S
Simon
Hi all,
I've had my Minolta Elite 5400 for a while now, and have so far only
made scans of individual images. However, in the near future I want to
start the task of scanning the large stack of negative films I've
accumulated over the years. To make sure I do this as efficiently as
possible, I investigated the quality and scan times for different
settings. I got some interesting results I'd like to share with you. I
also hope that somebody can tell me whether my conclusions make sense.
I'm running windows XP and have connected the scanner using USB2 (add-in
card). The scanner firmware version is 1.10. All scans are done with the
Minolta scan software, version 1.1.5, in 16-bit linear mode. I also have
Vuescan, but I did not use it, as it does not have ICE and its exposure
controls are more limited. The scans were all made in negative mode,
with manual exposure settings. I determined the proper exposure using an
empty leader image and making sure the background was close to white
without overexposing it (exposing to the right). In negative scanning
mode, this actually means moving the film base peak to the left of the
histogram, due to the inversion.
scan times
==========
The results I got were as follows:
GD off, ICE off
resolution CPU usage time
prescan 0:14
540/675 30% 0:17
900/1350 55% 0:33
1800/2700/5400 100% 1:05
GD on, ICE off
resolution CPU usage time
prescan 0:26
580/675 10% 1:07
900/1350 15% 2:14
1800/2700/5400 30% 4:27
GD on, ICE on
resolution CPU usage time
prescan 0:48
580/675/900/1350 100% 4:27
1800/2700/5400 100% 8:53
As you can see I grouped different resolution settings on the same line,
as they led to exactly the same CPU usage and scan times. It appears
that the scanner only has three different settings internally. Selecting
1800 ppi actually causes a 5400 ppi file to be sent to the computer,
which is downsized afterwards in software. A small quirk is that when
using ICE, there only seem to be two modes.
In going from one mode to another, the scan times roughly double. The
reason for this is somewhat of a mystery to me. The data stream for 5400
ppi is not 2, but 16 times larger than for 1350 ppi, so this seems to
imply that the data transfer is not a bottleneck. So my guess is that it
has to do with the scanning itself. However, as the difference is only a
factor two, it seems that the 1350 ppi mode actually is a lot closer to
the 5400 ppi mode than one might think. More evidence for this can be
found later in this message (see 'scan quality').
Another observation is that enabling GD causes the scan time to
quadruple, due to the lower light intensity. This time is doubled again
when ICE is switched on as well. This agrees with statements by Ed
Hamrick saying that the Minolta performs a separate RGB and IR exposure
- the exposure times are apparently equal.
Finally, I have reason to believe that the 100% CPU usage in many of the
figures above is not an actual limitation of the scanning speed. It is
more probably the result of sloppy programming. I believe this because
when I used the 5400 ppi setting without GD and ICE and I *decreased*
the exposure time (see below), the scan time did not decrease anymore.
In fact, the scanner came to intermittent stops as my computer could not
keep up. As this only happened then, I assume that in all other cases my
computer could keep up.
exposure controls
=================
Next, I played around with the exposure sliders a bit. I found that they
actually influence scanning times, so they are hardware controls. There
is no difference between adjusting the master slider and adjusting the
RGB sliders by the same amount. When scanning colour negatives, the blue
channel is limiting the scan time. It needs to have the longest exposure
time of all RGB channels due to the orange mask. Therefore adjusting the
blue exposure slider has an immediate effect on the scan time. The same
does not happen for the red slider, until the adjustment becomes very
large. This implies that the RGB exposures take place at the same time.
I also used the resulting scans at different exposures to check the
linearity of the scanner. I opened two raw scans with different
exposures into Photoshop as different layers, the top one using the
'difference' blending mode. In between I put a Levels adjustment layer
and used the sliders to scale the RGB values (no gamma adjustment!). I
could get an almost perfect black every time, implying very good
linearity of the scanner response.
I used the same method to compare the plain scans to those with GD and
ICE enabled. Again a (near) perfect match. So the GD and ICE functions
have a linear effect on the scan exposure. This is good, because it
means that besides the scan time increase, they have not detrimental to
scan quality.
scan quality
============
Intuitively, it is better to make a scan at 5400 ppi and downsample to a
lower resolution than to scan at the lower resolution directly. By
averaging over multiple pixels a lower noise floor might be abotained.
However, the scan times for 1800 ppi and 5400 ppi are exactly the same,
suggesting that they might not be so different after all.
To check this I scanned an empty frame at 1350, 1800 and 5400 ppi with
GD and ICE off. I compared the width of the film base peak in the red
channel using the photoshop CS histogram, both in the original
resolution and downsampled to the lower resolutions. Downsampling should
decrease the width of the distribution, producing a 'cleaner' file.
After reading about the scan times above, it might not be surprising to
you that the widths of the 1800 distribution and the 5400 distribution
resized to 1800 ppi were roughly the same, whereas the peak width at
native 5400 ppi was significantly wider. Apparently even for a 1800 ppi
scan the full 5400 ppi resolution is used.
More surprising is that when scanning at 1350 ppi the width got even
smaller. In fact, it was the same width I got when I resized the 1800
ppi and 5400 ppi scans to 1350 ppi. Even though the scan time is a
factor two lower, it seems to use the same amount of information!
Stange, right? I don't know what's going on here, but I do know that
selecting 1350 ppi gives pretty good quality at decent scan times.
workflow
========
After performing all these experiments, I think I will settle on the
following workflow for batch scanning. I will use 1350 ppi for most of
my scans, yielding files that are big enough to print on 10x15cm at 300
ppi. I will switch on ICE (and thus GD), as many of my negatives are a
bit dusty and underexpose by one unit (approx. 1.7 times shorter) to
keep the scan times down. My preliminary tests show that this hardly
damages the quality of scans of negatives.
This approach takes approx. 2:30 per image, or 15 mins for a strip of 6.
That is just doable, and not having to worry about dust-hunting is a big
plus. For the few images that are enlargement-worthy I will take
advantage of the full 5400 ppi + ICE + GD capability of the scanner and
spend 10 minutes to get the scan done.
I hope some of you found this information useful. If anyone can fill in
the gaps, I would be very pleased!
Regards,
Simon Tindemans
I've had my Minolta Elite 5400 for a while now, and have so far only
made scans of individual images. However, in the near future I want to
start the task of scanning the large stack of negative films I've
accumulated over the years. To make sure I do this as efficiently as
possible, I investigated the quality and scan times for different
settings. I got some interesting results I'd like to share with you. I
also hope that somebody can tell me whether my conclusions make sense.
I'm running windows XP and have connected the scanner using USB2 (add-in
card). The scanner firmware version is 1.10. All scans are done with the
Minolta scan software, version 1.1.5, in 16-bit linear mode. I also have
Vuescan, but I did not use it, as it does not have ICE and its exposure
controls are more limited. The scans were all made in negative mode,
with manual exposure settings. I determined the proper exposure using an
empty leader image and making sure the background was close to white
without overexposing it (exposing to the right). In negative scanning
mode, this actually means moving the film base peak to the left of the
histogram, due to the inversion.
scan times
==========
The results I got were as follows:
GD off, ICE off
resolution CPU usage time
prescan 0:14
540/675 30% 0:17
900/1350 55% 0:33
1800/2700/5400 100% 1:05
GD on, ICE off
resolution CPU usage time
prescan 0:26
580/675 10% 1:07
900/1350 15% 2:14
1800/2700/5400 30% 4:27
GD on, ICE on
resolution CPU usage time
prescan 0:48
580/675/900/1350 100% 4:27
1800/2700/5400 100% 8:53
As you can see I grouped different resolution settings on the same line,
as they led to exactly the same CPU usage and scan times. It appears
that the scanner only has three different settings internally. Selecting
1800 ppi actually causes a 5400 ppi file to be sent to the computer,
which is downsized afterwards in software. A small quirk is that when
using ICE, there only seem to be two modes.
In going from one mode to another, the scan times roughly double. The
reason for this is somewhat of a mystery to me. The data stream for 5400
ppi is not 2, but 16 times larger than for 1350 ppi, so this seems to
imply that the data transfer is not a bottleneck. So my guess is that it
has to do with the scanning itself. However, as the difference is only a
factor two, it seems that the 1350 ppi mode actually is a lot closer to
the 5400 ppi mode than one might think. More evidence for this can be
found later in this message (see 'scan quality').
Another observation is that enabling GD causes the scan time to
quadruple, due to the lower light intensity. This time is doubled again
when ICE is switched on as well. This agrees with statements by Ed
Hamrick saying that the Minolta performs a separate RGB and IR exposure
- the exposure times are apparently equal.
Finally, I have reason to believe that the 100% CPU usage in many of the
figures above is not an actual limitation of the scanning speed. It is
more probably the result of sloppy programming. I believe this because
when I used the 5400 ppi setting without GD and ICE and I *decreased*
the exposure time (see below), the scan time did not decrease anymore.
In fact, the scanner came to intermittent stops as my computer could not
keep up. As this only happened then, I assume that in all other cases my
computer could keep up.
exposure controls
=================
Next, I played around with the exposure sliders a bit. I found that they
actually influence scanning times, so they are hardware controls. There
is no difference between adjusting the master slider and adjusting the
RGB sliders by the same amount. When scanning colour negatives, the blue
channel is limiting the scan time. It needs to have the longest exposure
time of all RGB channels due to the orange mask. Therefore adjusting the
blue exposure slider has an immediate effect on the scan time. The same
does not happen for the red slider, until the adjustment becomes very
large. This implies that the RGB exposures take place at the same time.
I also used the resulting scans at different exposures to check the
linearity of the scanner. I opened two raw scans with different
exposures into Photoshop as different layers, the top one using the
'difference' blending mode. In between I put a Levels adjustment layer
and used the sliders to scale the RGB values (no gamma adjustment!). I
could get an almost perfect black every time, implying very good
linearity of the scanner response.
I used the same method to compare the plain scans to those with GD and
ICE enabled. Again a (near) perfect match. So the GD and ICE functions
have a linear effect on the scan exposure. This is good, because it
means that besides the scan time increase, they have not detrimental to
scan quality.
scan quality
============
Intuitively, it is better to make a scan at 5400 ppi and downsample to a
lower resolution than to scan at the lower resolution directly. By
averaging over multiple pixels a lower noise floor might be abotained.
However, the scan times for 1800 ppi and 5400 ppi are exactly the same,
suggesting that they might not be so different after all.
To check this I scanned an empty frame at 1350, 1800 and 5400 ppi with
GD and ICE off. I compared the width of the film base peak in the red
channel using the photoshop CS histogram, both in the original
resolution and downsampled to the lower resolutions. Downsampling should
decrease the width of the distribution, producing a 'cleaner' file.
After reading about the scan times above, it might not be surprising to
you that the widths of the 1800 distribution and the 5400 distribution
resized to 1800 ppi were roughly the same, whereas the peak width at
native 5400 ppi was significantly wider. Apparently even for a 1800 ppi
scan the full 5400 ppi resolution is used.
More surprising is that when scanning at 1350 ppi the width got even
smaller. In fact, it was the same width I got when I resized the 1800
ppi and 5400 ppi scans to 1350 ppi. Even though the scan time is a
factor two lower, it seems to use the same amount of information!
Stange, right? I don't know what's going on here, but I do know that
selecting 1350 ppi gives pretty good quality at decent scan times.
workflow
========
After performing all these experiments, I think I will settle on the
following workflow for batch scanning. I will use 1350 ppi for most of
my scans, yielding files that are big enough to print on 10x15cm at 300
ppi. I will switch on ICE (and thus GD), as many of my negatives are a
bit dusty and underexpose by one unit (approx. 1.7 times shorter) to
keep the scan times down. My preliminary tests show that this hardly
damages the quality of scans of negatives.
This approach takes approx. 2:30 per image, or 15 mins for a strip of 6.
That is just doable, and not having to worry about dust-hunting is a big
plus. For the few images that are enlargement-worthy I will take
advantage of the full 5400 ppi + ICE + GD capability of the scanner and
spend 10 minutes to get the scan done.
I hope some of you found this information useful. If anyone can fill in
the gaps, I would be very pleased!
Regards,
Simon Tindemans