Shel Belinkoff wrote:
Over the last couple of days I was thinking about noise that's generated in
digital photo files, and was wondering if longer exposures at lower ISO
gave more or less noise than a shorter exposure at higher ISO ratings,
assuming the overall exposure is the same in both instances. It seemed
like a good idea for some testing.
Now, just a few minutes ago, I came across this comment:
I believe (he) means that he's set the camera at
ISO 400 and then (using the exposure
compensation feature) deliberately underexposed 2
stops... thus yielding the same exposure as if the
ISO had been set to 1600 to start with. Then, plus
two stops of compensation is applied during
"development" (the conversion of the RAW data)
[...]
With some digital systems [...] it tends to yield
a bit more noise (the digital equivalent of grain)
than with the (camera's) native ISO 400 setting,
but much lower noise than obtained by
using the (camera's) native ISO 1600 setting.
So [...] it's a way of increasing the quality of shots
at higher ISO's.
This does not make sense to me. Assuming a perfect amplification and a
perfect digitization for a moment, then a shot that would have a
complete dynamic range at ISO 1600 would only go up to 1/4 the dynamic
range at ISO 400. So when you amplify this quantization (for 12 bits
this would be 2^12/4 = 1024) to the full range, you have "chunkier"
quantitization, as if you only had a 10bit sensor instead of 12. That
leads me to believe that there would be more noise associated with this.
This is similar to the arguments of keeping your image in 16 bit mode
when editing as much as possible, until the final conversion to JPG and
8 bits. Converting to 8 bits first then editing is going to cost you
alot of information.
I'm not taking into account the effects of Bayer interpolation or other
interpolation such as uprezing, etc. That just complicates the way the
information is interpreted, but it does not change the absolute
underlying numbers.
Well, I'm no expert on such matters, but I tend to believe what I see, so i
did a quick test. Unfortunately, the light was changing rapidly, and it
might be better to try this when the light is more stable. However, this
first Q&D experiment seems to indicate that lower noise is observable using
this technique. But don't take my word for it, try it yourself under
stable lighting conditions, and see what results you get.
Shel
