I recently wrote a little article about getting better B&W
negative scans from your scanner by scanning negatives in color and
discarding all but one color channel to get to grayscale. You can
read that article here.
I had just done that article when I went to a friend's house for
a visit. Lo and behold, he had just the thing I wanted to do a
bit more testing - a 2" x 2" chrome on glass USAF resolution test
target.
And, he was willing to let me borrow it. I did, and I took
it home and scanned it on my Microtek ArtixScan 1800f using both the
current version (6.52) of the Microtek scanner software for that scanner,
and version 6.1 of Silverfast Ai.
The first scans
All of the following were scanned on my Microtek
ArtixScan 1800f at 1800DPI, no sharpening, full range scans. The
test target is a chrome on glass positive (chrome pattern, clear glass
background) USAF 1951 resolution test target from Edmund Optics. I
scanned the entire test target, but I've cropped the scans down to the
center section (Groups 2 and below) to make it easier to present here.
Color scan, Microtek
Color scan, Silverfast
Grayscale scan, Microtek
Grayscale scan, Silverfast
At first glance, those Microtek scans look pretty
damn good. Nice and crisp, I thought.
But, look a bit more closely. In particular,
attend to the center portion of the scan, where groups 4 and below
are. In the Silverfast scan, we can easily pick out the
numerals identifying Group 4 and Group 5. Group 6 and 7 are
visible as smudgy blobs in the middle of the group 4 and 5
arrangement. They're entirely missing from the Microtek scan!
The USAF test target consists of bar patterns. The white bars
should be the same width as the black bars. They're certainly
the same in the Silverfast scan, but look at the Microtek scan - the
black bars are narrower than the white bars.
The reason why the Microtek scans looks so sharp is
that although I set the scanner software to do no sharpening, it did
some sharpening anyway. From what I can tell, it did a LOT of
sharpening - so much, in fact, that it threw away some resolution.
From my point of view, that's a Bad Thing.
Rather to my amazement, it turns out what scanner
software you use makes a difference, and the manufacturer's software
in this case is the clear loser.
Below is what we get if we sharpen those Silverfast
scans.
Silverfast red channel, sharpened 200% radius 0
threshold 0
Observant readers will notice I've picked the
red channel here, when I'd picked the blue channel as cleanest
in the previous article. The resolution test target makes
it a lot easier to pick out which channel is best!
As a side issue, we can (from the Silverfast scan)
get an estimate of the resolution of the scan. It might be
smudged up a bit on this web page due to jpg compression, but in the
original scans, I can pick out the lines for Group 5, Element 1
(which corresponds to 32 lp/mm) and can kind of make out Group 5,
Element 2 (35.9 lp/mm) (it's a bit blurry, and the vertical bars are
more easily made out than the horizontal ones). I can't quite make
out Group 5, Element 3 (which corresponds to 40.3 lp/mm)
Doing the arithmetic, and appealing to Nyquist's
theorem, we can translate these into DPI. 32 lp/mm = 812.8 lp/inch.
Each pair needs two samples (one black, one white) so group 5
element 1 corresponds to 1625 DPI. Hey, we get that with ease.
Group 5 element two works out to 1818. If we call this patch
the limit of what we can resolve, it works out to just about the
stated 1800 dpi resolution of the scanner. Color me impressed.
It's worth noting that the scanner seems to do a bit
better resolving things horizontally than vertically.
Horizontally in this case is perpendicular to the direction the scan
head travels. I suspect the better resolution horizontally is
basically an effect of this - no travel one way, travel the other.
It's also interesting that the color fringing occurs primarily on
the vertical axis and not on the horizontal axis.
But what about density?
Having convinced myself that scanner software matters when we
consider resolution, the question comes up about which software will
eke out the best performance in reading various densities.
We turn back to our friend, the Stouffer 31 step .1 density step
wedge. Here are two scans, one with Microtek software, one
with Silverfast.
Microtek full range, resized from 300 dpi scan,
red channel
Silverfast full range, resized from 300 dpi scan,
red channel
These two scans were both done with the range sliders set as wide
as possible, so that we're not clipping highlights or lowlights.
The first, important observation is that in the scan done with
the Microtek software, there's no difference between step 1 and step
2. This is the thinnest portion of the negative. If you
have densities in your negative that are below the step 2 in the
Stouffer step wedge, you're not going to get them using the Microtek
software. This is actually a problem I've known about for a
long time. I've contacted Microtek about it, and I've gotten
no response.
The second thing is to observe that the two scans have different
tonal distributions. That, in and of itself, is not a big
deal. You can move the tonality around with curves in
Photoshop. But to my surprise, there's detail deeper into the
densest part of the negative with Silverfast - note that you can see
hints of the steps in steps 22, 23, 24. Playing around in
Photoshop with the inverted image, my take on this is that it's
easier to get detail into the dense part of the negative with
Silverfast. Note that I'm completely ignoring the noise issue
here.
I'd also note that the density of Step 31 should be 3.0 logD
above step 1, which is essentially fb+f. Call it a density of
3.1 logD, and you'd be close. So much for the much touted 4.2
logD range of this scanner. Hmpf.
Conclusions
It turns out that I've been doing most of my scans with the
Microtek software, thinking that surely the manufacturer's software
would be tweaked out and get best performance from the scanner
hardware. Beyond that, when I first started scanning my 4x5
film, I wasn't very experienced, and honestly, I just didn't see the
difference between the results from the two different pieces of
software.
But these simple tests, taking far less than a day, I think I've
pretty conclusively demonstrated that there's a difference - a BIG
difference. Silverfast extracts more resolution from the
negative, primarily because when you tell it not to sharpen the
scan, it doesn't. In contrast, the Microtek software sharpens
the scan heavily, even when told not to, and in the process it
throws away resolution that I care about. The resolution
winner is clearly Silverfast.
On top of that, the Microtek software clips off all lowlights
(the thinnest part of the negative) below about .15 logD density.
Sure, a properly exposed negative has shadow density higher than
that, but why clip it off? On an underexposed negative, this
amounts to the difference between a scan that's usable, and one
that's worthless.
What a fool I've been. I wish I'd bought the damn test
target a year ago when I bought the scanner. I could have
saved myself a boatload of wasted time.
And from now on, I'm scanning things with Silverfast.
Silverfast scanner software can be purchased online at:
http://www.silverfast.com.
