Filmscanners mailing list archive (email@example.com)
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[filmscanners] RE: 8bits vs. 16bits/channel: can the eye see the difference
> From: Karasev, Alexander
> I think it is quite an assertion you are making there, Paul, that
> "the level
> of noise in a real-world image, either from film grain or CCD noise, is
> always greater than a least-significant-bit of an 8-bit value." There are
> situations when this is not the case. One perhaps is scanning a very
> fine-grained emulsion at a medium resolution; another, dealing with
> downsampled images (from medium or large format film?), and yet another,
> applying a smoothing algorithm to the sky and/or other areas that do not
> carry details or patterns. These are just a few off the top of my head.
Well, of course it's _possible_ to find a situation in which the noise level
is lower than that, but I've yet to see it. I've scanned lots of Kodachrome
25 slides, which are very fine-grained, with an LS-2000 in 12bpc mode, and
after truncating to 8bpc, I still get noise in blue sky.
As to downsampling, it's certainly true that if you reduce the linear
resolution enough, most of the noise will go away. However, I can't see any
situation in which I'd need to reduce the resolution where I wouldn't also
have to reduce the depth to 8bpc anyway (e.g., web images).
As to smoothing algorithms, if you get posterization from the resulting
noise elimination, then just don't smooth so much.
> Besides, once it is established that the eye *can* see the difference
> between any two RGB24 levels, which this experiment does, the doubt as to
> the advantage of the greater than 8 bits color depth for input
> and output as
> well (not just processing) goes out the window. If your input is noisy, it
> could mask the limitations of 8 bits/channel representation - OR perhaps
> even lesser representations, depending on how bad the image noise is. But
> the better the images, the more obvious will be the proven-to-be-visible
> advantage of the 16 bits / channel representation.
Well, sure, if the image has sufficiently low noise, then 8bpc won't be
enough. My point was that I've yet to see any real images where that's the
case. My digicams produce plenty of noise, even in broad daylight, and my
LS-2000 either produces or manages to find enough noise in the film, even
with K25 slides. I've never, ever seen posterization in a real image. Maybe
a high-end drum scanner might have low enough noise.
I still appreciate having 12bpc data and 16bpc processing, for those
situations where I have to apply strong curves.
Obviously, if 16bpc were free, then none of this would ever be an issue. But
16bpc takes twice as much storage uncompressed, more than twice as much
storage when compressed, and twice as much processing time. Adobe weighed
those disadvantages against the rarity of any real advantages, and that's
why Photoshop doesn't do everything in 16bpc.
Ciao, Paul D. DeRocco
Unsubscribe by mail to firstname.lastname@example.org, with 'unsubscribe
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or