I still am not sure of this one. If dust (or scratches, I just say
dust because it's one word) is on top of a transparent region of the
film, why do you say it won't be detected? I suppose it will undergo
some dimming, but that should be all, or should it?
Because the reflective scan will generally only register what's on the
underside of the film. Depending on film density some "stuff" on top
may come through but it will be distorted and unreliable not to
mention overwhelmed by the underside reflections. (Also, see next
segment about focus.)
BTW, I read below you actually open the lid for the reflective scan. I
understand you're doing this to reduce the "transparency effect" but
by the same token this will further reduce data from the top as the
light from the lamp has nothing to bounce off of once it's gone
through the film. Any reflections from the top scratches or dust
itself will be very tiny and then further diffused by the film
substrate by the time they reach the sensors. Finally, I expect that
whatever weak data survives all that it will then be drowned by the
strong reflection from the underside of the film.
I'm afraid you're thinking about a different kind of scanner than the
one I'm using. I don't move the film at all between the two scans - I
just leave it on the glass, held by means of the film holder provided
with my scanner.
Ah, then you have a different problem: Focus!
In general, flatbed scanners are "fixed focus" to the top of the
glass. By placing the film holder on top you're moving the film away
from the scanner's range.
That's unimportant when you're performing a transparency scan because
in that case it's the projected image on top of the glass that's being
scanned and not the film surface.
(Indeed, if you want to improve the resolution, you may wish to use a
slide projector and project the image onto the glass to fill the whole
scan area! Some scanners actually come with a holder to keep the
scanner on its side. In my case this was undocumented and I was
puzzled for weeks what that funny looking part was for!
Failing that you can open the scanner lid at a 90 degree angle and
have the scanner rest on the lid. Since it's hard to focus on the
glass you may wish to place a sheet of paper on the glass to help with
the focus and then remove it before scanning. I haven't actually tried
all this, but it should work.)
However, when you do a reflective scan, this time it's the bounce from
the film itself that gets registered. And since the holder makes the
film "float" several millimeters above the glass it will be out of
focus.
Now then, flatbeds generally have a certain amount of depth of field
(except some, like Canon Lido series, which use a different
technology). However, for such finicky work like removing dust (and
scratches) I fear this inaccuracy will impact the algorithm.
I guess, you could improve this considerably by not using the film
holder at all and placing the film directly on the glass.
The only mechanical operation I do between the two scans is opening
the lid for the reflective scan (or closing it for the transmittive
scan, if I do the reflective scan first), so that as little light as
possible is reflected down.
Of course, slamming the lid down is not a good idea since that
obviously *will* move something - but you just have to move it down
gentlier.
Unfortunately, at this level of accuracy no matter how careful you are
there will be movement.
But more importantly, any time you do multi-pass multi-scanning (which
is in effect what you are doing) there will be registration problems.
Even if the film was perfectly in place the stepper motor will never
be in the same place as on the previous scan. Furthermore, again due
to scanner mechanics, the sensor array is bound to move laterally as
well. The end result will be misalignment on both axis.
You can check this by simply scanning an image twice. You'll notice
that even though nothing has changed between the two scans there will
be major misalignment. When I got my flatbed this was a big revelation
to me! I then kept reducing the resolution expecting that at some
point the images will be in sync but even at the lowest resolution of
50 they were still off!
Actually, I've been wrestling with multi-pass multi-scan image
alignment in a different context (film scanner) where the slide
remains in the scanner between the scans and - still - there is major
misalignment. What's worse this is on sub-pixel level, so I have to do
sub-pixel alignment before I can merge the two scans, which opens up a
whole new can of worms...
BTW, on my flatbed I don't have to open the lid to switch between
reflective and transparency scans. The only thing I have to do is
connect the cable that provides the power to the light source in the
lid, as well as indicate that the lamp in the scanner should be turned
off. Couldn't you just do that? It would improve things quite a lot
because no matter how careful you are, you are bound to move the
holder if you open the lid. Not to mention that by opening the lid
you're not pressing on the holder anymore so it's bound to "float"
even further from the glass.
It does almost correspond on my scanner. Mind you, I had to do an
(extremely boring) manual alignment between two test scans in order to
measure the shift my scanner introduces between reflective scanning
and transparency scanning.
So now I know that every transparency scan will be x=-2, y=13 pixels
off every reflective scan, and I just have to adjust every scan
accordingly (of course I've written a script to do that).
I fear that, for the reasons outlined above, this will actually differ
with each scan. Furthermore, there will also be sub-pixel
misalignment.
Here's a little trick/workflow I use in Photoshop to check the
(mis)alignment of two images:
1. Click on the magnifying glass.
2. Make sure "Resize Windows To Fit" checkmark is *ON*
3. Load an image.
4. Double-click the magnifying glass to get 100%
5. Press Control/+ until the image is 300% or 400%
6. Repeat for second image.
7. Use Control/Tab to toggle between the images.
NOTE: This is the only reliable way I found to actually have the two
*windows* aligned. Any other method appears to skew the windows'
alignment.
To move the image within the window use Home, End, PageUp and
PageDown, then toggle with Control/Tab and repeat for the other image
to keep them in sync.
Now, it's true that there *will* be some alignment problems between
any two scans, since the motor stepping can't be 100% accurate.
However, if it's good enough for VueScan to offer a multi-scan option,
it can't be too bad for my purposes.
Ah... VueScan... Hmmm... I'm not a fan, to say the least... ;o) Far
too buggy for my taste.
As I like to say, VueScan's multi-pass multi-scan option is a very
time consuming and complicated way to blur an image... ;o)
In any case the scripts I've written dilate every noise spot found in
the reflective scan so that these alignment problems should go
unnoticed. Clearly, this way you lose more image data that you must,
since the interpolated region will be somewhat larger than exactly the
size of the dust particle / scratch.
That's exactly it! What you perceive as spot removal is really loss of
data. You could probably achieve a very similar effect but simply
applying a small amount of Gaussian Blur.
It all depends on whether that's a price one feels like paying.
Well... Yes... But doing stuff like this is so much fun! ;o)
By the way, a fluorescent lamp like you find them in scanners doesn't
emit much infrared, right? If it did, I suppose one could place an
infrared-pass filtering material between the lamp and the film,
thereby doing the same thing Digital ICE does. That's assuming a
scanner's CCD is sensitive to infrared light, and it looks like mine
is (I scanned a remote control's emitter with a button pressed -
definitely seems to pick up infrared).
I don't know the spectral characteristics of a garden variety flatbed
lamp or CDD response but I wouldn't be surprised if some IR data is
picked up. The problem is separating this IR data from the rest
because that's what you need to perform dust removal. This may be
possible by applying Fast Fourier Transformation to extract only
certain (i.e., IR) frequencies but that's only a wild guess and way
over my head...
Don.
