Theory and Insignificant Differences
Submitted by mitch on Sun, 07/11/2010 - 17:44.
There are a lot of theories that people argue about in photography forums like DPReview, Luminous Landscape, Photo.net, etc. Some examples you'll read about a lot:
- Always shoot RAW, never shoot JPEG;
- Always edit in 16-bits, not in 8-bits;
- Avoid going to L*a*b because it means a trip through the color engine.
These are just some of the issues that people have very strong feelings about and are willing to argue. Sometimes, the arguments get personal and really nasty.
I always shoot RAW. I understand why some don't. I know from editing thousands of photos that Dan Margulis is wrong as a factual matter when he says photographers can edit in 8-bits and not expect serious problems. I have seen too many photographs that degrade visually when the JPEG version is edited and do not when the 16-bit .PSD or .TIFF is edited.
Dan is right for many photographs. You can edit an 8-bit JPEG, print it, do the same with a 16-bit .PSD or .TIFF, and get equally pleasing results.
Dan Margulis is at the center of another argument. The desirability of working in L*a*b. He argues, you can go to L*a*b, make significanrt edits, and return to RGB without visible artifacts. On the skeptical side, there's Andrew Rodney. He warns against unnecessary L*a*b conversions and what can happen with a round-trip through the Adobe Color Engine.
In my recent video, Color: It's All Relative, Part 7, L*a*b and Back Again, I demonstrate that Andrew Rodney has the better argument with an 8-bit photo and Dan does wth a 16-bit photo.
The really important point, which I'm trying to expand on here in my blog entry, is that a lot of differences in theory turn out to be small differences in the typical case.
Take an 8-bit version of my pseudo Macbeth Color Checker Card, send it to L*a*b. and send it right back to RGB without any editing, and there will be demonstrable damage if you look at the histogram and the numeric information that goes along with it. Would someone notice it, if you print the original and the version with the round-trip to L*a*b. Probably not. Almost certainly not.
Andrew Rodney is 100% right: in theory. Will there be visible damage that anyone would notice or care about? Not very likely from a simple round-trip to L*a*b for sharpening or a boost in saturation. In practice, Dan has the better argument for many photos.
A savvy photographer will have to weigh the benefits that can be achieved from a round-trip to L*a*b against artifacts that might develop, especially if more editing is going to occur back in RGB.
That's the way with many of the arguments in digital photography. People get stuck into taking a position on theory that makes little or no difference for most photographers and most photographs.
Technorati Tags: misc- mitch's blog
- Login or register to post comments
(No subject)
Oh dear - another irrelevant test
Hi Mitch,
Read the blog and watched the video.
I'm afraid the whole thing still failed to prove there were any VISIBLE differences.
Firstly I understand and accept that you main point is:
"that a lot of differences in theory turn out to be small differences in the typical case."
The problem is that then you went right ahead and in the video your 'case' was not even on the same planet as far as being 'typical' - why did you do that?
BTW Readers might also like to see my responses to your previous blog which covers similar territory here:
http://www.thelightsright.com/LabForColorCorrection
(warning dear readers there's quite a bit of detail)
So back to the video and the above blog comments - here goes:
1 - You mention a number of times the effect on photos BUT...
2- your demonstration did not use a subject that bears any resemblance to a photo i.e. it was either a computer generated sRGB color chart (MacBeth type) or a photo of one - if it was the former that is even more naughty as no camera known can produce it's equivalent
3 - you forgot to mention whether you had dither turned on - which makes a difference to numbers and histograms
4 as per my post in the link above - Changing into Lab and back -WITHOUT DOING ANYTHING- is a CON. Ask yourself - why do these tests not involve doing anything that a typical 'good reason' Lab move would require. Answer - because the results would make the number/histogram junkies look foolish.
5 You mention damage a couple of times - why is it damage? - maybe it is an improvement (OK that's not right for your test - but only because the move to Lab was irrelevant) Any REAL WORLD change to an image will always produce 'differences' - a statement of the obvious, but that is my point. In any real change (i.e. edit tweak etc) to an image, even minor ones - the 'differences' are much more 'damaging' than an irrelevant RGB-Lab-RGB trip. But it's not really 'damage' at all is it? Because the person is making those changes for a purpose.
6 You mentioned artifacts which might develop - ? Other than computer generated gradients - which are not photos - what artifacts are these?
Lastly BTW I think this might be a useful discovery for anyone who has read this far.
I tried your test and found an interesting problem with PS. Since one needs to lay an original non Lab'd layer over the top, you need to open another copy. Interestingly I found that using the Place command is quite different to shift dragging a layer from an open copy - even after using the align layers top & left. Try using Place over an un Lab'd copy, change to difference, flatten, invert and then do the extreme levels move - ouch. Yet if you drag a layer if comes out perfect i.e. no difference.
Cheers
Brian - Melbourne Australia
And another thing!
Hi Mitch,
I also must disagree with your conclusion:
"I demonstrate that Andrew Rodney has the better argument with an 8-bit photo and Dan does with a 16-bit photo."
Really?
In your video you explain that you inverted the result of a difference layer but noted you had to 'whack the levels around to make it obvious which pixels were different' When I duplicated your test I found this "whacking' wasn't just a slight touch but a monumental and obscene move - one that would never be used in any real situation.
Excuse me? If you had to do such an obscene move to make it visible then doesn't that promote Dan's view over Andrew's??
You then did the same test using a 16 bit image and you state
"there ARE differences but you sure won't see them"
Huh? isn't that the point? Without monumentally smashing the levels in the 8 bit, you wouldn't have seen them there either Right?
The logic seems all twisted - IMO it goes something like this
Although you can't see any visual change, some change has happened therefore this must be bad - errr sometimes,maybe - although we don't know when it would be bad, and even if it was you probably couldn't see it any way and, ummmmm there would be no way to tell if it was the Lab move because when you actually did a REAL Lab move, i.e. for a proper reason, much more change would have happened then that in an IRRELEVANT Lab move, whether you did the changes via Lab or not.
I can see how this will be useful to many people ;)
The problem I find in this (and similar) debates is that:
- Both sides make 'extreme' claims
- The 'damage has occurred side' rarely, if ever demonstrate their point in a way that is actually VISIBLE to the human eye - i.e. in the actual image, no they always use counting colors, pixels, histograms or perverse changes to exaggerate the 'damage'.
- Nobody explains properly and simply what is actually happening, so a user can evaluate when (if ever) if might matter.
Here are my simple points:
a) Lab is difficult to grasp (for most) without unnecessary 'boogie man' claims scaring people off
b) Anyone making such claims MUST also show/prove when it makes a actual visible difference in the result- not some esoteric and meaningless 'test' that just proves simple arithmetic.
c) Plus they should demonstrate what REAL LIFE type of situations this visible difference will occur - yes yes we know about computer gradients - but other than that?
d) Please just explain to people that, due to simple maths, when you go to Lab and back that essentially infinitesimal changes will occur if you are stupid enough to go there and back for no reason at all.
e) Please also explain that if you go to Lab and back, making a meaningful change i.e. some improvement which works better in Lab, then the resulting changes are much much greater that those in (d)
f) If you made the same changes by staying in RGB and performing the same change (but obviously an inferior one) then once again much greater changes occur then in a silly RGBLabRGB do nothing move.
g) Wherever you see the word change in the above list, feel free to change it to DAMAGE, because some people like to misrepresent them as being the same thing.
For those who are interested, in simple terms the change that occurs in the pointless RG-Lab-RGB trip is quite simple to explain in a basic way. I'll have a shot at it:
Although Lab is a vastly 'wider' color space than RGB, (i.e. it can represent even impossible colors), when the conversion takes place (in 8 bit), occasionally an R or G or B value is changed by 1 digit, not all, just one of them , it's just simple arithmetical rounding issue.
Try this, fill a layer with any solid RGB colour, e.g. R86,G133,B67 then do the silly trip, the new value will be 87/133/67, do the silly trip 10 more times and nothing changes - GET IT? The change only happens once.
In a 16 bit image the values will (seem to) stay at 86/133/67 gosh that's good, NOPE it's just that PS only shows values as 0-255 in whole numbers which can't show the 16 bit number range - i.e. in the 16 bit it was 86.000/133.000/67.000 and now it might be 86.521/133.000/67.000 - note this is not the actual value but demonstrates what occurs.
The bottom line is that whilst there are theoretical situations where a Lab trip might cause problems, other than computer gradients I'm yet to find any photographic examples.
Which suggests that IMO it is not that it "makes little or no difference for MOST photographers/photos" but that it is so close to NEVER making a difference that it doesn't actually matter.
What these debates seem to achieve is that the fear of 'DAMAGE' is used as an excuse not to even try using Lab. Easier for the non Lab user to say 'oh no that's a nasty place', instead of admitting they've never actually been there cause they don't understand the lingo.
-
Brian - Melbourne Australia