ðòïåëôù 


  áòèé÷ 


Apache-Talk @lexa.ru 

Inet-Admins @info.east.ru 

Filmscanners @halftone.co.uk 

Security-alerts @yandex-team.ru 

nginx-ru @sysoev.ru 

  óôáôøé 


  ðåòóïîáìøîïå 


  ðòïçòáííù 



ðéûéôå
ðéóøíá












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

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

[filmscanners] RE: 8 bit versus 16



Andreas,

> Austin Franklin wrote:
> >>Its well documented in the 3D community that having
> >>24 bit colour internally in 3D processing engines
> >>can result in banding in certain scenes, and thats
> >>why Nvidia and ATI have developed 32 bit engines,
> >>and more.
> >
> > That's an entirely different issue.
>
> I don't think so. This is exactly the same problem.

I disagree, but that really doesn't matter to this discussion.

> When editing an image colorwise, then depending on the algoriths used,
> an 8bit value per channel can easily lead to banding on some operations.

It depends on what the source of the banding is.  Many sources can cause
banding, but we are specifically talking about posterization here that is
caused by tonal manipulation of the data, where the tonal transitions are
not smooth, and this is caused by missing intermediate values.  Banding
*CAN* be something completely different than this, and may not have a thing
to do with this specific issue.

> It will show up pretty clear in histograms, but might not be visible to
> the viewer, depending on where it happens and the visual sensibility of
> the viewer (and the monitor or whatever the outout device is).

What's important, is whether it's visible or not.

> If there is more room to work in, this banding does not happen or is
> less visible. That is the advantage of working with 16bits.
> Just plain old math. If you work in a small integer space some
> operations will produce losses.

Some operations *CAN* produce losses, and it depends on the visual
significance of those losses.  For 8 bit color images, as I've stated, those
losses are not visible for MOST images.  That's just plain fact.  If you
*WANT* to use 16 bits (which, as I've said, isn't really 16 bits
anyway...the N bits is expanded to occupy a 16 bit space, but fact is, there
are LOTS of holes in the 16 bit data.  On the histogram, you ONLY see the
upper 8 bits, so the holes aren't apparent).

> The bigger the space the less
> loss you have.

True, but again, there may be NO visual impact caused by the loss, as has
been readily and frequently proven by many an experiment.

> If the effects of working in the wider space affect your images visibly
> is something that only can be judged by looking at them and comparing
> the results of working in both ranges. For some it does so drastically
> for others not.

Exactly, but to claim that you "need" to use 16 bit data (for color) is
simply wrong, and was my point, and why I was very careful in what I said.
People can tout this, and espouse theory, all they want, but reality shows
otherwise.  If you want to argue this, it's important to understand what the
impact of theory has on reality.  As I've stated clearly, 16 bit data *MAY*
be beneficial for *SOME* images, but not for all.  For some people, it may
be more significant than others, depending on what it is they photograph.

What would be "nice" is if someone would post two snips of the same image,
showing this problem.  If it was such an issue, you'd think there would be
tons of web pages with this on it...but alas, I haven't seen any, and you'd
believe people would be ready to share their images...but I haven't seen
that either...  I do find that interesting.

Regards,

Austin

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



 




Copyright © Lexa Software, 1996-2009.