[Radiance-general] HDRI - Camera Response Curve

[Greg Ward]

Hi Kirk,

You've jammed a lot of questions in here, so let me try to take them  
one or two at a time....

> From: Kirk Thibault <kthibault at biomechanicsinc.com>
> Date: January 10, 2006 9:36:39 AM PST
>
> Makes sense.  I was curious about this simply because I wondered  
> how robust the Camera Response Curves are in accounting for camera- 
> based pre-processing.  Some cameras do things to the RAW data to  
> enhance contrast, boost saturation etc. and then write the JPEG.   
> Some cameras allow the user to define the amount of this processing  
> (I'm thinking of the Canon Digital Rebel that I shoot with, but I  
> imagine that many cameras now offer user control of these  
> parameters in some form).  Do the camera response curves  
> essentially account for these settings inasmuch as they affect the  
> way the camera "exposes" the RAW image data from the sensor  
> (referring to your statement below about second-guessing what the  
> camera is doing)?

Essentially correct.  So long as you stick to the same settings for  
your exposure sequences, and the camera is not automatically  
adjusting the tone curve, you should get reproducible results out of  
Photosphere/hdrgen.  Typically, the aperature-preferred or manual  
mode you will use for exposure bracketing precludes automatic tone- 
curves.

> I know that generating HDR images from multi-exposure LDR images  
> "demands" that auto-white balance not be used - are there similar  
> suggestions for the camera-based pre-processing parameters like  
> contrast, saturation, etc. that may affect HDR generation or is  
> that essentially what the curves compensate for (assuming that the  
> same exact pre-processing camera parameters are applied to each  
> image)?

Consistency is the main thing.  After that, you are better off  
without aggressive color and tone enhancement, and sharpening.   
Boosting color saturation (as many cameras like to do) intermingles  
the color channels in nasty, non-invertible ways.  Boosted contrast  
limits the dynamic range of each exposure.  Sharpening makes image  
alignment errors more noticeable.

> Because these pre-processing algorithms are automated, would they  
> necessarily be applied consistently across the entire range of  
> exposures or could they introduce some sort of changing response  
> similar to a changing white balance that might affect the  
> combination of the LDRs and the extraction of a response curve?

As I said above, the aperture-preferred and manual modes of most  
cameras disables per-image tone dickering.

> I suppose I could experiment by setting the parameters on my  
> Digital Rebel to combinations of extrema in a relatively controlled  
> lighting situation and see if it matters, but I'm not sure I would  
> know what to look for to measure any differences that might occur.   
> Clearly my limited understanding of the physics is being exposed  
> here (my limited understanding has a very large apparent dynamic  
> range) - which may mean that in my dimwitted approach, it may not  
> matter anyway!

I spent the better part of a day last week taking shots of a Macbeth  
ColorChecker chart in the many modes of the Canon EOS 5D I have on  
loan from LBNL, and concluded that there were small differences to be  
seen, particularly in terms of color accuracy, but for most  
applications this is not going to be a make or break consideration.   
Stick with the most neutral, faithful, unaltered mode your camera  
offers, and JPEG should work fine.

-Greg