[HDRI] Re: RGB values out of range

[Axel Jacobs]

Hi Raquel,

>> I think there is a misunderstanding here. I am sure that the
>> documentation you refer to is related to the source images. These are
>> 8-bit coded, and allow values from 0-255. This means that below 20 and
>> over 200 you get into the edges of this range. As the jpg's you get from
>> a digital camera are not linear mappings from luminance to pixel values,
>> and a slight change in these edge pixel values is related to a huge
>> change in the luminance of the recorded scene, it gets impossible to
>> reconstruct luminance values from such values. Have a look at the
>> s-shaped response curve of your camera, it should explain this problem.
>>
>> So hdr tools can use only this "middle" range, and this means that you
>> must make sure that the ranges of your overlap have a sufficient
>> overlap. Also, of course, the darkest (shortest exposure time) image
>> should not contain any very bright pixels, and the brightest one
>> (longest exposure time) no very dark ones. So if e.g. your brightest
>> image was taken with 1/2sec and has significant pixel values below 20,
>> take one more with 1sec time.

Lars is absolutely correct in what he is saying. WebHDR is mostly 
concerned with using HDR for (reasonably) reliable luminance 
measurements, so it's important that the entire dynamic range of the 
scene is captured in the photos. When you tone-map a scene with very 
bright lights (electric lighting fittings), then in theory you should be 
able to see details of those light sources: filaments in tungsten lamps, 
the uneven luminance of diffusers, mirror images of the lamps on the 
reflector etc. If, however, on the darkest image (shortest exposure), 
some or all parts of the bright sources are either over-exposed (they 
have a value of 255 in the JPEG), or even if they are above 200, then 
the HDR engine can't reliably work out the true luminance and will 
simply cut it off.

In reality, the HDR stitcher will even produce an HDR image from JPEGs 
that have only slightly different exposures. It'll work, but will be 
useless for our particular application. If you use HDR for anything 
other than luminance measurement, and find that the results are visually 
pleasing, you may simply ignore this 20/200 rule.

For a while, WebHDR did actually display the dynamic range of the HDR 
image (the more the better: I was even thinking about implementing an 
all-time highscore list!), but then decided to remove it. The lowest 
luminance values in an image (below desks, in dark corners) are too much 
affected by camera noise, so it didn't really say a lot about the 
quality of the HDR. So in a sense, the 200-rule is more important than 
the 20-rule, but depending on your application, feel free to ignore both.

I actually don't remember where I got this 20/200 from, maybe Greg 
hinted at it in a message on this list, or I might have read it in the 
HDR book. I didn't make it up, though. Honestly!

Regards

Axel