Thanks for the noise treatise. Well written and easy to understand.
Chuck Norcutt
Moose wrote:
> Ralf Loi wrote:
>
>> , summing 2 or more different scans into one single image in order to
>> minimize scan noise.
>>
> Fernando Gonzalez Gentile wrote:
>> Could someone more skilled than myself, explain theory and practice of this
>> technique?
>>
> Fairly simple. You are an audiophile. Imagine a single groove on a
> record that doesn't spiral in and with nothing recorded on it at all.
> Put the needle on it and turn up the volume. Any hiss or other noise you
> hear is system noise. If that noise is truly random and you capture each
> rotation, you can add them together. As the number of samples increases,
> the apparent noise decreases, until it becomes a tiny DC voltage. It's
> still not zero, but is unvarying and thus not heard.
>
> That's system noise, and may indeed be eliminated by summing multiple
> samples.
>
> Now imagine that the record has been sitting around and played often. So
> there is another kind of noise, source noise. As the location of each
> bit of dust, pit, etc. is mostly fixed, summing samples will actually
> appear to enhance this noise, by lowering random noise so that the
> non-random source noise stands out.
>
> Now imagine sound recorded on such a groove at a very low level. Factors
> such as latency in the cutter head and the resilience of vinyl mean that
> most of the signal doesn't actually result in a wave in the cut.
> However, the loudest peaks do manage to get recorded, so there are
> scattered little bits of sound. They aren't random, as they are
> amplitude related. However, like the non-random source noise, they rise
> out of other, random noise when multiple samples are summed. Without the
> missing parts, they appear to be random noise - and - may appear to be
> made worse by summing.
>
> That's dynamic range clipping noise.
>
> The situation with film is much the same. Random noise from the scanner
> electronics will be reduced by multiple passes. Tiny dust, scratches,
> dried processing chemicals, etc. may actually become more prominent with
> multiple passes. Fortunately, you have ICE for that.
>
> Perhaps the trickiest situation, however, may the one where
> underexposure is sufficient that only relatively scattered bits of the
> light reflected from the subject were bright enough to activate the
> silver halide in the film grains. As most of the area is not light
> activated, development results in an area of clear backing with a
> scattering of dark dots. Although theoretically containing information,
> like the tops of mountains poking through clouds seen from above, they
> don't tell us anything about the obscured part of the landscape.
>
> Again, multiple passes will do nothing whatsoever to recover the missing
> detail, only separate the bits from any random noise. With slightly
> greater initial exposure, the general nature of the subject may become
> discernible. However, the parts clipped in the photographic process are
> still not recoverable through any scanning manipulation.
>
> This results in what now appears to be noisy image. And it is. However,
> as the noise is dynamic clipping in the source, summing multiple scan
> passes will not improve it. It's even possible, depending on scanner
> characteristics, for this kind of noise to appear to get worse with
> multiple passes. I don't know how common that may be.
>
> It IS possible for software to interpolate missing data based on
> surrounding pixels. Various RAW developers for digital camera images
> this to different extents. some of it is remarkably effective. Adobe
> Camera Raw, for example recovers considerably more natural appearing
> highlight detail from a clipped image that Canon's SPP processor.
> Whether there is such software for film scans, I don't know.
>
> I have written all of the above assuming we are talking about negative
> film. Just to be clear, it also applies to reversal (slide) film. In
> reversal film processing, the negative image is first developed, at
> which any noise as a result of clipping at either highlight or shadow
> level is established. Later processing steps chemically reverse the
> luminosity of the image, creating a positive image. However, that
> chemical process increases contrast, actually decreasing the dynamic
> range of the original negative stage, and clipping more luminance
> extremes than the first development step. That's why slides have less
> dynamic range than negative films.
>> My scanner is a Nikon 4000ED. I used to scan each slide at the maximum
>> number of passes the machine is capable of
> In light of the above, I can't see any benefit to multiple passes beyond
> what may be necessary to reduce system noise. It should be easy enough
> to find that by scanning a completely dark piece of film and observing
> the pixel level with various numbers of passes.
>
> As the 4000ED is a very good scanner, with large dynamic range, I
> wouldn't be surprised to find that the number of passes needed for
> maximum system noise reduction is small.
>> , until I read Ken's post on June 2 when he says five passes are more than
>> enough.
> Ken says nothing whatsoever in that post about noise reduction. In fact,
> he says:
>
> "BTW, why five-pass? Because the extra passes help define the shape of the
> grain."
>
>> I don't think I will change my habits, though :-)
>>
> A compulsive behavior disorder? Ken says"
>
> "I'm doing up to FIVE passes. Is that necessary for all images? Hardly.
> Get real. .... I'd slice my wrists if I had to do five pass scanning on every
> image."
>
>> But there are scans which are problematic, those slides which have details
>> in the shadows and careful adjustment of curves and luminosity and USM don't
>> seem to resolve accurately.
> Whether the problem is simply missing data or less than perfect PS
> technique, I can't tell from here.
>> I was thinking about doing something like HDR scanning, merging the
>> resulting files using CS3 - but I don't know if this makes any sense.
>>
> It will make no difference if the detail is indeed clipped. Using HDR
> software to create extreme tonal shifts and LCE, as so many people do
> will, however result in a very different looking image. Combining
> multiple exposures taken when the subject exceeds the dynamic range of
> the film would certainly help create a natural image with greater
> dynamic range.
>
> Moose
>
> ==============================================
> List usage info: http://www.zuikoholic.com
> List nannies: olympusadmin@xxxxxxxxxx
> ==============================================
>
>
> ------------------------------------------------------------------------
>
>
> No virus found in this incoming message.
> Checked by AVG.
> Version: 8.0.100 / Virus Database: 270.0.0/1488 - Release Date: 6/6/2008 5:48
> PM
==============================================
List usage info: http://www.zuikoholic.com
List nannies: olympusadmin@xxxxxxxxxx
==============================================
|