What manipulations are adversely affected by 8bit images?

Hi,

I use Linux for pretty much everything, and am in the process of
getting a new camera after all these years. I'll be using RAW fairly
frequently, and want to ensure I take advantage of the increased bit depth.

However, the main software I use, Gimp, currently does all its
operations in 8 bits. Looking around, I often see people saying, "Do the
relevant operations where 16 bits matter (e.g. Levels adjustment) in
another program (e.g. Bibble, Cinepaint), and then use GIMP for the rest".

The question is, "For what digital manipulations does it matter if you
have 8 or 16 bits?" I know of Levels, and Curves. What else? How would I
know - is there a rule of thumb to follow?

Thanks.

--
Cartoon Law: Any violent rearrangement of feline matter is impermanent.


/\ /\ /\ /
/ \/ \ u e e n / \/ a w a z anl

 

Photoshop is really missing the boat not having a Linux version.
To my untrained but fairly critical eye, I can not see much difference
between 8 and 16 bits, using a Canon i960 printer. I don't use 16 bits.
It also slows things down a lot with my aging computer.

 

Photoshop is really missing the boat not having a Linux version.

I think generally, few can tell the difference. That's not the reason
people use 16 bits. When does some level adjustments - say to stretch
the histogram to get better contrast, a lot of interpolation goes on,
and the results aren't as accurate in 8 bit mode. Supposedly, if done in
16 bits, you get better (and probably more noticeable) results.

--
Cartoon Law: Any violent rearrangement of feline matter is impermanent.


/\ /\ /\ /
/ \/ \ u e e n / \/ a w a z anl

 

I know - the problem is that other than levels and curves, I don't know
what else creates such gaps (I suppose I could just "do" it and compare
before and after histograms...). Would sharpening do that, for example?

--
Laugh alone and the world thinks you're an idiot.


/\ /\ /\ /
/ \/ \ u e e n / \/ a w a z anl

 

Low light shots where you want to 'push process' to compensate for
several stops of under exposure end up less posterized if you use 16
bits instead of 8 bits. My rule of thumb would be that any software
manipulation that makes a large number of gaps in the histogram should
be done using 16 bits.

jpc

 

SNIP

With 16-bits/channel, you can avoid cumulating rounding errors at each
subsequent correction/adjustment you make. Then finally when it comes
to outputting the image, you'll have less chance on posterization
artifacts showing up.

 

Indeed - for this reason I requested that gimp have
an option to

1) save a level/curve/colour adjustment to disc, as floating point
2) apply a sequence of saved adjustement in one go.

The curves and maps could be combined ("composed", technically)
very efficiently in floating point, producing one "net" mapping,
which is then applied.

This operation is quick, and only introduces one stage of rounding,
thus avoiding the cumulative rounding problem you mention.

And withou the CPU costs of doing everything 16 bit.

Gimp have not implemented my suggestion :-(

BugBear

 

I doubt 16 bit sharpening would be that much different than 8 bit
sharpening.

jpx

 

One thing that doesn't seem to have been mentioned yet - RAW images may
contain more bits per channel (eg if it's a Canon it would probably be
12 bits). If you work in 8 bits then you're throwing away a fair amount
of data right at the very start.

However, if you can do the majority of your adjustments (contrast etc)
in your RAW converter then this will not be so much of an issue, as the
adjustments are performed before sampling down to 8 bits.

 

In general, the more 8-bit operations you do to your starting image,
the more damage you'll get, because it's due to clipping & rounding
errors at each operation, which accumulate with each operation. It
shows up in the form of posterisation in one or more colour channels,
most visibly in highlights & shadow areas.

I run Linux myself for many things, but unfortunately, Gimp just isn't
capable (yet!) of doing high quality photo work.

 

Yes, absolutely. Really, almost anything you're likely to want to do
to a photo (other than clipping) will introduce errors if done in 8
bits. This is why professional image or cine editing systems all
operate in 16 or 32 bits per channel, per pixel.

 

So what do you use?

--
Mary had a little lamb, a little beef, and a little ham.


/\ /\ /\ /
/ \/ \ u e e n / \/ a w a z anl

 

Photoshop, or one of several RAW image processing packages.