Luminous Landscape Forum
Raw & Post Processing, Printing => Colour Management => Topic started by: ErikKaffehr on February 13, 2014, 12:45:44 am
-
Hi,
A few photographers have recently tested profiles made for the IQ-250 with other sensors like the D800 and Canon and seems to have achieved more satisfactory results.
I was for some time thinking about camera profiles, and I have a few questions:
1) How are they made? Are they using spectral sensivity data to calculate response for different illuminants?
2) Do camera profiles convert raw data in CIE X,Y,Z or into working space like Prophoto RGB? Converting to X, Y, Z makes a lot of sense to me.
3) Having a good conversion to working space (like Prophoto RGB) and having rendering profiles (or presets) seems to me a better approach than combining both into a single profile.
4) How much of this is solid math and how much black magic?
I have some very basic understanding of how things work, but I feel that there is a lot of confusion about how profiles are handled, and what they mean, in raw conversion.
Would be nice the hear from those who know, like some barks from the forum terrier bit bull or the digital dog. Comments from the man in chains would also be most interesting.
Best regards
Erik
-
Would be nice the hear from those who know, like some barks from the forum terrier or the digital dog. Comments from the man in chains would also be most interesting.
I'll let Eric decide how much to say...so, are you calling me a terrier? A TERRIER? If you were referring to me, I would much rather be compared to a bit bull, just saying :~)
So, your question is pretty wide open. Are you referring to DNG Profiles or ICC camera profiles. There's a big difference...
If you are referring to DNG Profiles, at Adobe, Eric or Thomas (or maybe a new guy) use a pretty exotic piece of hardware to to generate the Adobe Standard DNG profiles. Exactly how they do it may be proprietary...but they are profiled at D65 and Standard Illuminate A (2850ºK, I think). In ACR/LR, the resulting dual illuminate profile is tweened between the two illuminates and extrapolated when above or below the K range.
As far as I know, the camera color is interpreted into CIE XYZ and the internal working space is linear gamma, ProPhoto RGB. The transform for the final output profile is also, I think, done in CIE XYZ. In the processing pipeline, a lot of "magic" occurs...some processes are only on luminance (like sharpening) other processes are in HSB (it says HSL but it's a misnomer, I'm pretty sure Eric told me it's actually working in HSB which is a tiny bit different than HSL).
Not sure what you are asking in Q3...wanna try again?
In terms of the math, well, Thomas Knoll figured it all out. Thomas knows the math really well (he invented some of it) but the thing about Thomas is that's he's a master of "magic numbers", which means he' knows what the results are that he wants and if the math doesn't quite get it, he's perfectly happy and willing to tweak the numbers the way he wants.
I watched Thomas reverse engineer the raw file from one of my cameras (an early Canon D30) before ACR was actually released...it was interesting but kinda like watch corn grow...it got boring pretty quick but Thomas was fascinated. Back then he shot two raw shots with an X-Rite ColorChecker DC in a SpectraView Jr light booth under D65 and Standard Illuminate A. The process has evolved a lot since then :~)
BTW, you can learn a lot about DNG Profiles if you read the full DNG Spec.
If you are asking about camera ICC profiles, sorry, I don't think they are particularly useful...(even though Capture One uses them).
-
Hi,
Terrier issue fixed…
Regarding Q3, there may be different profiles. My questions is really if those profiles are applied before or after conversion to XYZ. The question may make little sense, but may give cause to some interesting discussion.
Thanks a lot for thoughtful response! And I hope you excuse me for the small joke.
Best regards
Erik
I'll let Eric decide how much to say...so, are you calling me a terrier? A TERRIER? If you were referring to me, I would much rather be compared to a bit bull, just saying :~)
So, your question is pretty wide open. Are you referring to DNG Profiles or ICC camera profiles. There's a big difference...
If you are referring to DNG Profiles, at Adobe, Eric or Thomas (or maybe a new guy) use a pretty exotic piece of hardware to to generate the Adobe Standard DNG profiles. Exactly how they do it may be proprietary...but they are profiled at D65 and Standard Illuminate A (2850ºK, I think). In ACR/LR, the resulting dual illuminate profile is tweened between the two illuminates and extrapolated when above or below the K range.
As far as I know, the camera color is interpreted into CIE XYZ and the internal working space is linear gamma, ProPhoto RGB. The transform for the final output profile is also, I think, done in CIE XYZ. In the processing pipeline, a lot of "magic" occurs...some processes are only on luminance (like sharpening) other processes are in HSB (it says HSL but it's a misnomer, I'm pretty sure Eric told me it's actually working in HSB which is a tiny bit different than HSL).
Not sure what you are asking in Q3...wanna try again?
In terms of the math, well, Thomas Knoll figured it all out. Thomas knows the math really well (he invented some of it) but the thing about Thomas is that's he's a master of "magic numbers", which means he' knows what the results are that he wants and if the math doesn't quite get it, he's perfectly happy and willing to tweak the numbers the way he wants.
I watched Thomas reverse engineer the raw file from one of my cameras (an early Canon D30) before ACR was actually released...it was interesting but kinda like watch corn grow...it got boring pretty quick but Thomas was fascinated. Back then he shot two raw shots with an X-Rite ColorChecker DC in a SpectraView Jr light booth under D65 and Standard Illuminate A. The process has evolved a lot since then :~)
BTW, you can learn a lot about DNG Profiles if you read the full DNG Spec.
If you are asking about camera ICC profiles, sorry, I don't think they are particularly useful...(even though Capture One uses them).
-
Regarding Q3, there may be different profiles. My questions is really if those profiles are applied before or after conversion to XYZ. The question may make little sense, but may give cause to some interesting discussion.
Well, if you are referring to DNG profiles, well, it's complicated...
By default, when you open an image in ACR/LR, the metadata is read from the camera and ACR/LR picks a white balance starting point based on the metadata. If the WB is between the dual illuminates, ACR/LR interpolates between the dual illuminates. As you adjust the WB, the modifications are previewed. So, the way the DNG profile works is to first convert from camera color and then into the working space by way of the DNG profiles and the white balance selected and then transform the preview into the display color space (again using, I think, CIE XYZ). A lot of under the hood stuff happens up to the point where you process...it's pretty messy under that hood, Eric tried to explain some of it to me, but it made my eyes glaze over :~)
The final images after running through the pipeline is then transformed to the output profile selected. In ACR 8.x in Photoshop CC, that output profile can be any ICC profile including RGB, Lab, CMYK or grayscale. In LR only RGB color profiles are supported.
-
Well, if you are referring to DNG profiles, well, it's complicated...
it is described in http://www.adobe.com/content/dam/Adobe/en/products/photoshop/pdfs/dng_spec_1.4.0.0.pdf and for people able to read the code also DNG SDK code is a book to read...
-
http://forums.adobe.com/message/3183090#3183090
http://forums.adobe.com/message/3958507#3958507
just scan posts from Eric Chan and you shall see a lot of comments about how Adobe software works (DNG profiles)
-
Hi,
A few photographers have recently tested profiles made for the IQ-250 with other sensors like the D800 and Canon and seems to have achieved more satisfactory results.
I was for some time thinking about camera profiles, and I have a few questions:
1) How are they made? Are they using spectral sensivity data to calculate response for different illuminants?
Hi Erik,
Since you are referring to the new PhaseOne IQ250 ICC profile, as used in CaptureOne, I think the DNG profile info you already got is not going to answer your questions fully. Do note that the DNG profiles used in LR/ACR are something different from ICC profiles.
Although they may use a different starting point (e.g. a monochromator), I think PhaseOne starts out by shooting an image (more likely a series at different exposure levels) of a Digital ColorChecker SG chart (http://www.bhphotovideo.com/c/product/465295-REG/X_Rite_MSDCCSG_Digital_ColorChecker_SG_Card.html), which uses 140 color patches instead of the 24 of a common, or Passport, Colorchecker. They could also use an IT8 standard chart, e.g. one from Wolfgang Faust (http://www.targets.coloraid.de/ (http://www.targets.coloraid.de/)) which are usually of high quality.
Shooting such a target requires a very neutral shooting environment/studio/lab, because we do not want any ambient light influences/reflections to be mixed in with the main illuminant's spectral output. Also surface reflections from the illuminants must be minimized as much as possible (as in a good repro setup), since that will reduce the saturation of the patches and produce over-saturated/compensated results, and light-fall off must be miniimized.
The chart comes with an accompanying reference data set (usually as a file that comes with the profiling software, or one can read the actual chart colors with a spectrophotometer), and the profiling software will then compare the camera's recorded colors after White Balancing and demosaicing with the target colors, adjust the input and create look-up tables that convert the Raw demosaiced data into the desired/expected input data. That calibrated data can then be converted with a Profile to the Working space and finally to the Output colorspace. Such profiles are Scene referred profiles, and as such well suited for transformations through the imaging chain when the other components are also well profiled. The only drawback is that they are most accurate for the exact same lighting conditions as existed during shooting of the target. The more the actual scene lighting conditions differ, the more inaccuracies can develop (but then we usually do not want accurate colors but rather pleasing colors anyway, and our eyes do not respond exactly the same as RGB trichromatic sensors either).
The software that creates the (calibration) lookup tables and ICC profile, can influence/adjust several things in the creation of that profile, which will e.g. result in more (or less) contrast and saturation. But it also allows to shift the accuracy of interpolation for certain color ranges or even the response for those colors which can shift their colors, but at the same time compress or expand other colors. It is also possible to adjust the colors of the Raw demosaiced input file for the profiling software, which will then try to compensate in the opposite direction.
2) Do camera profiles convert raw data in CIE X,Y,Z or into working space like Prophoto RGB? Converting to X, Y, Z makes a lot of sense to me.
Camera profiles (ICC type) just translate from one set of coordinates to an other set of coordinates, and there are several almost lossless conversions between coordinates systems possible. XYZ or L*a*b are common device independent coordinate systems, and it is possible to convert calibrated RGB colors into and from those coordinates.
4) How much of this is solid math and how much black magic?
I think a lot of it is solid math, although the math is adjustable to allow and deal with imperfect (trichromatic) input and match with spectrally accurate data, which is already a bit of magic (some profilers produce better results than others, so it's not straightforward calculus). But the result will not be perfect, and the tweaking that will then be done could be coined as black ( or white?) magic.
A look at some tutorials (http://xritephoto.com/ph_learning.aspx (http://xritephoto.com/ph_learning.aspx), http://xritephoto.com/ph_learning.aspx?action=guide (http://xritephoto.com/ph_learning.aspx?action=guide)) for X-Rite's i1Profiler, or at the descriptions that come with the Argyll Color Management suite (http://www.argyllcms.com/doc/ArgyllDoc.html), may also help to understand better what happens in the process of profile generation.
Cheers,
Bart
-
hi Bart,
I am somewhat familiar with profiling as I have been making profiles for both scanners and printers, but I am a bit confused. The intention of a profile is generally to specify conversion to a 'connection space' that is to my understanding normally CIE XYX.
ICC profiles are normally definied by primaries, or colorants as far as I understand but cameras are quite different.
Another interesting question is profiling. Camera gamuts are very wide as they are sensitive to pretty much the whole visible spectrum. So a very large color gamut is needed for handling in camera colour. Using a printed target like Faust targets cannot cover the much larger sensor colour space as both traditional and chromogenic printing has quite small gamut.
It is very much possible that the ColorCheckerSG card has a wider gamut but the normal ColorChecker fits almost entirely in sRGB.
I am not so much looking for the answers but I hope to see a discussion about different approcahes, which would broaden our understanding of profile generation and usage.
Best regards
Erik
Hi Erik,
Since you are referring to the new PhaseOne IQ250 ICC profile, as used in CaptureOne, I think the DNG profile info you already got is not going to answer your questions fully. Do note that the DNG profiles used in LR/ACR are something different from ICC profiles.
Although they may use a different starting point (e.g. a monochromator), I think PhaseOne starts out by shooting an image (more likely a series at different exposure levels) of a Digital ColorChecker SG chart (http://www.bhphotovideo.com/c/product/465295-REG/X_Rite_MSDCCSG_Digital_ColorChecker_SG_Card.html), which uses 140 color patches instead of the 24 of a common, or Passport, Colorchecker. They could also use an IT8 standard chart, e.g. one from Wolfgang Faust (http://www.targets.coloraid.de/ (http://www.targets.coloraid.de/)) which are usually of high quality.
Shooting such a target requires a very neutral shooting environment/studio/lab, because we do not want any ambient light influences/reflections to be mixed in with the main illuminant's spectral output. Also surface reflections from the illuminants must be minimized as much as possible (as in a good repro setup), since that will reduce the saturation of the patches and produce over-saturated/compensated results, and light-fall off must be miniimized.
The chart comes with an accompanying reference data set (usually as a file that comes with the profiling software, or one can read the actual chart colors with a spectrophotometer), and the profiling software will then compare the camera's recorded colors after White Balancing and demosaicing with the target colors, adjust the input and create look-up tables that convert the Raw demosaiced data into the desired/expected input data. That calibrated data can then be converted with a Profile to the Working space and finally to the Output colorspace. Such profiles are Scene referred profiles, and as such well suited for transformations through the imaging chain when the other components are also well profiled. The only drawback is that they are most accurate for the exact same lighting conditions as existed during shooting of the target. The more the actual scene lighting conditions differ, the more inaccuracies can develop (but then we usually do not want accurate colors but rather pleasing colors anyway, and our eyes do not respond exactly the same as RGB trichromatic sensors either).
The software that creates the (calibration) lookup tables and ICC profile, can influence/adjust several things in the creation of that profile, which will e.g. result in more (or less) contrast and saturation. But it also allows to shift the accuracy of interpolation for certain color ranges or even the response for those colors which can shift their colors, but at the same time compress or expand other colors. It is also possible to adjust the colors of the Raw demosaiced input file for the profiling software, which will then try to compensate in the opposite direction.
Camera profiles (ICC type) just translate from one set of coordinates to an other set of coordinates, and there are several almost lossless conversions between coordinates systems possible. XYZ or L*a*b are common device independent coordinate systems, and it is possible to convert calibrated RGB colors into and from those coordinates.
I think a lot of it is solid math, although the math is adjustable to allow and deal with imperfect (trichromatic) input and match with spectrally accurate data, which is already a bit of magic (some profilers produce better results than others, so it's not straightforward calculus). But the result will not be perfect, and the tweaking that will then be done could be coined as black ( or white?) magic.
A look at some tutorials (http://xritephoto.com/ph_learning.aspx (http://xritephoto.com/ph_learning.aspx), http://xritephoto.com/ph_learning.aspx?action=guide (http://xritephoto.com/ph_learning.aspx?action=guide)) for X-Rite's i1Profiler, or at the descriptions that come with the Argyll Color Management suite (http://www.argyllcms.com/doc/ArgyllDoc.html), may also help to understand better what happens in the process of profile generation.
Cheers,
Bart
-
Hi Jeff,
Getting back to DNG profiles. It is my understanding that the camera profiles are built using solid color science, spectral data and a bit magic. In general I have been quite happy with Adobe Standard profile on my Sonys but not happy att all with the standard profile for my P45+.
So I generated a dual illuminant profile using the Adobe DNG Profile Editor. My understanding is that I still use the Adobe Standard Profile but making adjustments to it. So calibration is not based on my 24 ColorChecker patches but on all inputs used for the Adobe Standard profile plus my adjustments. Is that correct?
Best regards
Erik
Well, if you are referring to DNG profiles, well, it's complicated...
By default, when you open an image in ACR/LR, the metadata is read from the camera and ACR/LR picks a white balance starting point based on the metadata. If the WB is between the dual illuminates, ACR/LR interpolates between the dual illuminates. As you adjust the WB, the modifications are previewed. So, the way the DNG profile works is to first convert from camera color and then into the working space by way of the DNG profiles and the white balance selected and then transform the preview into the display color space (again using, I think, CIE XYZ). A lot of under the hood stuff happens up to the point where you process...it's pretty messy under that hood, Eric tried to explain some of it to me, but it made my eyes glaze over :~)
The final images after running through the pipeline is then transformed to the output profile selected. In ACR 8.x in Photoshop CC, that output profile can be any ICC profile including RGB, Lab, CMYK or grayscale. In LR only RGB color profiles are supported.
-
Camera gamuts
purists shall come and say that camera does not have a gamut
-
So I generated a dual illuminant profile using the Adobe DNG Profile Editor. My understanding is that I still use the Adobe Standard Profile but making adjustments to it.
you are always using some selected base profile (that's how Adobe tries to prevent you from getting absolutely ugly results - and one of the reasone why hoi polloi is running around screaming how good they are in building dcp profiles with a sub $100 24 patch target using simple tools... unlike building non dcp profiles where you do not have hidden suspenders in most/all cases).
So calibration is not based on my 24 ColorChecker patches but on all inputs used for the Adobe Standard profile plus my adjustments. Is that correct?
http://forums.adobe.com/message/1219195#1219195
http://forums.adobe.com/message/5198903#5198903
http://forums.adobe.com/message/3179876#3179876
http://forums.adobe.com/message/3395534#3395534
http://forums.adobe.com/message/3861785#3861785
http://forums.adobe.com/message/5299746#5299746
http://forums.adobe.com/message/4706036#4706036
http://www.luminous-landscape.com/forum/index.php?topic=56863.msg461536#msg461536
http://www.luminous-landscape.com/forum/index.php?topic=30747.msg249068#msg249068
-
Camera gamuts are very wide as they are sensitive to pretty much the whole visible spectrum. So a very large color gamut is needed for handling in camera colour. ...
Not sure that I agree. An achromatic camera might be sensitive to a broad spectrum, but would you need large color gamut for handling its colour?
I would rather think that a large color gamut is needed for handling files that stem from cameras with narrow pass-bands (little overlap).
So what is being said in this thread is that my humble xrite passport is a waste of time and money? I do feel that I have more "accurate" color now than either Adobe or camera manufacturers can provide. Granted, I do believe that what anyone claims to be a truth based on subjective impressions should be treated with a pinch of salt (that includes my own impressions of color accuracy).
-h
-
An achromatic camera might be sensitive to a broad spectrum, but would you need large color gamut for handling its colour?
if "I" decide to do a color transform of a certain kind then yes... "I" am who decides how to do the mapping and "I" might do some exotic ones, taste regardless
-
So what is being said in this thread is that my humble xrite passport is a waste of time and money?
depends on your goal... if you are trying to account for some unusual light then probably makes sense to some extent, however if you are dealing with a regular light then what you really is doing it is not a creation of an "accurate" profile, but rather you are just getting rid of Adobe's color rendering (LUTs) that they (whoever, Eric, etc) decided is good for populus and put in their profiles, that's it... and then why not just edit the profile directly in Adobe PE ? or gut it with dcptool ? or both ? some more advanced folks (like VitNovak from adobe forums) even have their own software for dcp profiles creation.
-
Hi,
I made the interesting observation that although the Adobe Standard profile was decently accurate on the ColorChecker it turned out some bad color on my first test shot. I did a calibration with Adobe DNG Profile Editor and got a good colour conversion with slightly improved accuracy.
So this calibration was completely mechanical, now tuning at all.
(http://echophoto.dnsalias.net/ekr/Articles/MFDJourney/Screenshots/P45+_AdobeSTD_vs_DNGProfiler_small.png)
I have done a lot of testing since than,some of it described here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition , but I decided to stay with LR 5.3 which suits me much better than C1.
My guess is that there may here are the Delta E mean values according to Imatest (sigma in paranthesis):
Capture One: 7.45 (8.4)
Adobe Standard (P45+): 5.65 (6.5)
DNG Profile Editor: 4.71 (5.31)
Adobe Standard (SLT 99): 4.1 (4.52)
For completeness: I did some manipulation in DNG Profile Editor later, reducing saturation on yellowish greens and moving from yellow to green.
But, I got a hint from Iliah that my initial shot used for calibration was bad so I made a new one with better controlled conditions and that is the one I am using on the P45+ now. That one is untouched.
As you have pointed out, DNG Profile Editor just tweaks the base profile, so I was in no way creating a profile from scratch. That is also the reason I still use Adobe Standard profile on the SLT, if it ain't broken why fix it?
By the way, Jeff posted a message where he said that Adobe Standard was less than optimal on his P65+, but he generated a DNG profile and was perfectly happy. I guess that there may be sample variation in P45+ that is contained in the calibration data in the raw files.
Best regards
Erik
depends on your goal... if you are trying to account for some unusual light then probably makes sense to some extent, however if you are dealing with a regular light then what you really is doing it is not a creation of an "accurate" profile, but rather you are just getting rid of Adobe's color rendering (LUTs) that they (whoever, Eric, etc) decided is good for populus and put in their profiles, that's it... and then why not just edit the profile directly in Adobe PE ? or gut it with dcptool ? or both ? some more advanced folks (like VitNovak from adobe forums) even have their own software for dcp profiles creation.
-
I made the interesting observation that although the Adobe Standard profile was decently accurate on the ColorChecker it turned out some bad color on my first test shot.
define "bad color" ? if you don't like it then you goal was to get rid of Adobe's idea of what colors shall be in some specific area (some reds or whatever) and then shooting a target is accomplishing just that (and not actually building a genuinely new profile from scratch - unless you are using tools that do not start from Adobe's base profiles at all), but then you could just change those colors in Adobe PE directly and/or play with dcp content using dcptool... now what was your goal really and did you achieve it consciously or just tried and liked, but w/o understanding of what happened really ? no offense meant
-
Hi,
The flower in my sample was red, but was reproduced as bluish red on the P45+. With the Sony Alpha 99 it reproduced correctly. In both cases using Adobe standard profile with each camera.
I than shot a ColorChecker and generated a DCP profile (based on the Adobe standard profile), using Adobe DNG Profile Editor. I did nothing to that profile at this stage and it reproduced the red flower as a red flower.
Adobe Standard profile did in this case not give the same rendition on the P45+ as on the Alpha 99, although Delta E on the ColorChecker was pretty good.
At this stage I did no tweaking manually at all. The profiles I am using now are all generated from ColorChecker exposures with no manipulation.
I am essentially satisfied with colour rendition on the Sony Alpha, so I use the Adobe Standard profile with that camera. Delta E is pretty decent on that one, too.
Best regards
Erik
define "bad color" ? if you don't like it then you goal was to get rid of Adobe's idea of what colors shall be in some specific area (some reds or whatever) and then shooting a target is accomplishing just that (and not actually building a genuinely new profile from scratch - unless you are using tools that do not start from Adobe's base profiles at all), but then you could just change those colors in Adobe PE directly and/or play with dcp content using dcptool... now what was your goal really and did you achieve it consciously or just tried and liked, but w/o understanding of what happened really ? no offense meant
-
Hi,
Pretty much my understanding. Nice to have it confirmed.
Best regards
Erik
you are always using some selected base profile (that's how Adobe tries to prevent you from getting absolutely ugly results - and one of the reasone why hoi polloi is running around screaming how good they are in building dcp profiles with a sub $100 24 patch target using simple tools... unlike building non dcp profiles where you do not have hidden suspenders in most/all cases).
-
Hi,
The flower in my sample was red, but was reproduced as bluish red on the P45+. With the Sony Alpha 99 it reproduced correctly. In both cases using Adobe standard profile with each camera.
I than shot a ColorChecker and generated a DCP profile (based on the Adobe standard profile), using Adobe DNG Profile Editor. I did nothing to that profile at this stage and it reproduced the red flower as a red flower.
Adobe Standard profile did in this case not give the same rendition on the P45+ as on the Alpha 99, although Delta E on the ColorChecker was pretty good.
At this stage I did no tweaking manually at all. The profiles I am using now are all generated from ColorChecker exposures with no manipulation.
I am essentially satisfied with colour rendition on the Sony Alpha, so I use the Adobe Standard profile with that camera. Delta E is pretty decent on that one, too.
Best regards
Erik
I think p45+ profile (Adobe Standard) is an quite old tech = matrix part doing the main color transform work + single ProfileLookTableData 3D LUT being applied postexposure corrections (and created who knows when and not recently reprofiled - what if P1 did some changes for P45+ CFAs during the life of the line ? ) while Sony A99 profile is the recent created w/ whatever current state of Adobe's tech is
-
Hi,
The approach solved my problem. I don't know if Phase One made changes to CFA on the P45+ but Kodak may have done that or simply have inconsistent CFAs. As far as I know Phase One does a long calibration of each back (several hours), I presumed that may include spectral calibration, too. But perhaps they just map out bad pixels/columns.
Best regards
Erik
I think p45+ profile (Adobe Standard) is an quite old tech = matrix part doing the main color transform work + single ProfileLookTableData 3D LUT being applied postexposure corrections (and created who knows when and not recently reprofiled - what if P1 did some changes for P45+ CFAs during the life of the line ? ) while Sony A99 profile is the recent created w/ whatever current state of Adobe's tech is
-
People might find the article starting on page 30 useful. It is about how your displays do things. Some of it is quite surprising.
Display Myths Shattered
BY DR. RAYMOND SONEIRA
CREATOR OF THE DISPLAYMATE TESTING SUITE
http://dl.maximumpc.com/Archives/MPC0710-web.pdf (http://dl.maximumpc.com/Archives/MPC0710-web.pdf)
-
Thanks!
Lot of good links giving insights!
Best regards
Erik
you are always using some selected base profile (that's how Adobe tries to prevent you from getting absolutely ugly results - and one of the reasone why hoi polloi is running around screaming how good they are in building dcp profiles with a sub $100 24 patch target using simple tools... unlike building non dcp profiles where you do not have hidden suspenders in most/all cases).
http://forums.adobe.com/message/1219195#1219195
http://forums.adobe.com/message/5198903#5198903
http://forums.adobe.com/message/3179876#3179876
http://forums.adobe.com/message/3395534#3395534
http://forums.adobe.com/message/3861785#3861785
http://forums.adobe.com/message/5299746#5299746
http://forums.adobe.com/message/4706036#4706036
http://www.luminous-landscape.com/forum/index.php?topic=56863.msg461536#msg461536
http://www.luminous-landscape.com/forum/index.php?topic=30747.msg249068#msg249068
-
Hi,
My understanding right now is that the ICC profiles used by Capture One are applied after conversion to an internal RGB colour space. So:
- ICC profiles cannot address issues of raw conversion
- The ICC profiles don't have a color transformation matrix (from sensor RGBG to CIE XYZ)
- Generating an ICC profile from raw data requires knowledge of the conversion process in C1
Are the above statements correct?
Best regards
Erik
-
Hi,
Anyone knows what is the internal colour space in Capture One?
Best regards
Erik
-
Hi,
My understanding right now is that the ICC profiles used by Capture One are applied after conversion to an internal RGB colour space.
Hi Erik,
No conversion, the R/G/B channel output of the sensor as quantized by the ADC is the input for CaptureOne, just like it is for all other Raw convertors. A profile is created based on the (maybe White balanced) RGB channel input, where known spectral input (Illuminant and XYZ coordinates) and resulting R/G/B channel output is characterized and calibrated. This is normal ICC input profile creation stuff.
So:
- ICC profiles cannot address issues of raw conversion
- The ICC profiles don't have a color transformation matrix (from sensor RGBG to CIE XYZ)
- Generating an ICC profile from raw data requires knowledge of the conversion process in C1
Raw conversion only plays a special role in reconstructing per pixel color and luminosity. The R/G/B channel data for uniform patches of a given color are not changed by Raw conversion. What can be changed is the assumption of the Illuminant that was used for the exposure, and the relative channel amplification and black points which control White balancing. Some Raw formats use a scaling to some channels to achieve high ADU values that may help with data compression and other calculations, but the part of the response curve of silicon that is used by the ADC input is almost linear, so simple offset and multiplier schemes are common. Sony uses a non-linear encoding though (for highlight compression it progressively skips ADU values), as does Leica with a Gamma adjusted set of ADUs.
ICC Camera profiles do not use a simple matrix conversion, they use the coordinates of an Illuminant, and three Color Primaries, and in addition they use an accurate LUT per color channel that characterizes the capture device (its non-linear behavior between the various coordinates in 3D gamut space). While much more accurate than a simple matrix conversion, it's only more accurate under the exact same shooting conditions (illuminant/lens/camera/CFA/sensor). Luckily, it's possible over a reasonably wide range to recalculate the color response for e.g. a different illumination.
Cheers,
Bart
-
ICC Camera profiles do not use a simple matrix conversion
you can have a profile stored in icc container for a simple matrix conversion, but certainly it is not what C1 uses.
-
you can have a profile storied in icc container for a simple matrix conversion, but certainly it is not what C1 uses.
Hi,
Theoretically yes, but it would be a rare case to not also include (inevitable) non-linearities, in the case of a real device instead of a conceptual model. I assumed we're talking about real cameras/scanners.
Cheers,
Bart
-
Hi,
I am quite a bit doubtful about some statements here. I have never seen a paper describing conversion methodology in C1.
I see two conflicts, one is that there has been a long discussion betweem Vladimirovich and Esben HR (who happens to be the engineer handling the color conversion pipeline at P1) and I think Esben quite clearly says that the ICC profile is applied after raw conversion: http://www.luminous-landscape.com/forum/index.php?topic=82891.msg672308#msg672308
I guess you need to read the full thread, to make any sense of it.
The other conflict is that it has been reported by "Synn" and also "Paul2660" that profiles for IQ-250 and and even for the CCD cameras give better results in Capture One than the original D800 profile. Now, assuming that the D800 has a more permissive or less "orthogonal" CFA design compared to the CCD cameras it would be a bit absurd that profiles generated for an entirely different CFA would give better results than a profile optimized for the camera (D800) and it's CFA.
If we assume that raw conversion is entirely separate from the IIC-profile, that is, the ICC profile is applied to a converted image, both conflicts would be avoided.
So my guess is that the ICC profiles in C1 are ignorant about CFA and CFA is handled in C1 workflow before ICC profiles are applied.
Regarding accuracy, I have compared C1 to Adobe raw engine a few times and the Adobe engine was always more accurate, at least when measured as Delta E n the colour checker. I am fully aware that the colour checker just represents 16 (or 24) of the several million colours we see. My take is that it is possible that C1 yields better colours the ACR/LR but they are less accurate.
Best regards
Erik
Hi,
Theoretically yes, but it would be a rare case to not also include (inevitable) non-linearities, in the case of a real device instead of a conceptual model. I assumed we're talking about real cameras/scanners.
Cheers,
Bart
-
Maybe it is your setup of C1. C1 is widely regarded to have excellent color. The fact your Imatest variances all show a blue vector on the errors and Bart noted a D65 vs D60 difference may let you fix how C1 operates on your system if you are interested. Just the fact so many people put up so many good images using C1 should tell you it is not the software. If C1 has a weakness relative to the other top raw converters it is in an artificial look to fine detail, sometimes even mazing. It is not on color.
-
and I think Esben quite clearly says that the ICC profile is applied after raw conversion: http://www.luminous-landscape.com/forum/index.php?topic=82891.msg672308#msg672308
Esben escaped w/o answering the questions fully (Eric Chan to his /Adobe's/ credit gives more clear answers)... and define "raw conversion" first... do you define that as (A) something that ends with demosaicking or (B) something that ends w/ color transform from "scene referred" camera/sensor coordinates into some colorimetric color space (like cieXYZ/D50 or whatever) ?
-
than a profile optimized for the camera (D800) and it's CFA.
I already suggested a conspiracy theory if there was a need of P1 to make an excellent profile for D800 that eats somewhat into their market (some segments of it)
-
Theoretically yes, but it would be a rare case to not also include (inevitable) non-linearities, in the case of a real device instead of a conceptual model. I assumed we're talking about real cameras/scanners.
if you are talking about profiles shipped by manufacturer like P1, but if you are talking about homemade profiles - not at all... from one's horse mouth = http://www.dpreview.com/forums/post/38768757
-
Hi,
Esbens writing indicated that the data are converted into an RGB space before applying ICC. The writing may have been less then clear.
The authors of QPcard have also discussed the issue, saying the same as Esben. They can generate an ICC profile for C1 but will be useless. It is also what Esben clearly has said, you can generate an ICC profile but it will not be based on the RGB that C1 operates on, so it will be useless.
Best regards
Erik
Esben escaped w/o answering the questions fully (Eric Chan to his /Adobe's/ credit gives more clear answers)... and define "raw conversion" first... do you define that as (A) something that ends with demosaicking or (B) something that ends w/ color transform from "scene referred" camera/sensor coordinates into some colorimetric color space (like cieXYZ/D50 or whatever) ?
-
Hi,
I don't think so. I am pretty sure that it is about pleasantness. Let's assume that we have a Delta E of say 7. It does say that the Lab coordinates are 7 units of, but the deviation could be bluish, pinkish, reddish, yellowish, reddish or so on. It tells about the amount but not about the direction.
What I have seen a bit is that skin that C1 rendition is a bit yellowish while LR 5 may go more in the direction of pinkish.
A photographer I have great respect for is Tim Parkin, he did not compare LR5 and Capture One, but he looked a lot at colour rendition of cameras. The worst offender he has found is the P45+ and the best one he knows is the Alpha 900, two cameras I also happen to have.
He compared something called SMI (Sensor Metamerism Index) which measures the ability of the sensor to achieve a metameric match for 8 different colours, of course taken from the ColourChecker. He has found a very strong correlation between SMI and his perception of colour quality.
Sony A900 87
Sony NEX7 85
Canon 5D 84
Nikon D5000 83
Nikon D700 83
Samsung GX20 82
Nikon D90 82
Panasonic G3 81
Panasonic GX1 81
Nikon V1 81
Phase IQ180 80
Phase P40 80
Canon 5dMkII 80
Olympus E5 80
Panasonic GH1 79
Nikon D3S 79
Nikon D3X 79
Hassleblad H50 78
Canon 7D 78
Panasonic GH2 77
Fuji X100 77
Phase P65+ 76
Fuji X10 76
Leica M9 76
Canon G11 76
Panasonic LX5 75
Hassleblad H39 75
Aptus Leaf 75
Samsung EX1 74
Phase P45 72
As you can see, the Sony Alpha 900 is on the top of the list and the Phase P45 on the bottom. What Tim says is that all the cameras he has access to fit well into this list. Now, this is about sensors and not raw converters, but it is notable that the MF cameras are not near the top of the list. It may also put the issue a little bit in perspective.
My guess is that MFDBs and raw converters are a bit optimised for skin tones and rendering that is perceived to be good under studio lighting, while cameras with high SMI and perhaps Adobe raw processing pipeline are optimised to give accurate, but possibly less pleasant rendition.
The most important factor is probably white balance, 99% of my shots don't include a WB card, and it is very hard to compare colour rendition without a good white reference. The images I used for evaluation included a ColourChecker which I used for WB.
These images show rendering differences between tools and profiles in landscape: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition?start=4
And this is a portrait in natural light: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition?start=5
In the portraits I added a small patch on my forehead using a sample of the hue of my skin, and I also added a small patch on the shirt. The shirt is a bit outside Adobe RGB, incidentally. The skin sample was taken in the winter and the portrait in the summer, but skin colour is said to be quite constant.
What it shows, I think, is that Adobe Standard is not really nice.
My first test shot with the P45+ resulted in somewhat blush red on the flower I used in the test, although the colours on the colour checker were quite accurate. Generating a DCP profile helped a lot. The accuracy of the ColorChecker card rendition was not improved but my rd flower was rendered red and not bluish red.
Best regards
Erik
Best regards
Erik
Maybe it is your setup of C1. C1 is widely regarded to have excellent color. The fact your Imatest variances all show a blue vector on the errors and Bart noted a D65 vs D60 difference may let you fix how C1 operates on your system if you are interested. Just the fact so many people put up so many good images using C1 should tell you it is not the software. If C1 has a weakness relative to the other top raw converters it is in an artificial look to fine detail, sometimes even mazing. It is not on color.
-
Hi,
Esbens writing indicated that the data are converted into an RGB space before applying ICC. The writing may have been less then clear.
well it is not clear if he meant demosaicked into scened referred RGB or something else ;D
The authors of QPcard have also discussed the issue, saying the same as Esben.
what are you referring to ? to http://www.qpcard.com/en_b2c/dcp_icc_profile ? they are saying
"...Both dcp and icc have a "translation mechanism" that should transform an image with colors represented by digital values from a [scene referred] white balance and [scene referred] intensities into a standardized [image referred] format... They can both do this with a straight, linear matrix type of transform - where the input image as represented by the channels [c1, c2, c3] and the output result is either the base color format CIE XYZ or a standardized RGB profile like Adobe RGB...."
and P1's ICC files indeed have the data guiding the color transform from scene-referred demosaicked data (and that is not in any colorimetric color space) into a proper colorimetric space...
They can generate an ICC profile for C1 but will be useless.
just because they don't generate it in a manner that C1 expects... nothing more implied there
It is also what Esben clearly has said, you can generate an ICC profile but it will not be based on the RGB that C1 operates on, so it will be useless.
I can only refer to http://www.dpreview.com/forums/post/38768757
Esben is playing smoke & mirrors game here, that's it.
-
I have seen some discussion more in depth from QPCard.
The writing below indicates that they will support Capture One in the next release.
Times are changing.
Best regards
Erik
The authors of QPcard have also discussed the issue, saying the same as Esben.
From: http://www.qpcard.com/en_b2c/applications/qpcalibration.html
Current version is v1.21b
This update: New cameras and an OSX Mavericks compatibility patch (v1.20), the latest cameras and a few bugfixes (v1.21)
Icc profiling for Capture One and Hasselblad Phocus is now in last stage beta, and will be published with the next update!
Raw Converter Compatibility: This application generates profiles for Adobe Photoshop ACR and Adobe Lightroom. Icc profiles usable in linear edit raw converters can also be generated with the ICC plugin. ICC compatibility does not yet include Hasselblad Phocus or Capture One, but the necessary adjustments are currently being reviewed for release.
http://www.qpcard.com/en_b2c/applications/qpcalibration-icc-profile.html
NOTE! Capture One and Hasselblad Phocus both require that you make this profile from tiff files preprocessed in the programs. Since we do not yet support tiff reading we can not, at this stage, support these applications.
Best regards
Erik