Dieter wrote:
How do you change the sensitivity of the sensor? I'd think it
was fixed. As far as I can tell my camera doesn't allow changing
the sensitivity.
That is what happens when you set the ISO on the camera.
I would think that the sensitivity of the sensor would be constant.
But then I know close to nothing about the sensors. Is there some
bias Voltage (or something) that changes the sensitivity?
The saturation sensitivity of the sensor is constant (the White level)
-- a given parameter of the sensor which is why you can't reduce the ISO
setting beyond a certain value (but why don't they make less sensitive
sensors?). However, since there is no Black level (it goes to minus
infinity) you can amplify the output to increase the ISO limited only by
the noise. So you have the base ISO for the chip times the factor used
to get the white in the scene to white in the ADC output and that gives
you the shooting ISO.
Sounds like they aren't really changing the sensitivity of the sensor,
just changing which portion of the available range they are recording.
Less sensitive sensors? Then you hurt the "available darkness"
capabilities. Reducing the light is easy, stop the lens down,
use a neutral density filter, ...
In theory a less sensitive sensor would have less noise. Since a camera
with a smaller sensor (<= 18x24mm) shouldn't stop the lens down past
f/11 and and using a ND filter has no great advantage, it would be
better for daylight shooting (less noise [grain]) to have ISO 25 to 50
as the lowest, if this could cut the noise in half. With a full frame
35 (which are currently in the $5K price range) ISO 100 is OK -- larger
pixels = less noise.
The Fuji sensor has pixels of two different sensitivities arranged in a
pattern so that they are all Green to increase the dynamic range.
Actually, a Kodak app note I read suggested that White should be
considered to be 106% reflectance rather than just 100% (ISO is actually
based on 18% Gray).
106% reflectance? Is this for parabolic mirrors?
Specular reflections reflect over 100% of the ambient light level.
Actually not. If there were no noise, this would be infinite. But, we
have noise.
Figure out how far into the noise we want to go, then design a format
that can handle the range.
It is a S/N ratio issue. When you choose a White point, then you can
only go so far into Black before the shadow detail gets lost in the
noise. Not only the image sensor noise but the noise in the buffer
amplifier. I doubt that you could get over 20 bits in a DAC in a
portable camera. Note that the image noise is a percentage of the
exposure while thermal noise is a constant. The result is that the S/N
ratio goes down as you increase the ISO setting.
Then there is read out noise which appears to be the noise in the buffer
amplifier. This noise is not stationary -- it is a function of frequency.
With a CCD, it is possible that (at least part of) the noise is close to
stationary stochastic noise and that much of it could be removed by
homomorphic deconvolution (no I can't explain how that works unless I
can find my DSP text to consult but I have seen and heard it work on
audio). Unfortunately, semiconductor noise is not stochastic -- it is
auto-correlation and is frequency dependent, making it much harder to
deal with.
--
JRT
_______________________________________________
Open-hardware mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-hardware
