Digital BW, The Print

> From: steve_bye [mailto:steve_bye@...]
>
> I have learned a great deal from Dan's Professional Photoshop
> book and many
> articles, but I've learned that I need to filter some of the
> things he says
> about color management and 16 bit processing. The Bruce Lindbloom
> link below
> describes a very significant flaw in Dan's 16 bit challenge. From Bruce's
> home page click the Info tab, then scroll down to the "Dan Margulis 16-bit
> Challenge - What's Behind the Controversy" article.
>
> http://www.brucelindbloom.com/

I had a run-in with Dan about this in the Color Theory list, too.

Although I agree that most of the time, 8-bit processing is fine, there are
times where 16-bit produces visibly better results. You'll only see the
difference in a featureless gradient with very little noise, but in real
life such a subject is commonplace: blue sky.

To prove the point, I begin with a raw image from my Canon 10D, which is a
pretty quiet camera. I choose a well-exposed ISO 100 image of an outdoor
scene in sunlight, with a clear blue sky. To live up to Dan' stringent
requirements, I make a copy of the image, convert both to 8-bit, then
convert one back to 16-bit.

I then record a simple Photoshop action, using two commonly used tools.
First, I go into Curves, and introduce a very slight darkening to the image,
by setting the midpoint to (128,123). Second, I go into Hue/Saturation, and
tweak the Hue a tiny amount to +3. Third, I go back into Curves, and lighten
the image a tiny bit by setting the midpoint to (128,133). Finally, I go
back into Hue/Saturation, and restore the original color with a Hue value
of -3.

At this point, I stop recording the action, and replay it on the other
image. This produces no significant difference yet. However, if I replay it
a few more times on both images, and examine the sky, I see noticeable
posterization in the gradient in the 8-bit image.

Of course, this action involves two pairs of operations that more or less
undo each other, and then involves the repetition of these exact operations
a few more times. But the point is that it doesn't matter what the
operations are: any slight tweaks to the image in 8-bit mode will introduce
quantization errors, and it is these that can build up and become visible. I
only chose those particular operations to make them easy for anyone to
repeat the test, and to guarantee that the resulting image doesn't look like
it has been modified in some unreasonable or atypical manner. Anyone who
wishes to repeat this test can feel free to use any long sequence of minor
edits, using Curves, Levels, Hue/Saturation or Color Balance. I expect the
results will be the same.

I can provide the image I performed this test on, but the steps should work
on any image that has the characteristics I outlined: low noise, and a
properly exposed blue sky.

There are other ways, besides resorting to 16-bit calculations, to avoid
this posterization. For instance, one can use adjustment layers, so that the
back-and-forth tweaks don't pile up quantization error. This is a good
argument for adjustment layers, especially in earlier versions of PS where
full 16-bit support doesn't exist, but represents a restriction on workflow.
If you prefer just to tweak images back and forth until you like the way
they look, in 8-bit mode you have to worry that eventually you'll see a
visible increase in quantization noise. In 16-bit mode, you don't have to
worry about it at all. Since 16-bit support is present in Photoshop CS, you
might as well relax and use it.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@...