Ad
Ad
Ad
Pages: « 1 [2] 3 »   Bottom of Page
Print
Author Topic: What is ACR calibration REALLY doing?  (Read 61362 times)
Barry Pearson
Newbie
*
Offline Offline

Posts: 11


WWW
« Reply #20 on: May 27, 2007, 04:28:46 PM »
ReplyReply

Quote
All of us abhor the undocumented file formats. But forcing an incomplete color model on the rest of the industry sounds a bit like burning the village to save it.
 
I have written an essay entitled Raw: The Adobe Way, The Microsoft Way and the Open Way which may prove entertaining to some here.[{POST_SNAPBACK}][/a]

Your article "Raw: The Adobe way, the Microsoft way and the Open way" advocates using dcraw:

"I suggest the Open Source community and maybe even the OpenRaw guys write or commission dcraw plugins for Microsoft Windows and Adobe's software, so that dcraw becomes an additional input module to all existing and future software, thereby future-proofing it. In effect, dcraw would become the ACR of the open source community".

dcraw is based on the Adobe colour model!

Dave Coffin: "Adobe Digital Negative (DNG) is a great format -- I totally redesigned dcraw for maximum DNG compatibility".
[a href=\"http://www.dpreview.com/news/0504/05042701davecoffininterview.asp]http://www.dpreview.com/news/0504/05042701...ininterview.asp[/url]

Dave Coffin: "Not only is Adobe DNG now supported, the entire codepath has been redesigned for it. Adobe's XYZ->CAM matrices allow color science to replace black magic, whether decoding DNG or the original raw files".
http://forums.dpreview.com/forums/read.asp...essage=20585662

dcraw appears to build-in inverses of the Adobe colour matrices as found in DNG files. Look for "void CLASS adobe_coeff".

ps: OpenRAW (which you mention) appears to be in terminal decline. They made themselves irrelevant by their hostility towards DNG.
Logged
Barry Pearson
Newbie
*
Offline Offline

Posts: 11


WWW
« Reply #21 on: May 27, 2007, 04:41:16 PM »
ReplyReply

Quote
I'm not aware of what's going on at the ISO with regards to a TIFF-EP update-other than there's something going on-finally. Which is good news I suppose.
[{POST_SNAPBACK}][/a]

You and me both!

ISO produced a draft for TIFF/EP in 1998. It was published as a standard (ISO 12234-2) in 2001. (Which costs a significant amount of money). It began its 5-year revision in 2006, and that is still ongoing. Perhaps the revision will be published within a year or two.

Anyone who believes that ISO rather than Adobe is the place to develop a format such as DNG should contemplate those dates.
[a href=\"http://www.barrypearson.co.uk/articles/dng/history.htm]http://www.barrypearson.co.uk/articles/dng/history.htm[/url]

Peter Krogh said: "As to the standard being owned by Adobe, I checked into that as well. The response I got was that the standards bodies don't want it until it's done. If you know someone else out there who is willing to fund and manage the effort, I'm sure the folks at Adobe would be interested in a discussion. At this point nobody is interested in doing that work".
Logged
eronald
Sr. Member
****
Offline Offline

Posts: 3980



WWW
« Reply #22 on: May 27, 2007, 04:41:59 PM »
ReplyReply

Quote
ps: OpenRAW (which you mention) appears to be in terminal decline. They made themselves irrelevant by their hostility towards DNG.
[a href=\"index.php?act=findpost&pid=119888\"][{POST_SNAPBACK}][/a]

Can't keep the troops from fragging their officers these days. Terrible world we live in

Edmund
Logged

Edmund Ronald, Ph.D. 
Neal.Stout
Newbie
*
Offline Offline

Posts: 1


WWW
« Reply #23 on: May 29, 2007, 04:12:52 PM »
ReplyReply

All this discussion about what ACR is doing, is technical and interesting,  but what about the visual psych?   I fight a constant battle to leave the color temp warm enough.  It has taken me a year to realize the Canon's "as shot" generally does a better (warmer) job than I do.  There is something appealing that happens to skin when when you take the "as shot" of 4900 and move it down to 4600.  And even though it is often very appealing, I don't think the lower setting is as correct as it often seems.  But why does it look good?  It's a trap.  It's quicksand.  It recenters your color judgement way off track.  It leads to that prutrid green that was discussed in this thread.  The mind is incredibly difficult to keep objective in this adjustment.

Maybe we are pushed to this because we are compensating for a camera that has not been calibrated.  The way Adobe processing things?  Hugh?  Couldn't that be verified by carefully comparing ACR with other RAW processors?  What about just a tendency in cameras because of some kind of technical limitations?

Neal
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9011



WWW
« Reply #24 on: May 29, 2007, 04:44:08 PM »
ReplyReply

Quote
However, regarding color models I do think that isssues of digital color have an existing venue for their discussion that happens to be the ICC, which has made a practice of releasing reasonably well documented standards and even sample code. ICC profiles are not closed technology as far as I can see.
Edmund
[a href=\"index.php?act=findpost&pid=119878\"][{POST_SNAPBACK}][/a]

Edmund, have you ever sat in on any ICC meetings about digital camera technology? I have, Thomas, Eric (from Bibble) are both regulars. The ICC has their hearts in the right place. But if a high tech company like Adobe moved at this pace, we'd be on about Photoshop 3 these days.

Look at what's up with the V4 spec (how long has that been around) and what its supposed to do to reduce ambiguity among profile vendors. They recommend using a specific colorspace for communication with the PCS now called PRMG but the vendor doesn't have to use it (they can do what they please). Tight standards are not something you're going to see from this body.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
eronald
Sr. Member
****
Offline Offline

Posts: 3980



WWW
« Reply #25 on: May 29, 2007, 05:13:42 PM »
ReplyReply

Andrew,

 I realize the ICC moves at the speed of a beard growing on a woman's face

 But looking at Adobe sidelining them on camera color with ACR while  Microsoft does an end run around them for screen and print (Windows Color System) should shake them up a bit, one would think.

Edmund

Quote
Edmund, have you ever sat in on any ICC meetings about digital camera technology? I have, Thomas, Eric (from Bibble) are both regulars. The ICC has their hearts in the right place. But if a high tech company like Adobe moved at this pace, we'd be on about Photoshop 3 these days.

Look at what's up with the V4 spec (how long has that been around) and what its supposed to do to reduce ambiguity among profile vendors. They recommend using a specific colorspace for communication with the PCS now called PRMG but the vendor doesn't have to use it (they can do what they please). Tight standards are not something you're going to see from this body.
[a href=\"index.php?act=findpost&pid=120155\"][{POST_SNAPBACK}][/a]
Logged

Edmund Ronald, Ph.D. 
Barry Pearson
Newbie
*
Offline Offline

Posts: 11


WWW
« Reply #26 on: May 29, 2007, 05:35:40 PM »
ReplyReply

In my original post in this thread, I identified that ACR Calibration using the THomas Fors script wasn't, as typically claimed, primarily concerned with variations between different cameras of the same model.

Instead, it appears to provide a systematic change to the colour balance compared with the Adobe default balance - typically reducing the tendency for reds to be too orange, and greens too yellow. (Camera variations are a secondary feature).

I have now obtained more evidence of a consistent bias in the settings resulting from running the Thomas Fors script. I often download raw files that people have posted so that I can examine them. Some of those files are of images of a GertagMacbeth ColorChecker. So I have just run the Thomas Fors script (version 1.0) against those raw files, using ACR 4.0. Here are just the results for the Red sliders:

Canon 350D: Red Hue: -18; Red Sat: 26
Canon 1Ds II: Red Hue: -16; Red Sat: 5
Leaf Aptus 17: Red Hue: -37; Red Sat: 100
Nikon Coolpix 5400: Red Hue: -26; Red Sat: 40
Nikon D70: Red Hue: -15; Red Sat: 8
Nikon D70s: Red Hue: -15; Red Sat: 8

All of those show a negative Red Hue and positive Red Sat, which is a typical result of running such a script. But there was a very interesting exception:

Sigma SD10: Red Hue: 51; Red Sat: -29

That is the first time I've seen a positive Red Hue or a negative Red Sat from one of these scripts. A characteristic of the Sigma X3F raw files is that Adobe use Sigma/Foveon code to do the first stage of colour handling. In other words, when code other than Adobe's is involved, the scripts no longer have the same consistent trend.

In fact, another exception came from the Hasselblad-Imacon H2D camera, which uses DNG as its raw file format, and so has colour matrices supplied by H-I not Adobe:

Hasselblad-Imacon H2: Red Hue: 2; Red Sat: -14

Once Adobe do not control the colours directly, either because Sigma/Foveon do so, or Hasselblad-Imacon do, different results appear.

I ask people here to post their results from doing an ACR Calibration using the Thomas Fors script or similar. Let's try to see what is happening.
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9011



WWW
« Reply #27 on: May 29, 2007, 06:05:54 PM »
ReplyReply

Quote
But looking at Adobe sidelining them on camera color with ACR while  Microsoft does an end run around them for screen and print (Windows Color System) should shake them up a bit, one would think.

Edmund
[a href=\"index.php?act=findpost&pid=120157\"][{POST_SNAPBACK}][/a]

What exactly has Microsoft done that anyone can use?
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
eronald
Sr. Member
****
Offline Offline

Posts: 3980



WWW
« Reply #28 on: May 29, 2007, 06:14:42 PM »
ReplyReply

Quote
What exactly has Microsoft done that anyone can use?
[a href=\"index.php?act=findpost&pid=120162\"][{POST_SNAPBACK}][/a]

Nothing that I know of - but they are certainly threatening to do something.

Old joke: Two chessmasters at a tournament, smoking has been agreed is forbidden. One GM pulls out a huge cigar and places it in his mouth, moves it from mouth to hand etc without lighting it. Second GM appeals to referee - who replies "But, he isn't smoking" And the GM, trembling "But he is threatening to smoke".

Edmund
Logged

Edmund Ronald, Ph.D. 
digitaldog
Sr. Member
****
Offline Offline

Posts: 9011



WWW
« Reply #29 on: May 29, 2007, 06:30:11 PM »
ReplyReply

Quote
Nothing that I know of - but they are certainly threatening to do something.
[a href=\"index.php?act=findpost&pid=120164\"][{POST_SNAPBACK}][/a]

Great. When you see them actually produce something that's not too messy, you let me know.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
eronald
Sr. Member
****
Offline Offline

Posts: 3980



WWW
« Reply #30 on: May 30, 2007, 01:22:40 AM »
ReplyReply

Quote
Great. When you see them actually produce something that's not too messy, you let me know.
[a href=\"index.php?act=findpost&pid=120165\"][{POST_SNAPBACK}][/a]

Non-messy Microsoft software ? hahaha.


Edmund
Logged

Edmund Ronald, Ph.D. 
bjanes
Sr. Member
****
Offline Offline

Posts: 2785



« Reply #31 on: May 31, 2007, 10:00:29 AM »
ReplyReply

Quote
A recent article posted here appears to assume that the purpose of calibrating ACR, (and by extension, Lightroom), using scripts such as the Thomas Fors script, is to cater for differences between cameras of the same model. I disagree!

I believe that the primary purpose of calibrating ACR or Lightroom using such a script is that the default Adobe method of calibration gives results that are unacceptable to many people, while the use of such a script gives results that many of those people prefer. To be very simplistic indeed, these scripts correct the Adobe tendency to give reds that are too orange, and grass that is too yellow.

Have a look at the many results of using such scripts for many camera models that have been published in various forums. They nearly always have a significantly negative Red Hue, and a significantly positive Red Saturation. (The values in the  ACTUAL cameras that Adobe used originally would also give such results!article are respectively: -9; 20). This tendency is obviously systematic, and not to do with within-model variations which would give values either way. I suspect that using such a script on the

These are not results from anti-Adobe people, or people with no knowledge of colour management. These are typically results from people who own a GretagMacbeth ColorChecker, who care enough about their colour balance to use the ColorChecker and a suitable script to try to get things right, and who then share their results so that others can benefit. We don't have all the answers, but surely we can raise serious questions!
[{POST_SNAPBACK}][/a]

Most of us agree that the purpose of calibration with the Fors script or using Bruce Fraser's manual method is to cause the pixel values of the patches to be as close to the published nominal values as possible. The above methods adjust for only the primary colors of the red, blue, and green patches, and one hopes that the values for the other patches will also fall into line.

I used the most recent Fors script (ver 1.0) and ACR 4.0 to calibrate my Nikon D200 and got the following values: tint -1, red hue -20, red sat 21, green hue -4, green sat 8, blue hue 6, and blue sat 2. The red values are in the range Barry has documented. If the purpose of the calibration is merely to compensate for variations from camera to camera in the same model, the values of many calibrations with different cameras should have a bell shaped distribution with a mean of zero. Pending further data, it would seem to me that Barry is on to something.

The standard way to evaluate color error is with ΔE*ab, which is the Euclidian distance between the measured and ideal (reference) values in the a*b* plane of the CIELAB color space as explained by [a href=\"http://www.imatest.com/docs/tour_colorcheck.html]Norman Koren[/url] on his Imatest web site. His Imatest program can plot the ΔEs for the Macbeth Color Checker. Results for many cameras are given on the Imaging Resource web site, and a link to results for the Canon EOS 5D is shown. Readers should refer to the Imatest web site and the Imaging Resource web site for information on the interpretation of these plots. Color error can be with Chroma (saturation), in which case ΔE lies on a radial line extending from the white point, or with hue, where ΔE is not on this radial line. Increased saturation (positive chroma error) is not necessarily bad: many people prefer saturated blue skies and green grass. However, hue error in the form of purple skies or yellowish grass is usually not desired.

Here are plots of my recent calibration for discussion. Exposure was in direct sunlight with the checker mounted on a black background and care was exercised to that no colored objects that could reflect colored light were near the target and exposure was  made according to the Imatest suggestions. The red, blue, and green patches are 13, 14, and 15 on the plots.

ACR default. Note the reds are in the yellow direction as Barry noted.

[attachment=2573:attachment]

ACR calibration with brightness 36 and contrast -34 as determined by the script for the low contrast target:
[attachment=2575:attachment]

ACR calibration with brightness and contrast at their default ACR values of +50 and +25:
[attachment=2576:attachment]

Nikon Capture NX for comparison:
[attachment=2577:attachment]

The Nikon Capture rendering is not that much different from that of the default ACR. The script calibrated results with the contrast and brightness as set by the script gave a very accurate result, but one does not use these contrast and brightness settings for normal photography. With the ACR default contrast and brightness but with the script calibration settings, the overall saturation is increased as indicated by the chroma of 134% as reported by Imatest. However, the chroma corrected color difference (ΔC) is quite small. The increased saturation could be handled easily in ACR or Photoshop with the saturation control, and the calibration was quite successful.

Comments and additional calibration results are welcome.

Bill
« Last Edit: May 31, 2007, 10:01:19 AM by bjanes » Logged
madmanchan
Sr. Member
****
Offline Offline

Posts: 2108


« Reply #32 on: May 31, 2007, 07:13:48 PM »
ReplyReply

Hi Bill,

Thanks for the data. BTW, the delta-E 1976 metric (the Euclidean metric you mention) is a bit dated since many recent studies have shown that CIE L*a*b* is not perceptually uniform. (I'm sure you've seen Bruce Lindbloom's article on this.) Does Imatest let you use some of the newer metrics such as delta E CMC 2:1?

Eric
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2785



« Reply #33 on: May 31, 2007, 09:02:51 PM »
ReplyReply

Quote
Hi Bill,

Thanks for the data. BTW, the delta-E 1976 metric (the Euclidean metric you mention) is a bit dated since many recent studies have shown that CIE L*a*b* is not perceptually uniform. (I'm sure you've seen Bruce Lindbloom's article on this.) Does Imatest let you use some of the newer metrics such as delta E CMC 2:1?

Eric
[{POST_SNAPBACK}][/a]

Eric,

Actually, I used the delta-E 1996, which is reportedly a bit better. Imatest also has a CMC metric, but I am not familiar with its specifics. Here is the  [a href=\"http://www.imatest.com/docs/colorcheck_ref.html#colorerr]Imatest[/url] reference.

Bill
Logged
PeterLange
Guest
« Reply #34 on: June 01, 2007, 11:15:01 AM »
ReplyReply

Quote
Most of us agree that the purpose of calibration with the Fors script or using Bruce Fraser's manual method is to cause the pixel values of the patches to be as close to the published nominal values as possible. The above methods adjust for only the primary colors of the red, blue, and green patches, and one hopes that the values for the other patches will also fall into line.
Out of interest, and with reference to your results: how many degree of HSB-hue does the Light-skin patch no2 become worse due to calibration? Could it be that the hue-accuracy of this memory color is sacrificed to get dead-on Red... At least (and as expected) this trend seems to be obvious from the imatest plots

Btw, great analysis.

Peter

--
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2785



« Reply #35 on: June 01, 2007, 02:18:59 PM »
ReplyReply

Quote
Out of interest, and with reference to your results: how many degree of HSB-hue does the Light-skin patch no2 become worse due to calibration? Could it be that the hue-accuracy of this memory color is sacrificed to get dead-on Red... At least (and as expected) this trend seems to be obvious from the imatest plots

Btw, great analysis.

Peter

--
[a href=\"index.php?act=findpost&pid=120657\"][{POST_SNAPBACK}][/a]

Peter,

Thanks for the kind words.

According to the HSB of the Photoshop color picker and the Imatest results output to a CSV file for analysis, the ideal HSB values for the skin patch are 32, 28, 61. The HSB values for ACR without calibration are 25, 24, 70 and with the calibration they are 20, 22, 76. So the calibration worsens the skin color by 7 degrees of hue. So your observation is a good one.

Since skin tones are a crucial memory color, the results of the calibration are not all good.  

Bill
Logged
juicy
Sr. Member
****
Offline Offline

Posts: 254


« Reply #36 on: June 02, 2007, 08:47:44 AM »
ReplyReply

Hi,

Thanks for the OP and others who have contributed to this interesting topic.

Skin tone reproduction sounds like a very probable and obvious reason  for overall behavior of the profiles built in ACR, especially if we think that the majority of photographs shot in the world represent people (or their cats/dogs   ). The interesting Imatest plots posted by Bjanes show what many have observed, that is a hue shift of reds and greens towards yellow. In commercial printed material  (at least in fashion related advertising) a trend can be seen towards a quite yellow or "golden" tone in skin color that many find very pleasing although it is rarely faithful to "reality". As Andrew Rodney has many times here pointed out most people do neither want nor need scene referred color representation or accuracy. Thus it could be the case that the default color rendering in ACR is directed to make it relatively easy for the "average" user (or a professional digital retoucher) to get at least decent skin color in most photographic situations. Also it would seem logical that the camera manufacturers would tweak the color response of their imaging sensors towards similar direction.

Although in most cases we indeed want pleasing not accurate color, there is a huge number of dslr-cameras used for both commercial product and fashion photography today. In many cases there are very different needs in color reproduction between these (and other) photographic pursuits. Especially in product photography the way to get relatively easily a good perceived match between the objects color and the color in both the file and printed material would be a highly desired situation. Some very succesful product photographers insist on using camera calibration and icc-profiles as the only way to create this match (although proper camera calibration is certainly not an easy or trivial task. Some people also claim that a well produced and edited daylight balanced camera profile may be used in all normal shooting situations to produce also pleasing color if carefully gray-balanced for ambient lighting conditions, ie. not only in studio).

Thus it might be useful to have a set of different "profiles" for different purposes available in ACR. This could be done by making it possible to use icc-profiles or some other system to have "presets" for different image content for different camera models.  Although any particular way of dealing with color may not be perfect for all possible situations, there are more people doing some sort of product or repro photography these days than most people think. Similarly the need for relatively accurate (scene referred) color reproduction without the emphasis on skin tone reproduction is also in great demand. Although one might say that scanning backs and multi exposure MFDBs are the only correct tools for that kind of work, there are many situations where the IQ of the best dslr's is certainly good enough if the color reproduction works well enough.

Personally I would like to hear real world experiences when the Gretag/ACR-calibration script method has worked well (especially for demanding product photography). Also it might be useful to hear comments if the ACR-calibration has failed.

I haven't been too succesful with this aproach (luckily I haven't really needed super accuracy yet...) and I have been wondering if the reason is in my colorchart (I haven't had it measured yet) or in something else I have done. I have tried 10D, 1Ds, 1Ds2, studio flash, sunlight (photographed on a top of a hill without any color reflecting objects near), color corrected daylight fluorescent, tungsten halogen, evennes of lighting within 0-2 rgb levels in the whole chart area in all cases, have tried different lighting directions and types of light (diffused vs harsh). The bottom line is that although the calibration settings are allways towards the same direction (and similar to the settings discussed in earlier posts) and the overal accuracy has been better after running the script, some colors and most notably the red and orange patches are allways off and usually visually worse after calibration. There is also a notable difference between different camera model's color response when photographed in the exact same lighting and the same lens, at least between 10D and 1Ds2 the latter being more neutral as a default. Usually after calibration the overal delta E has been around 2 according to the script.

So my question is if there is a way to edit the target values in the script if after measuring the particular chart in use there happens to be a meaningfull difference between the measured values and the script's default values?  

Thanks for any comments.

Cheers,
Juicy
Logged
Jack Flesher
Sr. Member
****
Offline Offline

Posts: 2595



WWW
« Reply #37 on: June 02, 2007, 12:02:07 PM »
ReplyReply

Quote
What this points to is I think there is a WIDE variation between sensors, even in the same lot and same model...[a href=\"index.php?act=findpost&pid=119583\"][{POST_SNAPBACK}][/a]

Maybe, but...  I daylight calibrated my first 5D, then 6 months later got a second.  I applied the original calibration to the new camera to test it and avoid having to repeat the calibration proceedure.  The converted files are indistinguishable using the same lens on either camera. The worst color difference was 2 points in the R channel and most other patches were either equal or only one point off in one or two channels. Expoure in Av Matrix was barely measurable at less than 1/10th stop different...  Moreover, I supplied that calibration to four of my shooting buddies who also shoot 5D's and they all claim it is "excellent" with their cameras.

So I supect some INTERNAL profiling may be happening with some cameras before they leave the factory -- the performance between this many different 5D's is too consitent to believe otherwise.

Cheers,
Logged

GregW
Sr. Member
****
Offline Offline

Posts: 305


WWW
« Reply #38 on: June 02, 2007, 08:56:45 PM »
ReplyReply

Hi Bill, thought I'd add some grist to the mill.  The calibration results are about 6-8 weeks old, and as you will see not so far away from yours.  The chart was inside in bright, naturally lit room generally free of reflection.

For what it's worth I've added the first few digits of the D200 serial.

D200 8022xxx

AcrCalibrator V1.0 (x89)
ACR 4.0

tint, -2
red hue, -20
red sat, 29
green hue, -2
green sat, 13
blue hue, 8
blue sat, 3
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2785



« Reply #39 on: June 02, 2007, 09:45:43 PM »
ReplyReply

Quote
Hi Bill, thought I'd add some grist to the mill.  The calibration results are about 6-8 weeks old, and as you will see not so far away from yours.  The chart was inside in bright, naturally lit room generally free of reflection.

For what it's worth I've added the first few digits of the D200 serial.

D200 8022xxx

AcrCalibrator V1.0 (x89)
ACR 4.0

tint, -2
red hue, -20
red sat, 29
green hue, -2
green sat, 13
blue hue, 8
blue sat, 3
[a href=\"index.php?act=findpost&pid=120848\"][{POST_SNAPBACK}][/a]

Greg,

Thanks for the feedback. The results are pretty similar and show a strong bias in the red hue and saturation as the OP noted.  FWIW, the SN on my camera is 3005xxx. I wouldn't want to over interpret the data, but if we got more data, we could calculate a standard deviation and confidence interval.

Bill
Logged
Pages: « 1 [2] 3 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad