Re: Speaking of CCD Accuracy…

[Kennedy McEwen <rkm@xxxxxxxxxxxxxxxxxxxx>]

In article 
<5.1.0.14.2.20020305211423.00a0ac40@postoffice.worldnet.att.net>, 
East75th <east75th@worldnet.att.net> writes

>Also I have some other questions:
>
>1.      Does CCD error have a normal distribution?  It seems that it 
>must for multi-sampling to be effective.

For multisampling to be effective the noise just needs to be significant 
and non-systematic, not a normal (gaussian) distribution - although such 
a distribution would permit infinite multisampling.  CCD noise generally 
is pretty close to gaussian and poisson distributions, but it is not 
uniform with illumination, which can be significant in choosing to scan 
negatives or slides.

>2.      How about PMT drum scanners?  Are more accurate?

More accurate in that the avalanche effect of the PMT means that they 
are much more sensitive to low light levels and have much lower noise 
ratings in this domain than CCDs, pulling out finer detail in the 
shadows.

>How about scanners such as Scitex which have cooled CCDs?  (Actually I 
>should try this.)

The main effect of cooling is a reduction of dark current, which itself 
has a noise component.  Dark current is usually removed from the scanned 
image by software and scanning of a dark calibration area (as with 
response variation, scanning a white area) however after removal in this 
way the noise from the dark current itself is still present in the image 
data.  Cooling the CCD will reduce this, and other noise sources as 
well.

Cooling a CCD is not trivial - you need to cool by several tens of 
degrees to achieve a significant effect.  A good rule of thumb is half 
the dark current for every 10deg of temperature.  To get a useful 
reduction in noise, you need to cool the CCD well below freezing - then 
you need to take serious measures to prevent condensation or snow 
forming on the front window of the CCD, obscuring the image and shorting 
out drive circuitry.  Not recommended if the scanner has not already 
been designed for that mode of operation.

>3.      From what I've learned from this forum higher CCD density 
>implies a trade-off in higher noise.  Yet clearly the LS-8000, which 
>has a res of 4000 dpi, displays less noise than the LS-2000 (2700 dpi) 
>for the scans that I've repeated.  What do you think Nikon has done to 
>achieve higher res and less noise?
>
Smaller CCD pixels certainly are subject to higher shot noise levels, on 
an almost linear scale with pixel area, however carrier shot noise is 
not the only noise present, though it usually dominates at high levels 
of illumination when the photocharge in the CCD wells is significant. At 
low levels of signal, when there is little photocharge in the CCD wells, 
the shot noise is also low but is dominated by dark current noise and 
the noise in the readout channel of the CCD.  There are many other noise 
sources, but these are the main three worth considering.  With really 
cheap devices, RO noise can dominate shot noise even when the CCD is 
near saturation.

To achieve a system with a higher density rating, such as the step from 
the LS-2000 to the LS-4000 & LS-8000 scanners, it is the RO noise which 
must be reduced, together with other noise sources in the analogue chain 
right up to the ADC.  Depending on how significant the RO noise was on 
the LS-2000 you might achieve an improved noise level right across the 
illumination scale even going to smaller pixels.
-- 
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers
-
Turn off HTML mail features. Keep quoted material short. Use accurate
subject lines. http://www.leben.com/lists for list instructions.