.

Should read, "not to deny"...

M

 

[Kennedy McEwen]

========
Oh, come on Kennedy...

If there were no interpolation happening to create the final square
pixels at 4800ppi, the scanner would have to be 4800x4800 rather than
2400x4800. Not to claim that 4800 scan on such a scanner often does
resolve more information than 2400.

I realize your high end knowledge of the science of optics and sampling,
both in theory and practice; but in more pedestrian terms, unless we
wind up with non-square pixels with different ppi rating for the two axes
OF those pixels, there is interpolation going on here after the hardware
does its thing.

As I said, any interpolation, when it occurs, is in the axis of the CCD
(the 2400ppi direction), not the scan motion axis, which is truly
4800ppi. Whether the scanner has enough optical resolution to exploit
the increased sampling density is no different a question than whether
it has enough resolution to exploit the 2400ppi sampling density in the
axis of the CCD.

Neither figure actually defines resolution, whether pedestrian or
Olympic sprint champion. Irrespective of that fact, no interpolation is
actually needed, simple pixel replication in one axis to write a
rectangular pixel in a traditional square pixel file format. Where it
is performed by the scanner driver, it is simply one dimensional linear
interpolation: add the two adjacent pixels and divide by two - an
operation most modern CPUs can perform in a couple of clock cycles.
Given Druid's 2.6GHz Athlon, just how long do you think that is going to
take, even for a few thousand pixels?

 

[Kennedy McEwen]

Should read, "not to deny"...

I am still unsure of your meaning. Presumably you are concerned that I
am suggesting that a 4800x2400ppi scanner specification means that the
scanner can resolve more than the limit of a 2400x2400ppi sampling
density. Would that even a scanner specified at 2400x2400ppi actually
resolve as much! Unfortunately, I have never seen a scanner
manufacturer actually specify its product's resolution at all, merely
the sampling density it collects.

However, my comment took issue with the specific statement that anything
beyond the lower spec figure is interpolated - it isn't and implying
such, particularly as an explanation for another issue, is misleading at
best.

 

[Mac McDougald]

However, my comment took issue with the specific statement that anything
beyond the lower spec figure is interpolated - it isn't and implying
such, particularly as an explanation for another issue, is misleading at
best.

Well, forgive the layman's misunder standing, but I still don't see how
anything over the optical capabilities of the sensors is NOT
interpolated.

Let's put it this way: will a 2400 x 4800 device like we're discussing
resolve more line pairs at 4800ppi than at 2400ppi? (assume that we don't
have optical limitations to even the 2400 figure). I'm sure the answer is
no.

Why don't we simply have 300 x 4800 scanners, if the longer axis is NOT
interpolated. Stepper motors are cheaper than CCDs I'd think. Seems your
premise would indicate that a 300 x 4800 would generate same quality
image as 2400 x 4800, since each would NOT use interpolation in
generating the final 4800 ppi image.

Mac

 

[Kennedy McEwen]

Well, forgive the layman's misunder standing, but I still don't see how
anything over the optical capabilities of the sensors is NOT
interpolated.

Well, what happens when you shoot Provia through a cheap lens. The lens
has a limited optical resolution, well below that of the film. There is
no numerical processor in the signal path to conduct any interpolation.
However the grain from which the image on the film is composed, is
certainly well beyond the optical resolution of the system. No
interpolation, but real data beyond the optical resolution of the
sensor.

Now, what would you call this if the same strip of film is scanned on a
Howtek at 8000ppi? Again, each sample is perfectly unique and a true
representation of what the scanner sees on the film. No interpolation
is undertaken, yet there are far more samples than are necessary to
resolve all of the optical information.

Interpolation is quite specific, it is the synthetic creation of samples
which have not been actually captured by the sensor. Samples are quite
independent of the resolution - the only criteria is that you need *at
least* two samples to resolve anything. Note the "at least" in the
above statement. That means that any practical system will always have
more samples than the optical resolution, since no system is going to be
perfectly designed so that every unit is critically sampled with exactly
two samples at the optical resolution.

In the case of a scanner which is specified as 2400x4800ppi, the CCD
samples 2400 points in the image every inch. The stepper motor steps
the CCD so that the image can be sampled 4800 unique positions every
inch in the orthogonal axis. Neither figure says anything about the
resolution of the image, the sensor or the entire system. However these
are real samples, they are *not* interpolated.

Let's put it this way: will a 2400 x 4800 device like we're discussing
resolve more line pairs at 4800ppi than at 2400ppi? (assume that we don't
have optical limitations to even the 2400 figure). I'm sure the answer is
no.

Well, your answer is wrong. Without optical limits then the sampling
density itself defines the resolution, the limit being half the sampling
density. So, of course the scanner will resolve more in the 4800ppi
axis than it will in the 2400ppi axis!

Why don't we simply have 300 x 4800 scanners, if the longer axis is NOT
interpolated.

No practical reason at all - but would you want a scanner which had 16x
as much resolution in one axis as it has in the other?

Many systems do have such a different resolution, in fact, every time
you go to the cinema you are looking at an image which has around 50%
more resolution in the vertical axis as it has in the horizontal - the
film camera uses an anamorphic lens to compress the image horizontally
on the film and the projector uses a inverse anamorphic lens to stretch
it back out. The result is an image with substantially less resolution
in the horizontal axis than in the vertical - but have you ever noticed?

Stepper motors are cheaper than CCDs I'd think.
Nope.

Seems your
premise would indicate that a 300 x 4800 would generate same quality
image as 2400 x 4800, since each would NOT use interpolation in
generating the final 4800 ppi image.

Not at all. Quite specifically, resolution and interpolation are
completely separate things and, in fact, have nothing whatsoever to do
with each other. A 300x4800 image can only support 1/8th of the
resolution that a 2400x4800 image can support. However, just because
the resolution is limited to 150cy/in, which a 300ppi scam can perfectly
adequately resolve, it does not mean that a 1200ppi, 4800ppi or
gazillion billion ppi scan is interpolated. If they are real samples,
captured from the image at a gazillion billionths of an inch apart then
they are NOT interpolated, they are real.

Furthermore, from a legal perspective, if a scanner is obtaining
additional samples through interpolation then these days it has to be
declared. That is why the specification for the scanner at the start of
this thread quite specifically states:
"
2400 x 4800 dpi (Optical)
12800dpi (interpolated)
"
The 2400x4800ppi is NOT interpolated, they are real samples. The
4800ppi may be much finer samples than the optical resolution of the
scanner, but then so can the 2400ppi - indeed the Nyquist Samplng
Theorem specifically states that they must be at least that fine.

 

[Leonard Evens]

The 2400x4800ppi is NOT interpolated, they are real samples. The
4800ppi may be much finer samples than the optical resolution of the
scanner, but then so can the 2400ppi - indeed the Nyquist Samplng
Theorem specifically states that they must be at least that fine.

I haven't seen the beginning of this thread, so perhaps I am missing
something. I understand the point you are making, but I have one
problem with it. If the scanner samples at 2400 ppi along the array
and steps so it samples at 4800 ppi, there is still the problem that all
the scanning software I've seen requires you to specify either that you
are scanning at 4800 ppi or at 2400 ppi. In the former case, it would
seem that interpolation needs to be done on one axis, and it the latter
case, it would seem that averaging has to be done on the other axis.
In practice, most people would treat 2400 ppi as the maximal obtainable
optical scanning resolution, I think.

Any comments?

 

[Kennedy McEwen]

I haven't seen the beginning of this thread, so perhaps I am missing
something. I understand the point you are making, but I have one
problem with it. If the scanner samples at 2400 ppi along the array
and steps so it samples at 4800 ppi, there is still the problem that
all the scanning software I've seen requires you to specify either that
you are scanning at 4800 ppi or at 2400 ppi. In the former case, it
would seem that interpolation needs to be done on one axis, and it the
latter case, it would seem that averaging has to be done on the other
axis. In practice, most people would treat 2400 ppi as the maximal
obtainable optical scanning resolution, I think.

Any comments?

Yes, read the thread! ;-)

This issue has been addressed specifically. The comment was made that
the 4800ppi specification of a 2400x4800ppi was in excess of the optical
resolution and, specifically that the 4800ppi resolution was
interpolated and consequently slowed the scan. All of these comments
are, simply, wrong. On the occasions when interpolation is performed
with a 2400x4800ppi scanner to produce 4800ppi scan, the interpolation
is in the 2400ppi axis, and is single axis linear interpolation which a
modern CPU can perform faster that the exposure time of the linear array
itself.

 

[Mac McDougald]

Yes, read the thread! ;-)

This issue has been addressed specifically. The comment was made that
the 4800ppi specification of a 2400x4800ppi was in excess of the optical
resolution and, specifically that the 4800ppi resolution was
interpolated and consequently slowed the scan. All of these comments
are, simply, wrong. On the occasions when interpolation is performed
with a 2400x4800ppi scanner to produce 4800ppi scan, the interpolation
is in the 2400ppi axis, and is single axis linear interpolation which a
modern CPU can perform faster that the exposure time of the linear array
itself.

I never said anything re speed of the process.
My simple claim was that anything above 2400ppi for the scan on a
2400x4800 device is using interpolation to generate the final pixel
count.

You strongly deny this in great detail in your posts, yet you just said
above,
"On the occasions when interpolation is performed
with a 2400x4800ppi scanner to produce 4800ppi scan"..

So you are saying here that indeed interpolation takes place above the
optical spec for the device.

So, hmmmm??

Mac

 

[Kennedy McEwen]

I never said anything re speed of the process.

No need to be paranoid, Mac, there are other people writing to the
thread than yourself and the first time that interpolation was raised in
the thread it specifically considered the effect it had on scan speed.
Indeed, speed was the issue in the thread and interpolation was
erroneously suggested as a cause. You indeed did not mention speed and
I was not referring to your comments when I addressed speed either.

My simple claim was that anything above 2400ppi for the scan on a
2400x4800 device is using interpolation to generate the final pixel
count.

You strongly deny this in great detail in your posts, yet you just said
above,
"On the occasions when interpolation is performed
with a 2400x4800ppi scanner to produce 4800ppi scan"..

So you are saying here that indeed interpolation takes place above the
optical spec for the device.

So, hmmmm??

Again, read the thread Mac. This has already been explained in
significant detail. Achieving 4800x2400ppi does not invoke
interpolation if the scanner is so specified. Converting that image,
with the non-standard rectangular pixels onto a conventional square
pixel graphic format of 4800x4800ppi *may*, but not necessarily, invoke
interpolation. When it does so, the interpolation is simple one axis
linear interpolation, and in the original 2400ppi direction, not the
4800ppi direction of the specification. However, many scanner drivers
simply pixel replicate rather than interpolate, writing two pixels to
the file for each 2400ppi sample.

 

[Mac McDougald]

No need to be paranoid, Mac, there are other people writing to the
thread than yourself and the first time that interpolation was raised in
the thread it specifically considered the effect it had on scan speed.
Indeed, speed was the issue in the thread and interpolation was
erroneously suggested as a cause. You indeed did not mention speed and
I was not referring to your comments when I addressed speed either.

Again, read the thread Mac. This has already been explained in
significant detail. Achieving 4800x2400ppi does not invoke
interpolation if the scanner is so specified. Converting that image,
with the non-standard rectangular pixels onto a conventional square
pixel graphic format of 4800x4800ppi *may*, but not necessarily, invoke
interpolation. When it does so, the interpolation is simple one axis
linear interpolation, and in the original 2400ppi direction, not the
4800ppi direction of the specification. However, many scanner drivers
simply pixel replicate rather than interpolate, writing two pixels to
the file for each 2400ppi sample.

I don't have any way to save an image except in ONE ppi, Kennedy.
It's either 2400 or 4800, in both directions.

I've never had an image that reported 2400 x 4800 ppi.
YOU can do that? Does PhotoShop CS, with its support for non square
pixels actually do that? Does the Image Size Dialogue box report vertical
x horizonal ppi?

Mac

 

[Kennedy McEwen]

I don't have any way to save an image except in ONE ppi, Kennedy.
It's either 2400 or 4800, in both directions.

That is why TWO pixels are written to the standard graphics image file
for each sample, you get 4800pixels per inch in the 2400 sample per inch
axis - no interpolation is necessary!

I've never had an image that reported 2400 x 4800 ppi.

Really? Assuming that you view NTSC or PAL video then you almost ALWAYS
see images which have originated in a digital studio in an anamorphic
format with more resolution horizontally (720 samples) than vertically
(480 samples in NTSC or 576 samples in PAL). Not only isn't it unusual,
it is downright commonplace these days.

YOU can do that? Does PhotoShop CS, with its support for non square
pixels actually do that? Does the Image Size Dialogue box report vertical
x horizonal ppi?

What to the limitations of your viewing software have to do with the
scanner performance?

 

[Mac McDougald]

That is why TWO pixels are written to the standard graphics image file
for each sample, you get 4800pixels per inch in the 2400 sample per inch
axis - no interpolation is necessary!


Really? Assuming that you view NTSC or PAL video then you almost ALWAYS
see images which have originated in a digital studio in an anamorphic
format with more resolution horizontally (720 samples) than vertically
(480 samples in NTSC or 576 samples in PAL). Not only isn't it unusual,
it is downright commonplace these days.

What to the limitations of your viewing software have to do with the
scanner performance?

Well, you know, Photoshop 7.
And 6, 5.5, 3 before that.
Reports an image in ppi. For entire image.
There has been something higher end than this for professional use all
along?

M