Re: 16 bits = 15 bits in Photoshop?
Re: 16 bits = 15 bits in Photoshop?
- Subject: Re: 16 bits = 15 bits in Photoshop?
- From: email@hidden
- Date: Tue, 19 Apr 2005 14:31:16 EDT
Roger writes,
>>I wish you would have kept the discussion on the grounds of bit depth
strictly, Dan. We can debate at lengths the effects of more or less
resolution in general, without making sweeping generalizations, that is. But
I'd like to see you elaborate on 8 bit vs 16 bit, >>
First, a general comment. I have to get ready for a European trip so I intend
these three to be my last posts.
I am sorry if I did not make the analogy sufficiently strongly. The effect of
extra bit depth is ALMOST EXACTLY analogous to the effect of extra
resolution. I believe that a person who does not understand the impact of more or less
resolution is very unlikely to understand what goes on with more or less bits.
I talk about resolution because its effects can be obvious. With bit depth the
same effects are visible, but on a much smaller scale, so people don't
realize what they're looking at.
>>My contention is that a 16 bit capture will conserve MORE of that original
sky
subtle gradation than an 8 bit capture. To me, that makes intuitive sense
but, I think you want hard facts?>>
The result is correct but the theoretical analysis getting you to it is not.
It is correct that the 16-bit file, if the differences are stressed to an
infinite degree, will give a more pleasing visual result in a section of sky than
an 8-bit file.
The more data you have to sample from, the less the variation will be from
the mean. This is true with respect to bit depth, resolution, or even tossing a
coin. If you toss a coin four times, the most likely result is two heads and
two tails, but three heads and one tails is entirely possible. Toss the coin
eight times, and six heads and two tails is substantially less likely. Toss the
coin forty times, and thirty heads and ten tails is exceedingly unlikely. Toss
it four hundred times, and three hundred heads simply cannot happen.
In imaging it works the same way. The more data, the closer to the average,
whether bit depth or resolution. Imagine that the picture consists not just of
the sky, but also of a large area of grass. Imagine also that you have two
separate original captures, one at half the resolution of the other.
In the sky, almost all the pixels will be close to a certain shade of blue,
but there may be some rogue pixels that vary from it considerably. In the
grass, same way with shades of green.
The individual pixels are not imaged on output. Instead, they are averaged
together to calculate a halftone dot or some other output function. In the
high-resolution version four times as many pixels will be used to make the average.
That will sharply reduce the impact of any rogue pixels. Therefore, the green
and blue areas will appear considerably smoother in the high resolution
version. The lower resolution one is more variable.
Unfortunately, you can't always say which one of these effects is preferable.
In this particular picture, we don't want noise in the sky, so for the sky, a
higher resolution is better. But in the grass, variability is what we need,
not something that looks like a green carpet. Nobody can resolve individual
blades of grass in that kind of image. But a green that's full of action
persuades the mind that the blades are there, and therefore we prefer the
lower-resolution version--up to a point.
Action and variability is good. Jaggedness and harshness, which is what we
get when we go to far in that direction, is bad. Smoothness is good. Softness
and lack of focus, which is what we get when we go too far in the other
direction, is usually bad, although it may be good in a sky.
With added bit depth, exactly the same effect is present. There's extra data,
which permits a more delicate form of averaging. In an area of very smooth
color like the sky, a pixel in a 16-bit file could vary from its neighbors by .1
pixels, whereas in 8-bit the same pixel under certain conditions could be 1.0
pixels away from its neighbors. Therefore, once again, the 16-bit sample is
smoother, less variable, than the 8-bit sample, although you'd have to go in at
500% magnification to see the effect in individual channels if you haven't
edited the file.
The extra smoothness in the 16-bit file is the beginning of a superiority in
the sky, provided that there's massive editing. We also need to admit,
however, that there is the beginning of a potential superiority in the grass in the
8-bit file.
A fundamental error that at least some of the 16-bit advocates make is to
assume that even if the extra data is not helpful, it can't hurt. I pointed out
in another post that I did some experimentation with B/W files, in which the
impact of heavy edits is a lot more visible than it is with color files. Viewers
were presented with alternate versions, B/Ws that were heavily edited in
16-bit as opposed to 8-bit. In most cases, as you'd expect, there was no
preference for either, but in some cases there *was* a clear preference. Sometimes the
viewers preferred the 16-bit version--but they were *twice as likely* to
prefer the 8-bit version. I'll bet that you can tell me the kinds of images in
which they preferred the 16-bit and the kinds in which they preferred the 8-bit.
Trying to engineer that level of preference into a color file is much harder.
If anyone ever succeeds in achieving it, it will IMHO be with sky, probably
around a quartertone in darkness and distinctly cyanish rather than blue. In
that case, it's at least theoretically possible that someone could trash the
8-bit file so completely that banding or noise or jaggedness or whatever you want
to call it might appear in those areas of the sky.
In practice, 1-level differences in pixels don't make much of a difference
except in computer-generated art. Getting them up to be 5 and 6 pixel
differences is a chore. (In high-res v. low-res captures, 5 or even 10-level differences
aren't all that common.) Also, remember that there will be significant
averaging on output which will lessen the irregularity.
Plus, the biggest item: suppose that the picture is not just of a sky, but of
a city skyline in front of the sky. And, suppose that this is one of those
hypothetical pictures where you edit it so much that you bring out the
variability in the sky in the 8-bit version to an objectionable point. Then, the client
may agree with you that the sky has been damaged--but turn right around and
tell you that the 8-bit picture is the one he wants to use because the 16-bit
version is too soft everywhere else.
This explanation may or may not appeal, but if not, you need to come up with
some other theory. Remember, the theoreticians were once highly absolutist.
The quote in my book in which I merged the comments of several "experts" in a
2001 thread on another list reads: "16 bit capability is critical during all
aspects of tone compression....The difference CAN be seen in the final output
very easily. Most definitely on the printed page, especially when using
high-quality halftoning and even more so to a film recorder....It's very easy to see
that substantial color & tone editing will eventually result in data loss and
banding....If it means the difference between taking a 16-bit image capture and
editing that to the final image and taking that same image in only 8-bit and
editing that to the final image then there is a difference like between the day
and the night...Yes, if a histogram full of holes has no impact on final
output, then throw away the graphs and just get on with the print run. However,
all of us have Real World Output showing the superiority of superior data
acquisition....My advice? Take the information you’ve read here to the bank. Stop
doubting and start applying what you've learned here....If you really start out
with a RAW image in high-bit form and a raw image downsampled to 8 bits, the
difference really is night and day....It's totally obvious to anyone who looks
that it's very advantageous to do the big moves on high-bit data."
If our long thread here as proved anything, it's that the above statements
are absurd overkill. The only question that remains is whether there are *any*
color photographs at all in which it's advantageous to do the moves in 16-bit.
I admit (and I suspect, actually) they may exist, but I haven't found any
myself and have given up trying. I hope someone else may find them and we can
proceed from there. But it is a useful exercise to try to figure out how so many
respected figures could possibly have gotten this so wrong. Hopefully the above
discussion may help explain why.
Meanwhile, as a peace offering to those who may have been offended, I have a
tip that pertains to the above discussion and would appeal primarily to
professional photographers.
Some of today's professional-level digicams are so powerful that they capture
much too much data for certain purposes. As discussed above, too much data
can actually harm image reproduction. So, next time you have to shoot a still
life or something else with a decided texture, and you know you have resolution
to spare, you might try a little experiment.
The natural tendency is to focus in on the interest object, taking in only as
much background as is needed. Do that, but also try a couple of exposures
where you deliberately include a lot of extraneous background. This reduces the
effective resolution of the shot because you will need to crop all the extra
stuff out.
Having cropped, go to the closeup and rez it down to match the size of the
second version. Check the two of them out against one another, viewing them at
different magnification. You should find that the higher-resolution one has
more detail when you look at it closely, but from a distance the impression is
that the lower-res one is more active, which is what one usually wants in that
kind of image.
Dan Margulis
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden