[HDRI] calibrating HDR pictures with real world luminance

[Gregory J. Ward]

As Mehlika points out, there are slight differences in destination  
color spaces to account for, but the major one is actually in the  
camera itself.  Most manufacturers spend a fair amount of time  
"tuning" their cameras' colors to have a more pleasing, photographic  
look, without regard to color accuracy.  In general, I have found the  
Canon DSLRs to stick fairly close to the sRGB specification,  
especially if a "Neutral" color setting is used.  Nevertheless,  
working from RAW images is a reasonable way to get around most of the  
in-camera processing that undermines absolute CIE color accuracy.

I do not believe that HDRShop does anything to the camera's color  
space, other than an attempt to linearize the three channels.  The  
same is true of Photosphere, although it does attempt to get an  
absolute luminance calibration, and provides features for inputting  
your own per-camera or per-image luminance calibration factor.  If  
you use dcraw.c or Photoshop's RAW converter in "don't touch" sRGB  
calibration mode, you should get out something reasonably close to  
the sRGB primary color space, provided you have shot with a daylight  
white balance.  (See note below on WB settings.)  I have had good  
luck using dcraw myself, and found it's color transformations to be  
fairly reliable.  (See Dave Coffin's website on dcraw.c at <http:// 
cybercom.net/~dcoffin/dcraw/> for code and related links.)  Neither  
HDRShop nor Photosphere can unscramble the eggs once the colors have  
been messed around in a typical digital camera, and I suspect this is  
behind the large delta's in Mehlika's report for primary colors.

Regarding white balance, shooting in daylight mode assures that your  
measurement condition matches the sRGB color space, so you have some  
home of getting out absolute colorimetry, which naturally will  
include any coloration due to the light source.  In other words, the  
camera will measure something like what you would measure with a  
chroma meter.  If instead you apply an appropriate white balance  
setting, e.g. incandescent under tungsten lighting, the camera  
performs some sort of von Kries transform (one would hope), bringing  
neutral colors back to the D65 white point of sRGB.  The problem then  
is that you would have to know exactly what transform was applied to  
get back to absolute colors, and in general you cannot know.  For my  
work, I shoot with the appropriate white balance when what I care  
about is appearance in my HDR results, and I shoot with D65 when I'm  
going for color (and luminance) measurements.  Remember that white  
balance will affect the luminance values as well.

-Greg

> From: Blochi <Blochi at EdenFX.com>
> Date: September 27, 2006 4:52:00 AM BDT
>
> Hi,
>
> That study is a very interesting read. Thank you for the link. And  
> thank you for taking the time to do this elaborate analysis in the  
> first place.
>
> It makes me wonder how closely your Mathlab algorithm for  
> calculating the luminance is related to the Photosphere's  
> algorithm. Does the determined 10% average accuracy apply to  
> Photosphere's luminance readout as well, provided we work off a  
> good calibration curve?
>
> Also, I find it interesting that the error is higher for primary  
> colors, with Red going up to 40% sometimes. Those really seem to  
> drive up the average a lot. Could the limited gamut of the Radiance  
> format itself have an influence here? Or is it really all due to  
> the cameras internal demosaicing and JPEG compression? Could this  
> error be minimized by shooting RAW pictures instead?
>
> Regards,
> Christian