[Hdri] HDRI capture program for Canon EOS cameras under Mac OS X

[Gregory J. Ward]

Hi Lisa,

Nikon really does make the greatest lenses.  LBNL has this f/2.8  
Nikkor fisheye that's just amazing, but unfortunately doesn't work  
with any of the Nikon digital cameras in the sense that the 1.5  
multiplier turns the 180 degree circle into a partial view.  It's  
vignetting is quite low for a fisheye -- just -20% at the edge at f/ 
4.0.  The 8mm Sigma lens we got for the Canon is pathetic by  
comparison, with vignetting of -80% from at the edge at f/5.6, and  
the eccentricity is not very linear, either.  (See attached graphs  
for comparison.)  Unfortunately, the Nikkor->Canon adaptor ring we  
bought doesn't seem to work for a Nikkor manual lens.

Regarding RAW captures, I've run into problems with pink highlights  
showing up under some lighting conditions, and I can't seem to shake  
them.  There are also issues with noise at the bottom end, and we  
avoid this as well if we just use JPEG or 24-bit TIFF output.  The  
reason is that cameras have a lot of built-in processing that cleans  
up the sensor range and shortcomings you just don't want to deal with  
yourself.  As long as the camera doesn't mess up the tone curve, the  
on board processing generally helps the HDR merge process more than  
it hurts it.  (Automatic white balance is one of the notable  
exceptions, but luckily we can disable that.)

The potential benefit you cite for RAW images, the bit depth, turns  
out to be an illusion.  Although many RAW files contain 12 bits/ 
channel, it's in a linear space rather than the gamma response space  
of JPEG or 24-bit TIFF, and doesn't actually encompass a greater  
dynamic range.  (See my page at <http://www.anyhere.com/gward/hdrenc/ 
 >  and scroll down to "What Is a Gamma Encoding?" to more.)  Since  
the camera and the A/D converter inside the camera are both linear  
devices, this is in fact why 12 bits is necessary for an 8-bit gamma- 
encoded output.  Otherwise, you'd end up with horribly visible  
quantization errors (banding) at the bottom end.

So what happens to the extra 4 bits of resolution?  The answer is, it  
gets *wasted* at the top end, giving us much finer steps than we're  
able to see or represent in an 8-bit, gamma=2.2 encoding.  Could this  
ever be useful in an HDR merging process?  Maybe, if your output  
format could capture these finer steps, and you spaced your exposures  
close enough that you could get the top ends of all your exposures to  
cover the full range of the scene, but by that time, exposure  
averaging will give you the same benefit from standard 8-bit/channel  
JPEGs.  In other words, a RAW process to HDR is a lot of work for  
very little benefit.  You don't get additional range at the top or  
bottom; white balance ends up as more of a problem rather than less  
of one, and the loss of noise-reduction processing in the camera  
makes the deep shadows look much worse.

I spent the better part of a week playing with dcraw.c and the Canon  
EOS 5D to find all this out.  It's possible that my conclusions don't  
apply to all cameras that produce RAW output, but in some ways,  
that's another argumentagainst it -- camera RAW files are all  
different!  If you're trying to build up or recommend a standard  
practice, I think RAW is a great big unknown in the equation, and  
much harder to work out than the camera response function, which is  
as far as I can tell, the *only* effort RAW saves you.

That's my 2 cents.
-Greg

-------------- next part --------------
A non-text attachment was scrubbed...
Name: SigmaVignetting.pdf
Type: application/pdf
Size: 18584 bytes
Desc: not available
Url : http://radiance-online.org/pipermail/hdri/attachments/20060120/61b52d2b/SigmaVignetting-0001.pdf
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NikkorVignetting.pdf
Type: application/pdf
Size: 23595 bytes
Desc: not available
Url : http://radiance-online.org/pipermail/hdri/attachments/20060120/61b52d2b/NikkorVignetting-0001.pdf
-------------- next part --------------

> From: "L.Yimm" <yimm at mac.com>
> Date: January 20, 2006 5:39:59 PM PST
>
> On Friday, January 20, 2006, at 04:18PM, Gregory J. Ward  
> <gregoryjward at gmail.com> wrote:
>
>  (I found RAW captures not to be the way to go for HDR merging,
>> for various reasons.)
>
> Could you elaborate on those, Greg?  I've been capturing TIFs for  
> HDR assembly but was thinking of using RAW for a wider dynamic range.
>
>> Rob G. asked about Nikon cameras, and we did look at them originally,
>> but there were two things that led us to use the Canon 5D.  One was
>> the fact that we needed to capture full 180-degree circular fisheye
>> images, and this requires a full-frame sensor.  The second was the
>> way Nikon created a separate SDK with a different programming
>> interface for every model camera, which makes supporting more than
>> one a real pain.  It's a shame, though, because we have this really
>> awesome f/2.8 Nikkor fisheye lens, and the adaptor ring that's
>> supposed to mount Nikon lenses on Canon bodies doesn't work for this
>> particular lens.
>
> Wow...I wish I could automate capture for my Nikon.    I've been  
> using the FC-E9 spherical fisheye with my 5700 and getting some  
> nice results with a custom made pano bracket designed just for the  
> lens.   I would love to be able to automate the capture and/or  
> processing of the images...Nikon capture doesn't and won't ever  
> (according to the reps I've talked to) include fisheye unwarping  
> for this lens.  Which is a shame really...it's a brilliant piece of  
> glass.