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: Sun, 17 Apr 2005 12:39:58 EDT
Bruce Fraser writes:
>>The only difference between the two is that on one, I edited the high-bit
data in Camera Raw, and continued to edit the high-bit data in Photoshop, where
in the other, I didn't do any editing on the high-bit data. >>
What you have devised is not a test of 16-bit vs. 8-bit. It is a test only of
Bruce Fraser's correction ability in Camera Raw vs. Bruce Fraser's ability in
Photoshop. The latter contestant did not acquit himself well.
Your test consists of correcting the 16-bit in Camera Raw (Fig. 2.3A) vs.
trying to duplicate the correction in 8-bit Photoshop (Fig. 2.3B). Having
declared that the two resulting documents are equivalent ("The two images shown above
appear quite similar") you apply a new set of identical curves to both,
resulting in Figs. 2.4A and B, and you say that 2.4A is the winner, with which
judgment I concur.
Now, even if someone who knows how to color-correct follows this workflow,
the whole exercise is invalid, because the Camera Raw correction can never be dup
licated precisely in Photoshop. But at least it would not be a total joke.
Granted that we have both gotten older and our vision is not what it used to be,
and that the print size is too small for the purpose, surely you must agree
that Figure 2.3B is much darker than Figure 2.3A.
This entire graphic should be saved as an example of why the use of
histograms should be restricted to those who know how to read them. You show these two
sets of histograms, and remark that the ones for Fig. 2.3B is rougher than
those of Fig. 2.3A. Somehow, you have missed the vastly more important
comparison, which is that every portion of the Fig. 2.3B histograms is pushed well to
the left (i.e., darker) than the corresponding part of Fig. 2.3A. The very first
thing that a beginning color-corrector learns is to set highlight. You
haven't done it in Figure 2.3B. It's considerably lighter in Figure 2.3A, especially
in the critical green channel, automatically making it the better image by a
wide margin.
According to your own graphic, the mean level value in Figure 2.3A is 150.61
and the median is 166. For Fig. 2.3B the corresponding values are the much
darker 139.54 and 148. These are misleadingly optimistic figures; the ones in
the more critical green or luminosity measurements would make you look even
sillier, if possible.
Now, there are a lot of people on this list whose business it is to measure
variations between two products. You may wish to ask them what Delta-E they
compute for these two images, or, in simpler terms, whether they consider that an
image can reasonably be said to be "quite similar" to another that is
typically 12 to 18 levels lighter in each channel.
After applying the second set of curves, you state that "[Fig. 2.4B] shows
much less detail, and displays some unwanted hue shifts." Agreed. Fig. 2.4B is
flat. You caused this yourself by failing to set an appropriate highlight when
you corrected Fig. 2.3B. The colors in Fig. 2.4B are duller. You planted the
seeds of this yourself in Fig. 2.3B by having a darker quartertone,
emphasizing the tones that contaminate the bright colors of the image. Then you
harvested them by applying an RGB master curve rather than channel by channel ones.
Maybe the second or third thing that a beginner learns about color correction
is not to use master curves, because they create color shifts of this type.
Anybody who doesn't know how to set a proper highlight, or who is using the
RGB master curve, has many better things to worry about than bit depth.
>>No, that's just a silly game. A much better test would be to get an
optimally-edited high-bit image from a much better photographer than
I am, then take an 8-bit version of the original uncorrected image
and try to edit it to produce a result that she finds equally
satisfying.>>
"Equally satisfying" would be a disappointment. I would hope to do much, much
better than she did. What you propose is once again solely a test of color
correction skills. If I'm better at it than she is, I'll win. If she's better
than I am, she'll win. Bit depth has zero to do with it.
>>The argument has never been that the same edits applied to 8-bit and
16-bit will produce better results on the 16-bit version (though they
sometimes may).>>
What on earth COULD the argument be about, if it isn't that? Anyone who is
saying that there's no known advantage to editing in 16-bit can ONLY be saying
that if you just converted your 16-bit files to 8-bit and went about your
business as usual, you would get substantially the same results as if you had
stayed in 16-bit longer and done *exactly the same things*.
>>The argument is that if you know what you're looking
for, you'll get it more easily out of the 16-bit than the 8-bit
version, typically with fewer edits,>>
Circular reasoning. People are telling you that the 8-bit data is
qualitatively equivalent to the 16-bit. That means that if you do the same things to one
as to the other, you should get qualitatively indistinguishable results. "More
easily" and "fewer edits" are inapplicable. The only way you can tell whether
16-bit gained you anything is to redo the edits in 8-bit exactly as they were
done in 16-bit, and compare the two final results. Same number of edits, same
ease of operation, probably faster in 8-bit because the computer doesn't have
to digest twice as much data.
As we discussed at Photoshop World, there are enough people around
misrepresenting both of our views that there is no great need for either of us to
contribute to it. So, I ask if the following is a fair summary of your position:
1) Bruce states that the argument has never been that identical edits applied
to 8-bit and 16-bit files would produce better results in the 16-bit version,
but he argues that it is possible that they might.
2) Bruce has not offered up any images that would demonstrate such a
superiority for 16-bit correction (as opposed to identical edits applied to 8-bit),
but he suggests that such images might exist.
3) Bruce's comments on "night and day difference" and "totally obvious to
anyone who looks" are based on his experience and perceptions; however, he has
never personally tested a series of corrections done to a 16-bit file on a live
image versus identical corrections done to an 8-bit one.
Dan Margulis
P.S. >>However, since I like to check my sanity every now and again, last
night I went through an exercise where I took an unadjusted scan and
edited it in high-bit. Then I applied the exact same edits to an
8-bit version. The 8-bit version shows large hue shifts that are
absent in the high-bit version. I'm happy to send you JPEGs of the
final results, just let me know where.>>
Based on this description I surmise that you have done what others have,
namely constructed a series of huge "corrections" that reverse one another. Or,
possibly you have generated an 8-bit file from some other source (e.g. a camera
output module). I've seen a lot of this kind of stuff, so I would not be
interested if either is the case.
If neither of these items is true, however, then of course I'll look at them,
because they would contradict about eighty other images that have been tested
and not shown any such shift. In addition to the two JPEGs you offer, the
items I'd need would be:
1) You confirm that the 8-bit version was generated in Photoshop by mode
change directly from the 16-bit and both sets of corrections were applied from
there;
2) You confirm that you believe that the corrections are real-world in
that they represent an honest attempt to improve, not degrade, the image, even if
you believe that it is a clumsy and amateurish attempt.
3) JPEG of the *original* 8-bit file;
4) Record of the exact corrections applied.
Can send right here.
_______________________________________________
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