Digital BW, The Print

Steve Kale wrote:


> that once one have
> determined the "size of the well" the ultimately delivered dynamic range is
> still affected by bit depth whereas you would say this is not true and that
> increasing bit depth is of no value.

I say that increasing bit depth beyond N bits is of no value, *unless* 
the sensor dynamic range is so great that N bits isn't a sufficiently 
fine sample of that range.

I say that there is no *inherent* advantage to adding bits beyond a 
certain number, unless other sensor attributes are *also* modified so as 
to take advantage of it.




Some thought experiments that might illuminate my thinking on this:

If, like Paul DeRocco reports, you can drop bits from your raw file and 
not see the values change, you are clearly outputting more bits than are 
significant. Does adding bits to this camera increase dynamic range? No; 
it only increases the number of bits you can drop without ruining the image.

If you have a perfect, noise-free camera that samples a ten stop DR in 
12 bits, and you increase that to sampling 16 bits, does this improve 
dynamic range? No. You get some sweet (high sample resolution) sampling 
in the shadows, but your sensor still only senses 10 stops.

If you have a perfect, noise-free camera that samples a ten stop DR in 
12 bits, and you increase that to 16 bits (max 65,535 ADUs) BUT white is 
still 4096 ADU's, does that increase dynamic range? No - it just gives 
you four bits you can truncate from your raw file without ruining the 
image (we're back to Paul DeRocco's camera again).

The number of electrons (stored in the photosite well) per 12-bit ADU in 
many cameras is very small - perhaps as few as a dozen electrons per ADU 
in some cameras. So if you add one bit, you will then be sampling six 
electrons per ADU (you don't magically get more electrons to sample, 
just because you've added one bit of resolution to the ADCs - the number 
of electrons in the well stays the same, and you are just sampling them 
with more dynamic resolution). Add one more bit, and you are sampling 
three electrons per ADU. One more bit, and you are sampling 1.5 
electrons per ADU. One more bit, and you are sampling 0.75 electrons per 
ADU.

At what point does adding bits gain you nothing? Even in a noise-free 
camera, sampling 0.75 of an electron is going to be a neat trick, right?


> I am interested in hearing a more direct retort to the
> material on Normen's site  ...

I don't think I have one for you, mostly for the reason you identify. 
Norman appears to me to be discussing what has to happen to the image at 
and after digital conversion in order to make that image look natural or 
good. I don't find error in anything he writes, but I'm not expert in 
that field.

I'm discussing the analog attributes a sensor must have in order to be 
physically capable of sensing a scene of high dynamic range without 
clipping. I have not made comments that are relevant once things have 
gone through the analog-to-digital converter.

That's why I say that Norman and I are discussing apples and duckbilled 
platypuses.

--
Jeff Medkeff
Eagle River, Alaska