[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]
Re: [APML] Couple of Wide-Angle shots on Konica Centuria 400(and scanner noise suggestion)
Thanks Ray; we all learn a bit of something from each contribution to this
list. Good point about the linear accumulation of dark current; I didn't
think that through when I was firing off the idea OTTH, but bias
subtraction might hold promise.
Jerry L seems pretty well versed on scanner esoterica, I wonder if he might
jump in here if he knows anything about scanner internal bias subtraction?
Steve Holak
Senior Software Architect
Brokerage Concepts IS Dept.
610-491-4879
email: steve.holak@bcitpa.com
Ray Butler
<ray.butler@nuigal To: Discussion of Film Astrophotography <astro-photo@seds.org>
way.ie> cc:
Sent by: Subject: Re: [APML] Couple of Wide-Angle shots on Konica Centuria 400
astro-photo-bounce (and scanner noise suggestion)
s@seds.org
08/31/2004 12:55
PM
Please respond to
Discussion of Film
Astrophotography
Steve.Holak@bcitpa.com wrote:
> Jason,
> With digital images, averaging multiple exposures is the best way to
> bring signal above the noise (median combines are noisier than averages,
> but you get rid of abberant pixcels by median combining), so scanning the
7
> multiples and average combining them is the correct first pass at
reducing
> scanner noise.
I would agree with this. If the noise source is "random" and follows a
Gaussian distribution, then averaging or medianing will reduce it. If
your scanner is based on a CCD, then its readout noise follows a classic
gaussian bell-curve distribution. Combining 7 scans would reduce the
scanner readout noise contribution by the square-root of 7, which is
2.6x better.
That doesn't make the whole image 2.6x better overall of course - it may
still be dominated by the noise contributions from film grain, source
faintness, and background brightness. Source faintness is bad (as
everyone intuitively knows), because low light levels translate to
higher Poisson photon-counting noise as a _percentage_ of the signal. A
bright background is bad, because high light levels in the background
translate to higher Poisson photon-counting noise in _absolute_ terms,
which is additive to the source's own Poissonian noise.
Median combination is just averaging with more robustness against skew
by an outlying data point - a pixel which may be much brighter or
fainter than the rest of the pixels in its little stack.
> I would next do final combining of the two images, any
> histogram / curve adjustments, and save smoothing in SGBNR for last;
> smoothing will cause some loss of the original data so this should go
last.
I agree with this too. Avoid (or keep to the very end) anything which
messes with the spatial resolution or spatial frequency content of the
image (deconvolution, smoothing, unsharp masking, etc.). That's because
these operations all have one thing in common - they correlate the noise
across adjacent pixels. Once that is done, it can't be undone by further
processing. "Don't f**k with the pixels", as one of my heroes memorably
said.
> This is the approach with CCD imaging, so after scanning film you're
> essentially starting at the same point.
>
> While composing this, I was wondering if the CCD imaging concept of
> "dark frames" would apply to help reduce scanner noise with film images.
> Has anyone here considered or played with this? I imagine a series of
> blank or unexposed scans could be median combined, just like CCD darks,
and
> subtracted from the raw scans to remove any consistent noise induced by
the
> scanner from scan to scan. If it did have some positive effect, faint
> detail could be brought out more effectively. The step order would then
be
> to scan the negatives, create the darks, subtract the darks from the
scans,
> and then proceed to average the scans, etc.
When people talk about CCD "dark frames" they are usually really talking
about bias+dark frames. Some even confuse plain bias frames (zero second
exposures) with bias+dark frames (long exposures with no external light
source). Anyway, it is a convenient short cut to perform bias
subtraction and dark subtraction in one pass, which is what I think
Steve is referring to. This works fine, if the bias level is not
changing over time, and if the dark frame is of the same effective
exposure time as the science frame. Strictly speaking, you should
subtract the individual bias levels from the dark frame, and call the
result the "dark frame"; or if you have several dark frames sum or
average them after bias subtraction, and call that the "dark frame".
Separately you would subtract the bias from the science frame. Then if
necessary scale the debiased "dark frame"to the same exposure time as
the science frame; and subtract it from the debiased science frame.
Doing this in a one-er pass might seem to be the obvious thing to do,
but then you can't average multiple bias frames for superior debiasing.
True (debiased) dark frames consist _only_ of dark current (thermally
generated noise), and in any given pixel this grows linearly with time.
But since the time it takes a scanner CCD to scan any given pixel in the
image is only a fraction of a second, there will be essentially zero
dark current in that pixel - insufficient time has elapsed for a
thermally generated electron to pop up in that pixel. The rate of dark
events is higher than in an astronomical imaging CCD, since scanner CCDs
are not cooled, but the extremely short exposure time per pixel makes up
for that.
So, basically, there is no point in worrying about subtracting true dark
noise. Having said that, there may be a point in investigating the
subtraction of scanner _bias_. I have not had a chance to properly test
this, but I am not entirely sure whether bias subtraction is done
internally in a scanner or not. The operating manual will never go into
such details. Bias being equally present whether the exposure is long or
short, it is something that might be hiding in all our scanned images!
> I don't have a film scanner other than a flatbed, but maybe it's
worth
> a try for someone to check it out.
Even with a flatbed (which is all I have too) all of these things apply.
So I think it is worth investigating alright.
Where I completely like the way Steve is thinking is that he is taking
universal concepts of signal and noise contributions (which most people
only come to grips with in digital formats) and applying them to
photographic imaging. This is something that I myself really believe in.
It is the best way to really understand what you can capture with any
image - including a photograph - and how best to process it from
scanning onwards. The experienced photographers and darkroom whizzes who
pioneered combining negatives in various ways, unsharp masking,
manipulation of film's characteristic curves, etc. were all getting at
this via the photographic idiom, but I'm not sure they all fully
understood the mixture of physics, statistics and information theory
which underpins it. I haven't mastered it yet either, but I keep working
at it.
Ray "sorry for the long lecture" Butler.
> Steve Holak
> Senior Software Architect
>
> Brokerage Concepts IS Dept.
> 610-491-4879
--
Dr. Ray Butler (ray.butler@nuigalway.ie || ray@physics.nuigalway.ie)
Lecturer, Dept. of Physics || Computational Astrophysics Laboratory
National University of Ireland, Galway, University Road, Galway, Ireland.
Tel: +353-91-524411 ext. 3788 FAX: +353-91-525700
_______________________________________________
Astro-Photo mailing list
Astro-Photo@seds.org
http://seds.org/mailman/listinfo/astro-photo
_______________________________________________
Astro-Photo mailing list
Astro-Photo@seds.org
http://seds.org/mailman/listinfo/astro-photo