SL said:
Can anyone explain the benefit of scanning in 16-bit vs. 8-bit? I have
tried both and can't seem to find any difference (other than file size)!
Theory:
You need 16-bits (which is about 280 trillion possible colours!) because
once you start adjusting levels and curves you will reduce the number of
actual colours used. If you only use 8-bit colour (16 million colours)
then it is possible, with very severe adjustments of level and curves,
to see the reduced number of actual colours as posterisation in the
image.
Practice:
Your eye can only discern around 2million colours, so you can be quite
severe on the levels and curves adjustment before this becomes a
problem. Even so, some adjustments (eg. gamma) will apply most of their
effect to a limited range of colours, often in the shadows so even
though you still have many more colours available than your eye can
discern, you should still be able to see posterisation in certain tonal
ranges. However, noise and image grain tend to obscure the effect
significantly, so in most (if not all) cases the posterisation is
imperceptible in the final image even if it is clearly visible on the
histogram. It is certainly possible to concoct a workflow where this
can become a problem (eg. smoothing the image to reduce noise then
downsampling to get a very clean noise and grain free image, then
applying curves and level adjustments, but its also possible to spend
all your life walking backwards - and equally stupid! ;-)
Bottom line:
If you want to avoid any limitations on your workflow then scan and save
all images at 16bpc, but if you are prepared to make appropriate
pre-scan adjustments (which can often be implemented automatically) to
get an image of almost the correct tonal range required and think about
the consequences of your subsequent workflow then 8-bit scanning and
saving is more than adequate.
