In my typical geeky fashion, I'm trying to wrap my head around all of the 
ramifications of adjusting sensitivity (ISO) on my camera. Please correct the 
errors in my understanding.

In the simplest form, it is a measure of how many LSBs per photon (or tens, 
thousands or millions of photons).

If the base is 10 photons, per LSB at ISO 100, 1,000 photons hitting a sensor 
site gives the value 100, at ISO 200, it'll give 200, and at 6,400 1,000 
photons will cause the raw fill to read 6,400 for that sensor site.

In a similar vein, if we have 14 bits of data, and at base ISO (100), 14 stops 
of dynamic range. That gives us a maximum data value of 16,384 which means at 
the above sensitivity, 163,840 photons will cause the sensor to read maximum 
value, and any more than that cause data to be clipped.
If we then increase the gain by a factor of two, then at ISO 200, we are 
expressing 13 stops of dynamic range in 14 bits of data, 5 photons will cause 
an LSB worth of change in the data, but it will only take 8,192 photons to clip 
in the data path.  

Since the data width is constant, every time we double the sensitivity, we 
force ourselves to only use the lower, noisier, half of the signal.  The 
benefit of this is that we can read that narrow band of the the sensor with 
more resolution so at ISO 12,800, rather than only having 7 bits of data for 
the lower 7 stops of dynamic range, we have 14 bits of data to work with. This 
is why using high ISOs give us more contrast.

It seems to me that if we are shooting a low contrast situation, such as clouds 
on a grey sky, or with a mediocre, low contrast lens, we could compensate by 
using a higher ISO to spread the fewer stops of dynamic range in the input out 
over more bits of data, at the cost of more noise, because we're constraining 
ourselves to the lower, noisier portion of the signal.

Is this basically accurate?

--
Larry Colen [email protected] sent from i4est





-- 
PDML Pentax-Discuss Mail List
[email protected]
http://pdml.net/mailman/listinfo/pdml_pdml.net
to UNSUBSCRIBE from the PDML, please visit the link directly above and follow 
the directions.

Reply via email to