At 2:51 PM -0500 12/14/03, Joe Gwinn wrote:
>In the January 2004 issue of the computer magazine "PC World", on page 146,
>there is an article "Serious Cameras for Serious Photos" that reviews [various
>cameras].
>
>The article can be found at <http://find.pcworld.com/38861>.
>
>"Noise Annoys. Noise--speckling or mottling, most often visible in broad
>planes of color, such as sky--is an inevitable fact of digital photography. We
>saw noise in photos taken by all of the cameras, although most of the time
>these blemishes were barely noticeable. The Nikon D100 showed the most noise,
>but we wouldn't say it has a big problem: The images it produced still looked
>great when enlarged, with the noise being only just visible at lower ISO
>settings. The Olympus E-1 has a noise reduction mode in which the camera takes
>the photo and then takes another with the same settings but with the shutter
>closed. Finally it subtracts this noise-only image from the first,
>eliminating much of the noise ."
>
>The above quote implies that the cameras are limited by dark current in the
>CCD image sensor, rather than by shot noise in the light or CCD, which is good
>news in that dark noise is not an unavoidable property of anything, and will
>be reduced as semiconductor technology progresses.
On thinking about this some more, it occurs to me that this answers the
question of how software (in camera or in photoshop) can eliminate the noise in
digital pictures, given that noise is random and therefore impossible for the
software to predict.
The answer is that dark-current noise, while different in each sensor
(different serial numbers), it's always the same in a given sensor (serial
number), so the software can very well compensate without losing the image
detail. The only cost is that the shot noise increases, which is likely a
price well worth paying.
Dark-current noise is one form of a larger class called fixed-pattern noise,
whose name is an apt description of its nature. As a class, fixed-pattern
noise can be eliminated by correlated double sampling, where one subtracts a
frame taken in darkness (shutter stays shut) from the image frame (shutter
opens); what's fixed cancels out in the pixel-by-pixel subtraction.
Dark frame subtraction is commonly done by astronomers, who also use "flat
frames" (uniform white illumination) to calibrate and compensate the light
sensitivity of each pixel in the sensor. I have not heard anything that would
make me think that digital cameras use flat frames. I do recall that I posted
the procedures some time ago, when we were discussing how to compensate for
vignetting in wide-angle lenses.
Joe Gwinn
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|