Matt Ion wrote in message news:9nhfd.23441$%k.13864@pd7tw2no...
Lorenzo J. Lucchini wrote:

What do you think? I *have* actually tried this stuff and it *appears*
to work and remove a number of defects from the final image - although
many dust spots remains, and I haven't been able to find out whether
the system works differently with dust that's *on the film* and dust
that's *on the scanner*.

Interesting concept, and should somewhat work in theory, although it
will only work for "noise" that's between the film/negative and the
scanner head. Dust that's on "top" of the film (facing the upper light
head) won't be detected by the reflective scan.

I'm not sure. It's true that there is dust that doesn't get detected,
but after a few experiments my opinion is that it's either because
they're too small, too dark to reflect enough light, or out-of-focus
(ie on the glass).

What you say is probably true if I do the "defect scan" with the lid
closed, which reduces contrast between image data and noise data, so
making the particles that lie on top of the film too dim for
detection; with the lid open, however, that doesn't seem to be the
case.

(Actually, it obviously *will* be the case for particles that lie on
top of a *dark region* of the film, but those particles won't do much
damage to the final image anyway)

Also, if the noise is
severe enough to block light on a transmissive scan, the reflective scan
may give you a "cancellation" image for it, but the information in the
film itself will still be obscured by it.

Sure, but AFAICS this applies to real Digital ICE as well - if
information is lost, it's lost, but we can try to mask the loss by
interpolation or some like technique.

On the topic of the noise-masking algorithm... what I do now is simply
- create a very blurred version of the scan
- superimpose the blurred version onto the original version, using the
"defect scan" as an alpha channel so that the blurred version is only
shown where there are defects

However, this doesn't seem to work very well, since the color of the
final image looks quite a bit different from what you'd expect in the
noise spots.

What's a better idea? I've read Digital ICE simply divides every image
pixel by the corresponding pixel in the "defect scan" (an IR scan in
ICE's case).
After trying this, though, it doesn't seem to work well at all: the
features of dust spots in my "defect scans" probably differ a lot from
ICE's IR scans.

Now I'm entertaing myself with the idea of using a median filter,
which seems to preserve pixel colors fairly well while removing most
defects; what do you think?

I'm not at home at the moment, but I'll upload some of my scans as
soon as I'm back.

by LjL