WARNING: The following is a rather long discussion of the "yellow stain"
effect seen on ACER Scanwit 2720s reported here a week or so ago. I am
posting this in hopes of getting more details on the actual working of the
scanwit from someone "in the know" - such as optics, sensor design, etc. If
you don't have a Scanwit or if you don't have this problem you may want skip
these ramblings of a somewhat inept newbie... :-)
...
I spent a full day Wednesday running scans using both VS and Mira and
believe I am closing in on what is going on. However, I still don't know if
the issue is with a limited number of Scanwit's or is a general
characteristic of the units. (If you own a scanwit, please look at some
sky's and let me know if you see this Yellow stain effect at all!) I have
notified ACER's tech support via email (in the US) and I am hoping they will
reply.
Meanwhile, where I am right now is this. On my unit (and Jerry's) it appears
that the light is "brighter" at the center of the negatives (I am looking at
negatives for now - I will look into slides this weekend) than it is on the
edges. While there appears to be a slight difference on the short edges, the
dominant effect is seen on the long edges. I have not verified yet if the
scanwit scans vertically (up and down in relation to the carrier) or
horizontally - but I am working on the assumption that it scans
horizontally.
The result could be characterized like this:
1111111111111111111111111111111111111111
2222222222222222222222222222222222222222
2333333333333333333333333333333333333332
2344444444444444444444444444444444444432
3444444444444444444444444444444444444443
3444444444444444444444444444444444444443
3444444444444444444444444444444444444443
3444444444444444444444444444444444444443
3444444444444444444444444444444444444443
3444444444444444444444444444444444444443
2344444444444444444444444444444444444423
2333333333333333333333333333333333333332
2222222222222222222222222222222222222222
1111111111111111111111111111111111111111
The numbers represent intensity of light getting to the CCDs on a scan of
clear film. They do not represent actual or relative intensity - just the
general pattern of darker on the edges.
This effect is very minor. It is only seen when the density of the image
causes the light to be attenuated to the point of reaching the threshold of
sensitivity of the CCD's.
It also appears that the sensing of blue is more affected than other colors
(or the optics in the scanwit attenuate the blue at the edges) - I don't
know if Scanwith uses separate sensors for each color or reuses the sensors
with different filters - if they use separate sensors, then I would say the
blue sensors are slightly less sensitive than the red and green at low light
levels. If they use filters, then I would say the filter for blue attenuates
the light slightly more than the red and green filters - resulting in a
lower signal level.
Just like the EV curve for film, semiconductors (CCDs) have an operational
characteristic curve which has a linear range and a low knee and I high
plateau. It is desired to keep devices operating in the linear portion of
the curve. When the light falling on a sensor is less than some threshold
unique to each sensor it will turn off (just before it turns off it will
"blink" on and off randomly resulting is random low level noise). Slightly
more than this and the device will behave in a non-linear fashion, slightly
more yet and it will begin operating in a linear fashion. Since each
sensor's absolute sensitivity is slightly different there is a very small
range of light values which will result in some of the sensors being on and
some being off. This is my theory on explaining the "1 pixel wide lines" I
am seeing in my (and Jerry's) scans. (I have also seen these lines in scan's
of the Q60 target on Ed. H's site from more expensive scanners - although
you have to look closely and use some imagination to "connect the dots".)
These occur only in areas of extreme density or lack of density - i.e.. at
extreme points on the operational curve of the device.)
The "yellow stain" appears to be a function of operating the sensors in the
non-linear portion of the bottom of the sensors curve. What I think is
happening is that this non-linear portion of the curve is slightly higher in
the blue sensors (they are less sensitive) than in the red and green
sensors. So, when you scan a negative that has a very dense area (bright
sky) the blue begins to have less output relative to the red and green at
the same light intensity levels. This would result in an overall color shift
in areas of extreme density towards yellow (less blue) in skies while the
rest of the image remained correct. I find that I need to add a touch of
blue in the highlight of my scans using the curves tool to accurately
reproduce the scene. I have tested this by taking a picture of a
Gretag/Mcbeth chart and scanning the resulting negative. The highlights tend
to be slightly low in blue - this of course may be a result of the film
used, which would invalidate my entire theory! :-0
I am very tempted to take my scanwit apart to see what the optic's look
like. I wonder how they get the light across the width (not length) of the
frame? If they are using a moving light source (I don't thing this is
reasonable), or if they use lenses. If they use a lense to spread the light
(vertically), then possibly this effect is due to a lense effect - I am out
of my water here, but something like "chromatic aberration?" where the
different colors are focused (bent) at slightly different places could
result in the blue being slightly out of focus at the extreme edges of the
scan, as you move in towards the center the three colors would converge.
This could also explain why some scanners appear to exhibit the problem more
than others, since the characteristics of this lense (if in-expensive) would
vary from unit to unit. I could certainly understand how an engineer being
asked to reduce cost could consider a less expensive lense in this function,
since it's job is simply to spread the light out across the frame and
another lense (higher quality) would be responsible for actually focusing
the light on the image.
That is my theory on what is happening - an inexpensive lense is being used
to spread the light vertically and is attenuating the blue slightly at the
edges. The BIG question is "why is VS different than Mira?" (Jerry has
reported that he is seeing this in both VS and Mira - to me this implies a
difference between his scanner and mine in degree of effect.)
I have made quite a few scans of the same negative using both programs and
the results are fairly consistent.
1. Mira does have the yellow stain but it is very attenuated (on my
scanner) - in most cases it takes some imagination to find evidence of it.
2. VS always has the stain, but scans can be made by adjusting the white and
black balance manually such that the effect is almost as good (not present)
as with Mira.
3. Focus does not appear to play a part in the stain being present. However,
I have noticed on my unit that Mira appears to give a slightly better
(sharper) focus using auto-focus compared to VS. VS can be focused to match
the Mira by setting the focus manually. (This was a surprise, I expected VS
to have the better focus than Mira since it consistently gives more detail!)
If my assumptions are correct then I would expect this effect (the
difference between VS and Mira) to be a result of using different operating
bias's (voltages) on the CCD's or light source (which would affect the
sensitivity of the devices.) However, I don't know if the software is able
to set the operating points on the devices, or if that is controlled in
hardware. Another way this could be explained would be if the software were
affecting the exposure (level or duration) on the devices - but again, the
Scanwit has automatic exposure control, which I assume can not be affected
by the software. (Maybe the exposure is affected indirectly via the way VS
focuses?)
One of the things frequently said about VS is that it has superior ability
to "pull detail out of dark/dense areas" of scans. This implies to me that
somehow Ed H. is using a wider portion of the operational curve of the
device (i.e.. closer to the low and high ends.) This would agree with my
findings, in that by pushing the operation of the device closer to the lower
non-linear region of the operating curve you would get more detail in the
dense areas, but you would also be more susceptible to variations in
individual sensors (pixel to pixel across the scan.) This appears to be true
also. In looking at very bright and very dark areas of images on this
negative, when I turn up (or down) the level in these areas I can see more
variation in the pixels with VS than with Mira. In most cases this variation
can be attributed to more detail - however, it also contains "noise". A
quick easy test to see this noise is to over sharpen an image using unsharp
mask. Oversharping a scan made with Mira is "slower" to result in
blatant/random noise "sparkles" in dark areas than the equivalent amount of
USM on a VS scan of the same image. (Note this effect can be compensated for
somewhat by using multiscans in VS.)
what does all this mean -
1. I have a broken scanner and none of it matters - I don't think so, but
maybe...
2. On images which have a relatively narrow contrast range (4-5 stops?) it
probably doesn't matter at all since the entire range of signals (data)
generated from the image will fit nicely within the linear operating range
of the sensors, and the auto exposure system will adjust the exposure to use
this linear range. Remember, this effect only comes about when you are
scanning at the extreme ends of Dmax or Dmin of the scanner. This is one of
the spec's that you pay for in buying better equipment - so, the scanwit
should - since it is less expensive - exhibit this more than a $2,000 (or
$10,000) scanner.
3. On images which have wide ranges of contrast (high Dmax/Dmin - 5-7
stops?) values, this becomes a problem, since the scanwit has auto exposure
control and we cannot set it to choose the operating range of the sensors
(i.e., select to blow out the highs or lows). These scans a doomed to
undesirable artifacts resulting from operating in non-linear regions of the
sensors.
4. Since there is a difference between the output of VS and Miraphoto I have
hope that there is a software solution (or at least improvement to be made)
to the problem. Ideally I would want a program with the workflow, color
accuracy, and detail resolution of VS without the color stains! Right now I
have to pick between the two.
I still have the examples of this effect posted at my website
( ) This weekend I will be posting a couple
of methods for hiding/correcting the yellow stain effect using Photoshop.
The methods are not perfect, but they can salvage an unusable photo in most
cases.
/fn
