Hi everyone,
In my research of camera profiling, I discovered a rather puzzling behaviour with the Adobe DNG Profile Editor (I'm using 1.0.0.45). Before I proceed to describe this phenomenon, I would first like to acknowledge Eric Chan's and Sandy McGuffog's invaluable help in this journey. I am sure that I can speak for many of us, who are indebted to Eric for the monumental work he has contributed to improve the tools we use daily for creating better images. Eric also wrote the DNG PE program that I speak of, and although I will quote him a number of times in this post, it is with all due respect to Eric, and I only wish to clarify his statements and my own findings. I am not accusing Eric of anything - alas, we are all prone to mistakes, and it is my hope that I grossly misunderstood how all this works and that perhaps Eric can step in to clarify matters.
This issue I speak of concerns the illuminant used by the DNG PE. As seen in the screenshot attached, the PE allows one to build color table(s) for a custom camera profile in three different ways: "Both color tables", "2850 K only" and "6500 K only".
Adobe's documentation states that to create a dual illuminant profile, one would have to take two photographs of the color checker, one taken under a 6500K illuminant and one under a 2850 K illuminant. One builds the dual illuminant profile in the PE by first selecting the 2850 K photograph and "2850 K only" color table, then the 6500K photograph and "6500 K only" color table or vice versa. The order of creating the tables do not affect the result.
Choosing "Both color tables" in this situation is decidedly wrong! In fact, I am not exactly sure when the option of "Both color tables" would be useful...
To simplify matters, I decided to test the difference in the result of creating a camera profile using "Both color tables" against "2850 K only" followed by "6500 K only" to achieve theoratically identical profiles. I used the same photograph of the colorchecker (taken in noon daylight, sufficient exposure of the capture), and left the control "end points" in the PE in exactly the same position for all the profiles. After generating and exporting each profile, I went to Edit>clear All Adjustments to be absolutely sure that I started from a clean slate again to build the next profile. In total, four profiles were built:
1. Both color tables
2. 2850 K and the 6500 K color tables
3. Only the 2850 K color table
4. Only the 6500 K color table
From my understanding of how this works, 1. and 2. are dual illuminant profiles, while 3. and 4. are single illuminant profiles. I used
Sandy's dcpTool to decompile the 4 camera profiles to xml format. I am making the decompiled versions of the profiles available
here.
Observations:3. profile has a HueSatDelta1 table that correcponds exactly to the HueSatDelta1 table of the 2. profile. 4. profile has a HueSatDelta2 table that corresponds exactly to the HueSatDelta2 table of the 2. profile. 3. profile contains a HueSatDelta2 table, but it's values are null. 4. profile contains a HueSatDelta1 table, but it's values are null. This is expected so all is good.
My expectation was for 1. and 2. to be identical. However, they were not. Applying profiles 1. and 2. in Adobe Camera Raw 7.1 (processed to Prophoto RGB and 16 bits/color channel) to the same image of the colorchecker used to build these profiles, I noted a differences in the color values of the colorchecker. None of the neutral patches shifted in hue or saturation, but all the 18 color patches did (verified in 16 bit precision). In 8 bit rounded integer values, a maximum of 2 values difference was observed in the blue channel for some patches, while the red and green channels mostly differed by one value. Decompiling the profiles, 1. profile has HueSatDelta 1 & 2 tables different from the 2. profile. That explains the difference in application. This is unexpected behavior to me.
I initially put this down to the illuminant assumed for the calculations to generate the HueSatDelta tables, and chromatic adaptation to D50 if necessary. So I tried to verify this thinking against Eric's comments on how the DNG PE works:
http://www.luminous-landscape.com/forum/index.php?topic=59688.msg482472#msg482472"Yes, that's right. If you build a single-illuminant profile in DNG PE, the target values used by DNG PE are from a set of averaged ColorChecker charts with adopted illuminant of D50. If you build a dual-illuminant profile in DNG PE, the illuminants used by DNG PE are A and D65."http://www.luminous-landscape.com/forum/index.php?topic=70762.msg561604#msg561604"For the single-illuminant profile, DNG PE assumes a D50 illuminant (which is what the normal published numbers assume, too), and for the dual-illuminant profile, DNG PE assumes Standard Illuminant A for the first table, and CIE D65 for the second table (same calibration illuminants that Adobe uses for its Adobe Standard profile)."But I got confused here:
http://www.luminous-landscape.com/forum/index.php?topic=70762.msg562781#msg562781"Yes, if you do the 6500k table only, it's D65. If you do the 2850K table only, it's Std illuminant A. Regardless of the order you choose to do them in individually, the mapping is always the same."Questions:1. I am confused as to what is really happening. Is D50 assumed for the "Both color tables" option and Std illuminant A for the "2850 K only", D65 for the "6500 K only" options?
2. Also, which resulting profile is more "accurate" in theory? Say if I was trying to create a profile for use under daylight (single illuminant), would I have a more acurate profile by creating a 6500K only profile, or a 6500K and 2850K profile from the same capture of the colorchecker, or a "Both color tables" profile?