[filmscanners] Re: Newish Digital Tech

["David J. Littleboy" <davidjl@gol.com>]

"Paul D. DeRocco" <pderocco@ix.netcom.com> writes:

>>>>>>>>>>>>>>
> To make your argument
> fly, you have to show that in reusing the smaller number of measurements,
> Bayer increases the noise. I don't think it does...

Oh, but it does, because correlated signals add linearly (i.e., in the
voltage domain) while uncorrelated signals add root-mean-square (i.e., in
the power domain).

For instance, consider a simple Bayer decoding algorithm that computes each
missing green value as the average of the four adjacent real green values.
(Real algorithms are more complex, but no amount of complexity will escape
the fundamental limitation of information theory.) This means that noise
from each green sensor is duplicated and divided by four into the adjacent
pixel locations that don't have sensors. If at a particular point in time, a
green sensor is generating, say, one millivolt of noise on top of the signal
itself. This will result in 0.25mV of noise added to each of the four
adjacent pixel locatons, so the total amount of noise added to the image is
2mV, spread out over five pixels.
<<<<<<<<<<<<<<

Huh? The noise added to a given green pixel is 4*(1/4) which used to be 1...

But there's also the problem that the result has a different distribution
than the original noise; the noise may add or cancel; averaging four values
usually _reduces_ the noise, in rms terms, as you point out below.

>>>>>>>>>>>
 In a Foveon chip, the four adjacent pixels
have their own sensors which generate their own noise, but it will be
uncorrelated, meaning that it may be positive or it may be negative. When
you add the noise of one pixel to one quarter of the noise from the adjacent
four pixels, instead of getting 2mV, you get 1.414mV on the average, because
of the way two Gaussian bell curves add together.
<<<<<<<<<<<<<

In the Foveon, you never add adjacent pixels, so the above applies to the
Bayer, not the Foveon. Since the noise is uncorrelated, it's there as is in
all its glory in the final image.

In any case, to actually reduce measurement noise, you'd have to measure
more photons, and the Foveon doesn't do that.

The bottom line, though, is that the technologies are different, and how
they actually compare can't be second guessed. If they cough up a
high-sensitivity, low noise sensor, more power to them. But before they do,
it's just hot air.

David J. Littleboy
davidjl@gol.com
Tokyo, Japan

----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe 
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or 
body