The GMB ColorChecker coordinates in Adobe's HSL space.

From RAW conversion to image editing and printing
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Those tables contain the coordinates in the arbitrary HSL space used in the Adobe 'DNG Profile Editor'. Could be of great use for building camera profiles for ACR.

The colour-boxed patches in the 5000K table mark the equal lightness of the respective target values.
Attachments
Tungsten.jpg
That's the 2800K target coordinates
(120.05 KiB) Downloaded 8751 times
D50.jpg
That's the 5000K target coordinates
(139.05 KiB) Downloaded 8751 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Above you may see two images of GMB checker taken at about 2800K and 5000K.
The values shown over those patches are the hue and saturation values in the arbitrary 'units' used in Adobe 'DNG Profile Editor'. These values may be used as target values when adjusting camera profiles with that editor.

The 5000K values are quite standard, they are the same as those derived from the 2005 GMB data. You may also consult this page - http://www.babelcolor.com/main_level/ColorChecker.htm - for the ColorChecker coordinates data plus some quite serious analysis

The 2800K values presented above have not been borrowed from any source, they are my personal subjective translation. Those values work very good for tungsten lighting and they also allow optimal results under mixed lighting conditions, like tungsten plus fluorescent plus some daylight.
As you may notice, the target values are considerably different from the 5000K ones. They are also considerably lower in saturation compared to 5000K targets – that's necessary to avoid clipping of weakest individual channels in the saturated colours, especially in the reds, under naturally pretty much coloured low-K tungsten light; so the saturation may be added at later steps, if needed.

Some pseudo-scientific sources boldly state that RGB values (and thus the HSL too) of the GMB checker do not change with change of illuminant. The worst example may be found at this site - http://www.brucelindbloom.com/index.htm ... lator.html . Don't trust them, take a look at the ColorChecker for yourself and then trust your own eyes!

In both the 2800K and 5000K tables the colour patches marked with coordinates in white text are sampled and adjusted. The coordinates in black text are not sampled/adjusted, but must still fall pretty close to the stated values. The coordinate values in orange text are for general reference only, they must not be sampled/edited as they are derived from badly rarefied raw data in case of 12-bit raw; so their point values may deviate wildly.

The 5000K table also contains some additional relative data on the lightness coordinate of the patches in the Adobe's editor version of HSL. The patches outlined with boxes of the same colour must have the same (very close at least) values of the lightness coordinate.
That lightness coordinate is best controlled by the 'saturation' slider under the 'Color Matrices' tab in the editor. It works like this – moving that slider in either direction for one given axis colour changes the lightness of the both other basic colours in the same direction, e.g. lower red saturation lowers green and blue lightnesses and so on. Do not ever use the 'lightness' adjustment at the 'Color Tables' tab – the 12-bit raw and Adobe's 15-bit editing do not allow doing it smoothly, some bad noise/posterization are readily delivered.
Last edited by agorabasta on Sun Feb 13, 2011 11:53 pm, edited 1 time in total.
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

So the workflow is as follows -

1. Export 5000K and 2800K raw DNG GMB image files made with matrix profile (requires temporarily removing all related profiles from the registered programme folders). Better have them well WB-adjusted in ACR.
2. Open the 5000K file in 'DNG Profile Editor', check neutrality of the greys. Adjust the pure R/G/B patches hues using the controls under 'Color Matrices' tab. Adjust the relative lightnesses of matched patches using the 'saturation' controls there. Better adjust that 'saturation' towards the lower values. Readjust the hues if necessary.
3. Uncheck the 'Edit Both Color Tables Simultaneously' under the 'Options' menu.
4. Open the 2800K file. Do the 'create color table' procedure at the 'Chart' tab for '2850K only'. Then kill the slots for unneeded colour patches. Do full adjustments for the remaining colours till good match.
5. Return to the 5000K file and do the 'create color table' procedure at the 'Chart' tab for '6500K only'. Kill the slots for unneeded colour patches. Do full adjustments for the remaining colours till good match.

Checking the adjustments made to the profile must be done in another instance of the 'Editor'. Open the same DNG images there using the intermediate profile exports and check the effect of adjustments. The whole procedure may take a few hours – the adjustments are too convoluted in that twisted proggie...

One more comment on the raw profiles – making anything other than 'standard-type' is a very bad idea. Don't make any 'vivid' profiles – there's simply not enough data levels in the raw data; e.g. it's pretty normal to have no more than 100 populated raw levels of the full 4000 of them in the blue channel under low-K lighting.
Don't build-in any overly special curves as those curves would be applied at the absolute values of the raw data and hence would change with further exposure/brightness/contrast adjustments in the ACR – that's not what is normally needed, e.g. a negative exposure compensation could move your curve's highlight compression into the midtones etc. Use HSL and custom curve presets instead.
User avatar
UrsaMajor
Imperial Ambassador
Posts: 650
Joined: Sat Nov 22, 2008 6:36 am
Location: Southern California

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by UrsaMajor »

agorabasta wrote:. . .
In both the 2800K and 5000K tables the colour patches marked with coordinates in white text are sampled and adjusted. The coordinates in black text are not sampled/adjusted, but must still fall pretty close to the stated values. The coordinate values in orange text are for general reference only, they must not be sampled/edited as they are derived from badly rarefied raw data in case of 12-bit raw; so their point values may deviate wildly.
Thank you for the interesting post. I have a question abut one thing you said above and would appreciate it if you could clarify that for me.

You said " . . . derived from badly rarefied raw data . . . ". Did you mean "rarefied" or did you intend to say "refined"? If you intended to say "refined", I think I follow what you are saying, but if you intended to say "rarefied" then I am not sure that I follow you - unless you mean that there are very few data points and these points are not guaranteed to be representative of the actual information.

Thanks in advance for your help.

With best wishes,
- Tom -
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

I surely mean what I say :)

Those 'foliage' and 'dark skin' patches raw values being mapped onto the output colour space produce very few possible data points. The precision of the HSL coordinates in H&S channels for those patches may fall as low +/-10 units, and is never much better than +/-5. Those are huge steps, and if you try correcting individual points so distantly spaced, you would be also affecting the colours of a range too broad.
User avatar
UrsaMajor
Imperial Ambassador
Posts: 650
Joined: Sat Nov 22, 2008 6:36 am
Location: Southern California

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by UrsaMajor »

agorabasta wrote:I surely mean what I say :)

Those 'foliage' and 'dark skin' patches raw values being mapped onto the output colour space produce very few possible data points. The precision of the HSL coordinates in H&S channels for those patches may fall as low +/-10 units, and is never much better than +/-5. Those are huge steps, and if you try correcting individual points so distantly spaced, you would be also affecting the colours of a range too broad.
Thanks for the reply. I thought that might be what you were saying, but I do not know enough about the subject to see why there would be fewer data samples for those patches, so I was wondering if you were meaning that the data was not refined (not accurate) and therefore might give bad results.

With best wishes,
- Tom -
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

In those two patches at least one of the red and blue raw channels is reduced to about 3-bit precision, and even that is mostly noise. Thus even a 1-bit change in that channel already produces large hue deviation.
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Here are the archives with profiles exported for the following bodies -

Sony a500/550/580/700
Sony Nex5/C3/5N/7
Sony a55/33/35/65/77
Canon 1DMk4/7D/550D/600D/60D
Nikon 7000/5100/5000/300s

The recipe files used to export the profiles are included in the archives. The a500, a700, a55, Nex5, Nex5N recipes are created independently and the respective camera profiles are thus 'native'. All other cameras' profiles were all exported with the Nex5N recipe, but they apparently work pretty well even for the bodies from other manufacturers.

The 'Deep' profiles are created using the procedure described above in the thread and may be considered as standard colour profiles; I call them 'Deep' because the most realistic-looking style included in a700/850/900/77 was called 'Deep', so there is some semblance.
Attachments
SonySLR.zip
(138.64 KiB) Downloaded 623 times
SonyNEX.zip
(172.9 KiB) Downloaded 508 times
SonySLT.zip
(203.72 KiB) Downloaded 852 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Here continued with Canon and Nikon profiles.

I would really like to have some feedback on the Canon profiles performance. It seems that starting with 1DMk4 and 7D Canon colours have really improved, so please let me have your opinions.
Attachments
Canon.zip
(163.88 KiB) Downloaded 384 times
Nikon.zip
(131.42 KiB) Downloaded 365 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Here's a good example of difference a properly built profile makes, and also an example of how poor really are the Adobe's 'standard' supplied profiles. Mind that the whites/greys are at the same levels in both samples, only the lightness of coloured areas is different.
The raw file was borrowed from DPR review of Nikon D5100.
Attachments
d5100_deep.jpg
(237.71 KiB) Downloaded 7384 times
d5100_std.jpg
(236.72 KiB) Downloaded 7384 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Adding 'Deep' profiles for A900/850 exported using the A700 recipe. Colours seem about right, but please check for yourselves.
Attachments
Sony A900_850.zip
(65.68 KiB) Downloaded 414 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: The GMB ColorChecker coordinates in Adobe's HSL space.

Unread post by agorabasta »

Here's the same kind of profile and its recipe, all native for the rx100 -
Sony DSC-RX100 Deep.zip
(36.34 KiB) Downloaded 348 times
Post Reply

Who is online

Users browsing this forum: No registered users and 3 guests