Impossible Camera !!

David J. Littleboy wrote:


> Noise implies a loss of information, and while the gross appearance of the
> noise can be hidden, you don't get the information back.
>

Noise does not imply a loss of information - usually exactly the reverse
applies - images which are heavily processed to remove noise lose
information.

That's why I shoot with noise reduction turned OFF on the Dimage A2. The
result is a picture with sharper detail.

The Kodak DCS Pro SLR/n also offers a noise reduction option, and sure,
you can have a smooth looking mush like very fine painting by numbers if
you really want that from an image. Or you can have a sharp, highly
detailed image which has a grain structure considerably better than
scanning Velvia 100 35mm.

I know which I prefer.

I was very unhappy with the Minolta A2 at first because I expected the
Extra Fine JPEGs to be just as good as the 7Hi but larger - on my 7Hi
they are impossible to tell from TIFFs, to the extent that 2X size
doubled TIFF saved files from JPEG shots have got through Alamy's QA
without question.

But after three months of testing, I conclude that the A2 files have
just that little extra edge in resolved detail. Comparing an 8 megapixel
A2 shot with a perfectly focused and calibrated 2400 dpi Leaf scan from
35mm, I am left in no doubt; the A2 is better. And it's WAY better than
a desktop scan even at 5400 dpi. (The Leaf is maybe not quite up to drum
scan standards but still beats any of the desktop scanners).

I guess we are all incredibly hung up examining 'noise' on digital files
when for the most part, even the noisiest ones around are far superior
to the 'noise' created by grain (or grain/scan artefacts) on film.

David

"David Kilpatrick" <iconmags3@btconnect.com> wrote in message
news:c9kn30$kj1$1@titan....
> David J. Littleboy wrote:
>
> > Noise implies a loss of information, and while the gross appearance of
the
> > noise can be hidden, you don't get the information back.
>
> Noise does not imply a loss of information

Of course it does. The presence of noise means that the bits that hold that
noise hold noise, not information. You can zero those bits or do more
sophosticated averaging of adjacent values, but the information that those
bits would have held had you captured without noise was lost when the image
was created. That's the information that was lost.

> - usually exactly the reverse
> applies - images which are heavily processed to remove noise lose
> information.

Of course, that doesn't mean that some particular noise reduction system
doesn't lose even more information. FWIW, I was quite impressed with how
nicely NeatImage cleaned up 4000 dpi scans of Reala without damaging the
image. Of course the 25 minute processing time per 645 frame made it rather
impractical.

> I guess we are all incredibly hung up examining 'noise' on digital files

Noise trashes fine/subtle textures and makes rescuing shadow detail
harder/impossible.

> when for the most part, even the noisiest ones around are far superior
> to the 'noise' created by grain (or grain/scan artefacts) on film.

Yes, dSLR ISO 50/100 noise levels are what make digital competitive with
film at radically lower limiting resolution levels.

But consumer camera ISO 100 noise levels aren't all that much better than
Velvia 100F and Provia 100F 4000 dpi scans, and downsampling to 2400 dpi
(enough for 8x enlargement) reduces it further.

David J. Littleboy
Tokyo, Japan

"Lionel" <nop@net> wrote in message
news:7qjvb0peu6ldtjdbo1o850ov7vsu0gcche@...
> Kibo informs me that "David J. Littleboy" <davidjl@gol.com> stated that:
>
> >
> >"David Kilpatrick" <iconmags3@btconnect.com> wrote in message
> >news:c9kn30$kj1$1@titan....
> >> David J. Littleboy wrote:
> >>
> >> > Noise implies a loss of information, and while the gross appearance
of
> >the
> >> > noise can be hidden, you don't get the information back.
> >>
> >> Noise does not imply a loss of information
> >
> >Of course it does. The presence of noise means that the bits that hold
that
> >noise hold noise, not information.
>
> That's incorrect. They hold information *plus* a noise value. The
> distinction is subtle, but very important.

How so? Noise is both positive and negative in sign. Since you don't know
what the sign was, the correct value of the pixel is somewhere +/- the
average magnitude of the noise from the measured value. Noise is noise. A
complete lack of information.

David J. Littleboy
Tokyo, Japan