Apache-Talk @lexa.ru 

Inet-Admins @info.east.ru 

Filmscanners @halftone.co.uk 

Security-alerts @yandex-team.ru 

nginx-ru @sysoev.ru 




      :: Filmscanners
Filmscanners mailing list archive (filmscanners@halftone.co.uk)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[filmscanners] RE: Pixels and Prints

Hi Paul,

> > But that doesn't mean that every combination of film/scanner has
> > noticeable
> > noise generated by these things in sky regions.
> I assume drum scanners do much better, but they're a heck of a lot more
> expensive than a Canon Digital Rebel.

As do high end CCD scanneras as well, and scanner operation is critical as
well.  Anyone can make a mess of most anything.

> > Noise does not have to be random.  It can be random, or
> > deterministic.  It's
> > still noise.  Anything that decreases fidelity is considered noise.
> >
> > Distortion is noise.  I really don't care what you want to call
> > it, and I'm
> > surprised you're arguing semantics here...instead of arguing the points.
> This is not a semantic issue. Noise is _fundamentally_ different from
> distortion.

Being a professional EE, and having designed many systems that deal
specifically with noise (signal testing, audio and video), I disagree.  I
know of no precise definition of noise that would exclude distortion from
being noise.  Noise can have many sources, as can distortion...and
fundamentally, they are the same...both are a reduction in the fidelity of
the signal.

> But the point here is that if a Bayer pattern generated "noise" then it
> would be filling in the pixels in some unpredictable, and therefore
> ultimately useless, manner. But it doesn't.

No one said the Bayer pattern generated noise.  It's the Bayer pattern
reconciliation (as in the interpolation used to fill in the missing color
information) that can induce noise.  As I've said, noise does not have to be
unpredictable.  It's simply a reduction in the fidelity of the reproduction
of the original image, period.  Call it what you want, I really don't care.

> On the sort of image
> detail that
> matters, modern Bayer interpolation algorithms do the Right
> Thing, and do so
> consistently and effectively.

You say that, but I KNOW it's simply not true, and you're using amorphous
terms like "do the right thing"...that's VERY unscientific, and hardly
quantifiable.  You also say it providing no basis for your claim, but your
claim.  If you have some actual data, please, I'd love to see it.  I have
direct personal experience with this, and I'm not sure you do.  If you do,
I'd like to hear it.

> So while it's theoretically possible that a Bayer camera will
> miss a red dot
> on a white wall, because light from the red dot happens to fall only on a
> red pixel, who cares?

That isn't what's being talked about.  It's not that it'll mis a red dot in
the middle of a white wall, but that a red wall is made up of many
gradients, and is not an even field.  It may interpolate a point 98,132,12
that is in reality 112,138,12.

> Qualitatively, Bayer sensors work extremely well, so
> it's closer to the truth to say that a six-million sensor Bayer chip
> produces a six megapixel image than to say that it really only produces a
> 1.5 megapixel image.

I agree completely with that, and have never said any differently (nor has
anyone else in this discussion that I am aware of).  But, to claim the Bayer
patter reconciliation is %100 spot on is simply wrong.  It may be "decent",
and in fact, quite "decent", but it's not perfect.  It is still, unarguably,
a reduction in fidelity.  The significance of that reduction is debatable,
but it's still a reduction.



Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe 
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or 


Copyright © Lexa Software, 1996-2009.