Digital BW, The Print

Julio Fernandez writes:

> Indeed the number of colours possible under 8 bits is mega smaller than in
> 16 bits and somehow the differences have to show.  The question is when?

Consider this:  In an 8-bit image, if you isolate level 149 in your
histogram and try to expand it to fill the entire image, you'll get a
single color.  If you do the same thing in a 16-bit image, you'll get a
completely new image, because that one channel in the histogram contains
256 levels itself in a 16-bit image.

The main reason this isn't always obvious in practice is that very few
images make use of all sixteen bits.  For example, digicam images or
scans may only have 12 or 14 bits at most.  As a result, they tend to
posterize almost as badly in 16-bit manipulation as they do in 8-bit
manipulation.

However, if you really do have a full 16-bit image (one that contains
data in all 16-bits, like an extremely high-quality scan or digital
capture), you'll see the difference if you have to make sharp changes in
curves or levels.  It also reduces cumulative errors if you make many
consecutive changes to colors or levels.

> For one thing does the color space used not have something do with whether
> you see the difference or not?

It has a very slight effect.

> I wonder then if using small color spaces like sRGB or Color Match
> will limit the number of colours and put many of those colors which
> actually reside in film and in 16 bit RGB files out of gamut anyway
> so that the small color space becomes the limiting factor?

The biggest limiting factor is the original capture.  If the original
image capture doesn't actually contain a full 16 bits of data, you won't
see much difference.

> By the same token, inkjet printers have a typically smaller color gamut than
> true photographic printers like a Lightjet.  As I understand it, inkjets
> like the Epson 2200 can only print to 8 bit colour so in fact, 16 bit files
> are printed to 8 bit,  is that true?  If it were, the differences between 16
> and 8 bit files printed in ink jets may not be readily evident, that has
> been my experience.

The weakest links in imaging workflow today are at the front and back,
because these are still (and must remain) analog processes.  Thus, image
capture and image display/printing are the weakest links.  Display and
printing are worse than capture.  No commonly available display or
printing technology allows you to clearly distinguish more than 8 bits
of tonality in the result.  No commonly available image capture system
allows you to capture more than about 12-14 bits in the original image
(with electronic cameras, the limitation is the image sensor; with film,
the limitation is the scanner).

> Are there any other advantages in large bit files besides non-banding?

Large files are an advantage only for image manipulation.  If you go
straight from capture to printing, you don't need more than 8 bits.  If
you do a lot of retouching and adjustment to your images, having 16 bits
dramatically reduces the chances of posterization and cumulative errors
that might produce image artifacts.

Thus, while it's best to do your image editing and retouching in 16-bit
if possible, you only need eight bits for printing or display, and if
you are sure you'll never need to manipulate an image again, eight bits
is plenty for archiving, too.