🚀 go-pugleaf

In article <18e1vpINN9m2@network.ucsd.edu> spl@szechuan.uucp (Steve Lamont) writes:
>In article <jecome01.715798061@starbase.spd.louisville.edu> jecome01@rigel.spd.louisville.edu (Eric Comer) writes:
>>	if the lens caps is place on a ccd camera and an image is capture,
>>	why aren't all of the pixel values 0 (black)?  when the image is
>>	normalized it seems like there are faint light diagonal lines
>>	throughtout the whole image.  the image acquired was a gray scale
>>	image the min value was 0 and the max value was 6.  any ideas?
>
>Noise, mostly.  There may also be imperfections in the CCD -- a
>perfect chip, in fact, is fairly rare (and expensive), particularly at
>higher resolutions.  Finally, there is a charge accumulation (dark
>current) that occurs even when the shutter is closed or the lense cap
>is on.
>
>Excerpting from the Photometrics Series 200 manual, "[all] electronic
>circuitry generates noise.  Photon shot noise, dark current, and read
>noise are primary CCD camera noise sources.
>
>"Photon shot noise is a funamental property of the quantum nature of
>light...
>
>"Dark current, or thermally generated charge, cannot be distinguished
>from photon induced charge.  Dark current can be measured and
>subtracted, but its noise component cannot be ignored...
>
>"The third primary source of noise, read noise, sets the detection
>limits of a CCD camera.  Read noise is generated by the on chip output
>ampilfier...
>
>"Three types of integration measure the charge that has integrated on
>the CCD:  exposure, bias, and dark integrations. ..."
>
>There's a whole lot more to this, obviously, but that should at least
>give you some ideas on what's going on.
>
>The bottom line to all of this is that you have to be certain to
>remove the effects of the bias, dark current, and, possibly, light
>source non-linearities and artifacts when you grab an image.
>

There are ways of "beating the noise" to some degree:

Combining any noisy signal with any other noisy signal always results in
increased noise.  So you will increase the noise of the dark-subtracted
and bias-subtracted images unless you take precautions.  You improve the
precision of the dark frame by combining many darks together and then
scaling the pixel values by the ratio of image exposure time to total
summed-dark exposure time.  This scaled frame must then be subtracted
from the image of interest.  Since you've added many darks together and
the dark count/unit time should not change as long as the detector
temperature remains constant, the increased precision in the composite dark
dark frame will result in less noise per unit of dark signal being added
to the image when you subtract the dark count (the composite dark frame
has a higher signal to noise ratio, S/N).  The same principle applies to
bias as well.  In addition to the older, brute force approach of taking
bias *frames*, newer chips will be over- or under-scanned (= reading
fictituous pixels to measure the electronic bias and noise in the absense
of signal) to determine the bias to be applied to each particular row of
the image.  By combining many over/underscan pixels per row, you can beat
down the uncertainty in the bias level for each row just as you would by
combining many bias *frames* together to beat down the uncertainty of
the bias in each *pixel*.  Subtracting this higher S/N bias frame or over/
underscan vector will do less damage to the resulting bias-subtracted image.
Note that the bias must be subtracted from *every* frame before you use it
for anything, and this includes the individual dark frames I was suggesting
be combined (discussed above).  Finally, note that if your dark count
appears constant and featureless in a relatively long exposure, you are
better off subtracting a *constant* number (which represents the dark count
expected during the exposure time of the image) from the entire image as this
is an almost noiseless quantity and hence adds virtually no noise to the data.

I've written a paper which addresses these noise issues and discusses
maximizing the signal to noise ratio for data obtained with CCD's.
Although this is aimed at astronomers, it is very general in nature; just
replace references to the word "sky" with the word "background".  See
"Publications of the Astronomical Society of the Pacific", vol.103, p.122,
January 1991.

Mike Newberry
Steward Observatory
newberry@astro.as.arizona.edu