More on the Subject of 8-bit Versus 16-bit
This week, I wanted to expand on what I wrote a week ago regarding the benefits of 16-bit as compared to 8-bit editing. Here then are the answers to a few questions that have come up since last week.
Can I still save space by keeping files in 8-bit, but converting them to 16-bit whenever I open them for editing?
Yes, but this is not recommended. Assume you start with a 16-bit image filled with lots of good information with up to 65 thousand values per channel. Once you convert to 8-bit, you're down to only 256 possible values per channel, and you will lose all the extra shades that 16-bit had allowed you. Every shade will get collapsed down to the nearest 8-bit value and fractional data will be truncated. Even converting back to 16-bit later won't give you those shades back again. Your file will still only have 256 discreet values in it. Adjust the Levels as we did last week and you'll end up with the same gaps that you would have had you left your file in 8-bit mode. The data simply doesn't exist to avoid the gaps. Photoshop has no way of knowing what fractional values might have originally been there since they were lost when the data was converted to 8-bit.
If you are truly done editing a file, it would be safe to convert to 8-bit since you can't use more than that for printing, but how would you know you won't want to tweak something a bit six months or a year down the road?
If my Levels histogram has gaps in it already, what can I do to fix it?
Not much, but I do have one trick for you that may help if you find yourself in this predicament. Detail in an image is defined by edges in the image. Areas lacking edges are by definition lacking in detail. Skies and other smooth expanses of similar color with no edges are where banding and other artifacts resulting from 8-bit editing can be most apparent. All the detail in other areas helps to obscure such problems.
After we convert back to 16-bit, we can easily add some Gaussian blur to such open expanses with little loss of detail since little was there to begin with. Doing so has the benefit though of "mushing" our data sufficiently to make it no longer fall just on the 8-bit values. To a degree, the gaps get filled in with random data from the blur, and the data previously bunched up on the 8-bit values gets spread back out. Surprisingly, even a single pixel of blur is sufficient to do the trick on even the largest of images.
Of course you can't really blur parts of your image with detail in them, but then these parts won't show most problems resulting from 8-bit data loss anyway.
It may be a bit tedious to selectively blur only smooth, featureless parts of your image, but there's no other good remedy I know of. This would need done right after converting to 16-bit mode too, before any tonal edits on the image or the problem will just get worse.
Does all this mean my image will never get damaged if I always keep it in 16-bit mode?
Unfortunately, no, but it does greatly lower the possibility of damage that will show. There really is only so much you can do to an image, and Photoshop should not be viewed as a substitute for getting it right in the field. Push your pixels around too much and you will eventually have problems no matter what.
16-bit editing allows you to do reasonable edits to optimize an image, but it won't let you turn a bad image into a good one.
If 16 bits per channel is good, is more even better? Is this even possible?
Maybe some day, but not till computers get much faster. Editing a 16-bit image requires a heck of a lot more number crunching than does 8-bit editing. This is the primary reason why 16-bit support has only relatively recently become practical on home computers.
Also, with each extra bit per pixel you store, you will reap progressively diminishing returns. A one-bit image would allow only two possible values per channel, or eight colors total. This might make for usable clip-art, but that's about it. The jump to 8-bits per channel therefore is huge as it gives us sufficient precision to record the subtle tonal transitions that make an image look photographic. Going to 16-bits per channel doesn't change what an image looks like, but it does give us the extra precision needed to effectively edit an image without visible data loss (within reason, of course). Going beyond 16-bits per channel may offer moderate incremental gains in image editing, but that would be about it.
For now at least, 16-bit is where it's at.
Update 7/31/2005 - I ishould probably add a note here that the next new thing is indeed on the horizon. The new Photoshop CS2 supports a 32-bit per channel format called HDR ("High Dynamic Range") that is quite intriguing. You can find out more about it here.