[Leica] IMG: Roofers at 2500 ISO

[hlritter at bex.net (Howard Ritter)]

Your point seems well taken, but that process would be in addition to what I 
was talking about.

I believe that when the shutter button is pushed, what happens first, and 
without noticeable pause, is that the sensor is flushed of all thermal 
electrons (noise) accumulated since the last readout. Then during the 
exposure, both thermal electrons and the electrons freed by image photons 
accumulate at each photosite. Finally, when the shutter closes, the 
electrons are read out by the electronics and the corresponding numbers 
stored as the image file. This file contains contributions from thermal 
electrons (and other sources of noise, such as read noise). The longer the 
exposure (and therefore the interval between the pre-exposure flush and the 
post-exposure read), the more thermal electrons will accumulate at each 
photosite. Part of the thermal noise is random from one photosite to the 
next and so is not unambiguously removable, though statistical and 
contextual algorithms can be applied to make the picture look less noisy 
(Noise Ninja etc). 

So longer exposures produce more noise, and this is invariant with the 
number of "image" electrons that accumulate (unless a photosite saturates). 
If two images are made, one with a short exposure and low f/ratio and the 
other with long exposure and high f/ratio, both accumulating the same number 
of image electrons, you can see why the image made with a long exposure will 
contain more noise electrons than the short exposure. What I don't know is 
whether this difference is noticeable under usual working conditions, i.e. 
whether 1/60 or 1/30-sec exposures accumulate noticeably more noise than 
1/1000 sec in an image that is adequately exposed?in the middle of the 
histogram. Certainly it's important as exposures get longer. I suspect it's 
noticeable in dark areas of subject matter in ordinary exposures at high 
ISOs. (Higher ISOs are obtained simply by amplifying the signal being read 
off the CCD, and some of the noise inherent in the process gets amplified 
too, so the image is noisier than one taken at low ISO?though not all of the 
noise, otherwise what's the point?)

The random part of the thermal noise cannot be removed by the automatic 
dark-frame subtraction that digital cameras perform after a long exposure, 
because it's unpredictable and variable from one photosite to another and 
from exposure to exposure at each individual photosite. But there is another 
 component to thermal noise, one that depends on fixed, unique 
characteristics of each individual photosite, and is therefore consistent 
from one exposure to the next. These systematic aspects include almost 
atomic-level differences in the photosites acquired during manufacturing, 
differences in temperature across the sensor produced by the camera's 
electronics ("amplifier glow"), and other factors. This systematic thermal 
noise can be replicated by exposing a dark frame of the same duration and 
subtracting its values pixel-to-pixel from the image file. These things are 
well known to "astroimagers", who count practically every single photon from 
extraordinarily faint deep-sky objects using CCD cameras that are cooled to 
30-60 degrees below ambient and apply dark frames, bias frames, flat-field 
frames, etc, to refine the result. Professional astronomers cool their CCDs 
for spectroscopy and imaging to the temperature of liquid nitrogen!

And as you said, selectively bringing up the raw signal from 
under-represented pixels will selectively bring up noise of that color.

I'll do some high-ISO comparison shots of short exposure at high f/ratio and 
longer exposure at low f/ratio and post them.

?howard

On Oct 31, 2010, at 5:11 AM, Jeff Moore wrote:

> 2010-10-30-22:32:57 Howard Ritter:
>> In answer to your question about the low noise level in bright-light
>> exposures at ISO 2500, here's my suspicion: The amount of noise in a
>> given exposure depends on a number of variables, one of which is the
>> duration of the exposure.
> 
> Well, maybe.  But I also note that when there's bright light, it's
> usually sunlight, while dim light is often something with a warmer
> color balance like incandescent lights (or candles).  When I
> particularly notice noise, it's often when I start with something with
> very little blue energy in it naturally, and try to balance the color
> temperature out to look not-entirely-orange.  That requires cranking
> up the underexposed blue pixels, which respond by peppering the frame
> with blue noise.
> 
> Or is it some entirely diffrent phenomenon you're talking about?
> 
> -Jeff
> 
> _______________________________________________
> Leica Users Group.
> See http://leica-users.org/mailman/listinfo/lug for more information