Photoshop knows two data modes:
• 8 bits per channel
• 16 bits per channel
WHY USE 16 BITS PER CHANNEL?
Many digital SLRs allow 12 bits per channel capture and this is
later represented as 16 bits per channel in Photoshop (mostly 16 bit
TIFF). If the images coming from your camera were perfect, these 4 bits
would not make much of a difference as most printers only use a
maximum of 8 bits. However, an image that does not require some
minor correction is very rare. Once the saturation or contrast of a
color image is corrected, the actual color information is soon
reduced (due to rounding errors) by one or more bits. If you started with eight bits, you might end up having only
six bits of real color information per channel left. The not so
smooth color gradients indicate this loss. Starting with 12 bits
ensures there is still a sufficient amount of color information
available after even major corrections.Note: Stay as much as
possible in 16 bits as you can afford (memory, disk space and
performance). The test.Here is a test you can perform to experience
the 16 bit vs. 8 bit difference. Open a photo in Photoshop, which is in 16 bits. Now perform some
curves or levels. When finished changing the levels have a look at
the histogram in the “Levels” dialog box. Open a photo in Photoshop
but convert it to 8 bit just after you open the file and then do the
same curves and levels operations. In the above figure you see is a
histogram with many gaps. This indicates lost data and shows up in
an inferior gradient smoothness, or posterization. Note: On the
other side these gaps should not make you nervous. If the photo and
print look fine, that’s all that counts. Conclusion: Remain as long as possible in the 16 bit mode.
Why would you ever use 8 bits per channel if 16 bits is so much
better? Using 16 bits:
• Takes longer
• Needs more memory
• Creates larger files
• If you work with versions of Photoshop prior to CS many
operations (e.g. layers) are not available for 16 bits/channel.
Before Photoshop CS was released, all our work was done in 8 bit
layers and we also got good results. But if you have enough disk space, memory
and fast processors, 16 bit is the best way to go.