What is ACR calibration REALLY doing?
Barry Pearson
the Thomas Fors script (various versions) since I first saw it
mentioned in an Adobe forum, with success. But I believe that the
typical explanations of why to use it aren't true!
We are told that the ACR Calibrate tab exists to enable variations
between individual cameras within a model to be catered for. So if my
cameras are significantly different from the ones tested by Adobe to
create the ACR profiles, I can "normalise" them. Bruce Fraser
described a process for setting the sliders in his books, (which I've
read), and Thomas Fors wrote a script to automate the process based on
target values for GretagMacbeth ColorCheckers.
But ... if the PRIMARY effect of using the script were to cater for
variations between cameras, we would expect to see slider values
spread from negative to positive. For most of them, we typically do -
but not for the Red sliders. I've been noting calibration values that
people have obtained by running this script (or sometimes the Rags
Gardner script) and then posted to various forums. And these are
typical (Red) results I've seen:
Nikon D200: Red Hue: -24; Red Saturation: 42
Nikon D2X: Red Hue: -20; Red Saturation: +20
Pentax K100D: Red Hue -27; Red Saturation 40
Pentax K10D: Red hue: -13; Red sat: 7 (mine)
Pentax *istD: Red Hue: -24; Red Saturation: 27 (mine)
Pentax *istDS: Red hue: -26; Red Sat: +35
Panasonic LX1: red hue -20; red saturation +42
Panasonic LX1: Red hue: -26; Red sat: +19
Konica Minolta A2: Red hue: -26; Red sat: 0
Canon G3: Red hue: -1; Red sat: 20
Canon 300D: Red hue: -5; Red sat: 20
Canon 20D: Red hue: -9; Red sat: 9
Canon 10D: Red hue: -5; Red sat: 30
Canon 5D: Red hue: -5; Red sat: 11
Canon 5D: Red hue: -11; Red sat: 4
I have never seen a positive Red Hue, or a negative Red Saturation,
posted to a forum after using one of these scripts! I'm sure they
exist, (one of the above has a 0 Red Saturation), but the bias is
systematic. Is it a real effect, or some sort of surveying bias, or
what? Is there a technical explanation?
I accept that a SECONDARY effect of using the script is to cater for
camera variations, as seen in the values above. But I believe the
PRIMARY effect is to compensate for the fact that the standard Adobe
profiling method results in reds that are often too orange, and greens
that are often too yellow. And these are criticisms of ACR's colour
rendering that have been posted to forums. In other words, it might be
possible to derive a generic set of slider values for (say) the Pentax
K10D which gives results that are prefered by many to the Adobe
defaults, yet are non-zero.
I have posted a similar article to the Luminous Landscape forums, as a
result of an article there about the Thomas Fors script. The point I
made appears to be ignored, and perhaps more data is needed.
http://luminous-landscape.com/forum/index.php?showtopic=17064
I ask that people who have used the Thomas Fors (or Rags Gardner)
script post their results here, so that we can understand why I see
the systematic bias in results posted to forums. (It would be useful
to say whether the result is prefered to the Adobe default! I
certainly prefer the results from the script to the Adobe defaults).
Steve Sprengel
| 300D ACR4 Calibration | |||||||||||||||
| Conditions | Date | CRW_ | Color | Tint | Exp | Shad | Bright | Cont | Shad Tint | Red Hue | Red Sat | Grn Hue | Grn Sat | Blue Hue | Blue Sat |
| Shade at Dusk | 05/10/07 | 8426 | 9700K | 12 | 0.00 | 1 | 55 | 2 | 1 | 0 | 0 | -4 | 18 | 2 | 10 |
| Shade at Dusk behind house | 05/16/07 | 8855 | 7800K | -2 | 0.00 | 1 | 60 | 20 | 0 | -1 | 5 | -9 | 12 | 2 | 1 |
| Northlight shade from full sun | 05/08/07 | 8330 | 6600K | -15 | -0.05 | 5 | 67 | 31 | -2 | -3 | 8 | -11 | -3 | 1 | 10 |
| Cloudy/Drizzle/Day | 05/03/07 | 8034 | 6550K | 2 | 0.20 | 1 | 46 | 20 | 0 | 0 | 9 | -5 | 4 | 1 | 5 |
| Sun behind house | 05/11/07 | 8586 | 5550K | -1 | -1.05 | 2 | 74 | 19 | 0 | -5 | 14 | -5 | 18 | 4 | 1 |
| Fluorescent Tube | 05/01/07 | 7984 | 3650K | 22 | 0.10 | 0 | 61 | 6 | -2 | -13 | 37 | 39 | 20 | -6 | 3 |
| Fluorescent Bulb | 05/02/07 | 8028 | 2850K | 24 | 0.00 | 0 | 64 | 31 | -2 | 12 | 14 | 28 | -31 | -9 | 4 |
| Incandescent Reveal Ceiling | 05/02/07 | 8029 | 2850K | -5 | 0.05 | 0 | 71 | 31 | -2 | 8 | -12 | -24 | 22 | 4 | -1 |
| Incandescent 100W Ceiling | 05/13/07 | 8655 | 2700K | 1 | -0.25 | 3 | 52 | 23 | -1 | 7 | -9 | -13 | 17 | 2 | 3 |
| Sodium Vapor Night - grn prob | 05/13/07 | 8658 | 2000K | 8 | 0.20 | 0 | 42 | 16 | -1 | 11 | -2 | 116 | 49 | -27 | 14 |
Barry Pearson
> The positive/negative values depend on the initial lighting situation. According to Bruce Fraser (on a page linked to from Tom Fors' site) ACR has one internal profile for Incandescent and another internal profile for everything else. In my testing, Cloudy lighting 6500K seems to be the closest to zero. If your survey respondents are not all the same lighting and white-balance, then they cannot be compared with each other.
Adobe use "D65", for daylight, and "Standard-A", for tungsten.
Where possible, I identified a daylight calibration run. That is
certainly the case for the two of mine that I posted, and for some of
the others as well. I don't believe that is the factor here - my non-
daylight values also show the same bias towards negative Red Hue and
positive Red Saturation.
Please post your results, qualifying them as daylight or whatever, and
we can see what the factors are, and how much the illuminant affects
the result. I think what we need now are plenty of cases - then we can
see where to go next.
Steve Sprengel
Thomas Fors
These are interesting observations. One possible explanation I've heard before is that Nikon and Canon (for example) optimize their cameras for reproducing Asian skin tones.
--Tom
Barry Pearson
> Hi Barry,
>
> These are interesting observations. One possible explanation I've heard
> before is that Nikon and Canon (for example) optimize their cameras for
> reproducing Asian skin tones.
Wow! That is thought provoking! And if Silkypix (Japanese) cooperated
with this optimisation, it would make sense - "sort of". (There appear
to be many "western" people who prefer Silkypix colours).
But that still doesn't explain why your script appears to give results
that are SYSTEMATICALLY different from the default Adobe profile. The
colours resulting from use of your script could obviously be achieved
by Adobe if they chose to - after all, you exploit ACR controls that
are available to Adobe too!
Suppose, as I believe but can't prove, that using your script, with
the actual cameras Adobe themselves used to generate their profiles,
ALSO gave these systematically different results. That would be a clue
that the difference is because Adobe and you actually have different
colour intentions/objectives. In effect, you try to ensure that if you
photograph a ColorChecker and have a properly colour-managed system it
continues to look just like a ColorChecker throughout.
But ... what do Adobe attempt to achieve? Do they use a different
target and try to make that consistent? Or is it a more theoretical
calibration? (I know they identify 2 colour matrices that map between
camera colour space and CIE XYZ colour space, but I don't know whether
that is sufficient by itself, or whether there is another factor in
the equation). The joke in some forums is "Adobe must like orange"!
Have you been able to see the values generated for lots of different
cameras by your script? If so, what are the results? What do you get
from your own cameras? I am trying to gather more data in the hope of
being able to identify how consistent this is and what the cause may
be.
I've just rerun the script for own cameras using ACR 4.0 and the 1.0
version of the script. The photographs were in daylight at the start
of March, just minutes apart.
Pentax *istD
Shadow Tint: 1
Red Hue: -24
Red Sat: 27
Green Hue: -5
Green Sat: 0
Blue Hue: 5
Blue Sat: 11
Pentax K10D
Shadow Tint: -3
Red Hue: -13
Red Sat: 6
Green Hue: -4
Green Sat: 9
Blue Hue: 10
Blue Sat: 8
Barry Pearson
> In your request, here, you didn't specify the lighting conditions, so knowing "Daylight" is somewhat useful, although the results for my 300D are a bit different between 5550K and 6500K, so knowing the illumination source as well as the white-balance values for the next-to-lightest patch would be good. I don't have access to D50 or D65 workstations, so my sources are only natural or normal indoor.
I would like to know how much difference the illuminant makes. For
example, are the results still consistently Red Hue negative and Red
Sat positive, or does it change the centre of the spread considerably?
Please post them.
A problem is that it is hard to know what the colour temperature
actually was at the time of capture. I took photographs with my Pentax
*istD and my Pentax K10D within minutes of one another in stable light
conditions, and the Thomas Fors script records the *istD Color Temp as
6550 and the K10D Color Temp as 6000. So: "daylight"!
> As far as why all saturation adjustments seem to be positive for your survey, perhaps ACR's internal calibration is for a non-linear toning curve, medium contrast, for example, but Tom Fors script uses a luminance curve fitted to the neutral patches and would tend to undersaturate things. It would be nice to hear Knolls tell everyone how they calibrate ACR. My 300D's calibration is from ACR 2.4 and I wonder if they have changed how they do it for the newer canons. I see the 5D is almost the same as mine so maybe they haven't.
My *istD values changed a little, but not much, since ACR 2.4.
> I am intrigued that your Pentax camera is so much different than mine. Can I ask you to send me a RAW file that corresponds to the results you posted? I would like to use ACR Calibrator and plot the color error with Imatest to see how your camera differs from mine. If you have GMB CC shots from cloudy and incandescent sources those would be similarly interesting.
[snip]
I'll try to get some suitable raw files to you. I'll need to take some
more shots. In the meantime, please post your own script values,
together with the best statement of the illuminant that you can. I
believe that this effect is so strong that details of the illuminant
are a secondary issue. I think we needs lots of initial data before we
can home in on the areas needing further examination.
Steve Sprengel
Barry Pearson
> Perhaps you are talking about something else, but a couple replies ago I showed my results all the way from < 2000K sodium vapor to 9700K bluish dusk lighting--the first two "yellow" columns. I don't have any way to do other illuminates besides what I find outdoors in various wather and indoors at my home.
I apologise. I didn't pay enough attention to your earlier post, nor
have I made it sufficiently clear enough what I am trying to achieve.
I am trying to identify what the results of the Thomas Fors script are
in comparison to the default Adobe profiling. So I am looking for
pretty-much unaltered results of the script. The script is trying to
match the real values of a ColorChecker, so different values must be
trying to do something else. I believe Adobe is trying to do something
else - but I don't know what.
You said "The following table lists the results with a modified ACR
Calibrator script that uses RGB and CMY color patches which keeps the
color-error from going wild for the yellows". I don't know how to fit
a modified script into what I am trying to do. It is an extra variable
or more that I don't know how to deal with at the moment. I am trying
to identify such things as differences in yellows - you have corrected
for differences in yellows using your own algorithms, and so have
(presumably) hidden the effects I am trying to evaluate.
For the time being, I'll have to concentrate on results from the
standard script.
Steve Sprengel
| 300D ACR4 Calibration - ACR Calibrator using RGB error plus CMY error | |||||||||||||
| Conditions | |||||||||||||
| Color | Tint | Exp | Shad | Bright | Cont | Shad Tint | Red Hue | Red Sat | Grn Hue | Grn Sat | Blue Hue | Blue Sat |
| Shade at Dusk |
| 9700K | 12 | 0.00 | 1 | 55 | 2 | 1 | 0 | 0 | -4 | 18 | 2 | 10 |
| Shade - hazy day | 8150K | 2 | 0.20 | 12 | 67 | 43 | -6 | -10 | 18 | 2 | -40 | 0 | 10 |
| Shade - hazy day | 8150K | 2 | 0.20 | 12 | 67 | 43 | -6 | -8 | 18 | -10 | -14 | 1 | -8 |
| Northlight shade from full sun |
| 6600K | -15 | -0.20 | 13 | 79 | 17 | -2 | -5 | 5 | -1 | -19 | 1 | 19 |
| Northlight shade from full sun |
| 6600K | -15 | -0.05 | 5 | 67 | 31 | -2 | -3 | 8 | -11 | -3 | 1 | 10 |
| Cloudy/Drizzle/Day | 6550K | 2 | 0.10 | 0 | 51 | 25 | 0 | -1 | 18 | -4 | -13 | 2 | 5 |
| Cloudy/Drizzle/Day |
| 6550K | 2 | 0.20 | 1 | 46 | 20 | 0 | 0 | 9 | -5 | 4 | 1 | 5 |
| Sun behind house | 5550K | -1 | -0.75 | 0 | 58 | 22 | 1 | -4 | 30 | -7 | -7 | 5 | -5 |
| Sun behind house |
| 5550K | -1 | -1.05 | 2 | 74 | 19 | 0 | -5 | 14 | -5 | 18 | 4 | 1 |
| Fluorescent Tube |
| 3650K | 22 | 0.10 | 0 | 61 |
| 5 | -2 | -9 | 85 | 29 | 1 | -3 | -18 |
| Fluorescent Tube |
| 3650K | 22 | 0.10 | 0 | 61 | 6 | -2 | -13 | 37 | 39 | 20 | -6 | 3 |
| Fluorescent Bulb | 2850K | 24 | 0.05 | 0 | 61 | 29 | -2 | 12 | 35 | 27 | -54 | -6 | 1 |
| Fluorescent Bulb |
| 2850K | 24 | 0.00 | 0 | 64 | 31 | -2 | 12 | 14 | 28 | -31 | -9 | 4 |
| Incandescent Reveal Closet | 2850K | -5 | 0.10 | 0 | 70 | 31 | -2 | 8 | -7 | -40 | 31 | 9 | 15 |
| Incandescent Reveal Closet |
| 2850K | -5 | 0.05 | 0 | 71 | 31 | -2 | 8 | -12 | -24 | 22 | 4 | -1 |
| Incandescent 100W Ceiling |
| 2700K | 1 | -0.25 | 3 | 52 | 23 | -1 |
| 8 | 5 | -34 | 35 | 8 | -20 |
| Incandescent 100W Ceiling |
| 2700K | 1 | -0.25 | 3 | 52 | 23 | -1 | 7 | -9 | -13 | 17 | 2 | 3 |
| Sodium Vapor Night (WB bad) | 2000K | 8 | 0.30 | 0 | 37 | -10 | -1 | 29 | 28 | 91 | 159 | -16 | -28 |
| Sodium Vapor Night (WB bad) |
| 2000K | 8 | 0.20 | 0 | 42 | 16 | -1 | 11 | -2 | 116 | 49 | -27 | 14 |
MacUser
I don't have much experience with the ColorChecker and the calibration
script but I found out that the camera profile is changing a lot, even
when I thought the conditions are the same.
I do take care of reflections of walls and windows and I avoid colored
shade under trees etc but still I don't get a consistant profile.
I thought that as long as the light source is the same that it didn't
matter how bright the light is, clouded or not, shadow or in bright
sunlight and that always the colors of the output are always fine, so
just 2 or 3 profiles would be enought to cover all outside and inside
images.
I was wrong and it turned out that using a calibrated camera profile
is better then using nothing at all but for sure a camera profile is
not as usefull as I expected because of the huge variations.
I gues the topic starter have a point in thinking that the goal of
Adobe isn't to provide an absolute neutral output but that other
issues (color tone of the skin of people for instance) are involved.
If so then I strongly disaproove. I want to have full control over my
RAW and I don't want that application or scripts makes a color balance
for me. Unless the application of scrupt has a good reason for it and
tells me why/when I should choose it or not.
I tried Steve script as well (Steve I hope you don't mind that I
didn't ask it before installing the script).
There defenitely are differences in the output, I hate it to say
because it sounds ungratefull to Tom (and I am gratefull for his
efforts) but I prefer the modified script of Steve.
The results can be seen here:
http://www.dmmdh.nl/calibratie/standard_script.png
http://www.dmmdh.nl/calibratie/modified_script.png
The RAW was taken a few days ago under a clouded sky on the middle of
the day in the Netherlands with a 5D+Tokina 10-17@17mm fish eye lens.
I used a iMacG5 and a MacBookPro for processing, hence the processing
differences in time.
I hope this info is of some use.
Regards Wim.
Steve Sprengel
MacUser
Thanks for the explanation and the analyzing, it confirms my own
thoughts about it.
The image of the screenshot was zoomed-in at 200% to give a good look
at the colors, the card was actual filling up approx. 1/4 of the
height and width of the image so the vignetting or the curved lines
are not so important because the card was placed in the center of the
image. The card was facing up to the sky almost horizontally placed
and far from the ground on a garden table, I took the image with
stretched arm to prevent as much reflection of my self as possible and
the sun (behind the clouds) was in front of me to avoid any shade on
the card.
So I think that I did what has to be done to take a good shot.
As said before I am new in the Color Calibration scene but I can see
the strenght when using it because it makes post processing so much
easier.
Regards, Wim.
On 29 mei, 15:31, "Steve Sprengel" <s...@sprengels.com> wrote:
> No need to ask permission from me, since the script you are using is from Simon Tindemans who is much more knowledgeable than myself: http://www.xs4all.nl/~tindeman/raw/color_reproduction.html
>
> I posted the link to his script as something I felt was more sophisticated than the modifications I had done, myself, and therefore there was no need to give anyone my script. It looks like his script runs quite a bit faster, too.
>
> Don't be concerned about the discrepancy between the two different scripts, because Tindermans' script uses all the color patches in its calculations, not just the RGB patches used in the Fors' script, so I would guess there would still be a slightly more error with the RGB patches, but less error in other ones, which seems better.
>
> Besides the trees and buildings, your clothing can color the image of the colorchecker, and it probably should be up off the ground to keep the grass from reflecting on it.
>
> Also, the fact you're using such a wide angle lens on a full-frame camera means there could be an issue with vignetting and it might be better to back off and use the middle half of the image.
>
> Here is the color-error analysis from a program called Imatest, where the modified script at the right converges on a better solution from what I can see, particularly in the blues. Things were initially a bit off, but then I realized your screen capture PNGs were posted with your monitor profile instead of sRGB, and once I converted them to sRGB the numbers were closer. These are a-b-error plots and as such the saturation errors are generally differences in and out relative to the center the center and hue errors are a rotation around the center. I would say the modified script gives very good calibration values with both the average and maximum error being less.
>
> The results can be seen here:http://www.dmmdh.nl/calibratie/standard_script.pnghttp://www.dmmdh.nl/calibratie/modified_script.png
> MacUser_combined.png
> 109KWeergevenDownloaden
Barry Pearson
> I am intrigued that your Pentax camera is so much different than mine. Can I ask you to send me a RAW file that corresponds to the results you posted? I would like to use ACR Calibrator and plot the color error with Imatest to see how your camera differs from mine. If you have GMB CC shots from cloudy and incandescent sources those would be similarly interesting.
You didn't say which Pentax. I've uploaded the PEF from my K10D that I
used for my latest calibration results using ACR 4.0 and the 1.0
script that I recently published here. I'll keep it here for just a
few days:
http://www.barrypearson.co.uk/photography/temp/_IGP0101.PEF
I look forward to seeing your analysis. If it is the *istD PEF you
wanted, I'll upload that instead.
Thomas Fors
I tried Steve script as well (Steve I hope you don't mind that I
didn't ask it before installing the script).
There defenitely are differences in the output, I hate it to say
because it sounds ungratefull to Tom (and I am gratefull for his
efforts) but I prefer the modified script of Steve.
Hi Wim,
No worries about sounding ungrateful. When I first developed the script I explored using more than just the R,G,B patches to do the color calibration. For some cameras it gave better results and for others it made things worse. Using just R,G,B seemed to work fairly reliable for all the cameras I tested on.
I am willing to revisit this though based on the interest it has gained here. So, your feedback about which results you prefer is greatly appreciated.
--Tom