alundeb wrote:
Estimate of the photon full well (saturation) of the 7D
The background of the still life scene used by image-resource.com has some areas that are uniformly lit, and clean.
For high ISO\'s (6400-12800), the photon shot noise is practically all that contributes to the Standard Deviation (or variance, or Sigma) in those areas.
We can then use the mean-variance method to calculate the system gain.
The system gain is the number of photons collected per ADU (A/D conversion unit)
The mean-variance method utilizes the fact that photon shot noise is the square root of the number of collected photons.
The patch I used, had a mean value of 1652 ADU (Black level subtracted)
Std.dev at ISO 12800 was 288 ADU
The system gain at ISO12800 then becomes 1652/(288*288) = 0.02
Converting to ISO 100, the system gain is estimated to 2.56
If the data set is contaminated with other contributions to Std.dev, like a non-uniform surface or other noises, the estimated system gain will be lower than the actual. So my method should be conservative, regarding optimism about performance.
I got from the ISO 100 histogram that the saturation level is 11532 ADU (Black level subtracted)
The maximum number of photons a 7D pixel can collect, will be 11532 * 2.56 = 29500
For comparison, the 50D does 27300
The \"betterness\" per pixel is 1.08
For the whole sensor, 1.3 times more photons can be collected.
Take this with a grain of salt, it is only an estimate, and I apologize for the tech language.
let\'s be optimistic and say there was some texturing and call it 1.4x
well the APS-H surface area is 1.56x larger (I think) so that is almost giving everyone their desire for an APS-H sensor then plus this one takes EF-S and has normal walk-around zooms made for it And since the 1D2 photon collection stunk this should be better for both DR and SNR than the 1D2 APS-H and even the 1D3 was not quite as efficient per area as the 50D I don\'t beleive so let us now bump the 1.4x to 1.45x and compare to 1.56, pretty close
of course this is optimism placed on top of rough calculations....
and again I did take some optimistic liberties above!
(anyway my initial crazy estimates from looking at the jpgs that it might go half to a FF in terms of SNR appear to have beem a bit much; it didn\'t really make sense that it could do that much better really, just caught up in teh initial release enthusiasm)
alundeb wrote:
Estimate of the photon full well (saturation) of the 7D
The background of the still life scene used by image-resource.com has some areas that are uniformly lit, and clean.
For high ISO\'s (6400-12800), the photon shot noise is practically all that contributes to the Standard Deviation (or variance, or Sigma) in those areas.
We can then use the mean-variance method to calculate the system gain.
The system gain is the number of photons collected per ADU (A/D conversion unit)
The mean-variance method utilizes the fact that photon shot noise is the square root of the number of collected photons.
The patch I used, had a mean value of 1652 ADU (Black level subtracted)
Std.dev at ISO 12800 was 288 ADU
The system gain at ISO12800 then becomes 1652/(288*288) = 0.02
Converting to ISO 100, the system gain is estimated to 2.56
If the data set is contaminated with other contributions to Std.dev, like a non-uniform surface or other noises, the estimated system gain will be lower than the actual. So my method should be conservative, regarding optimism about performance.
I got from the ISO 100 histogram that the saturation level is 11532 ADU (Black level subtracted)
The maximum number of photons a 7D pixel can collect, will be 11532 * 2.56 = 29500
For comparison, the 50D does 27300
The \"betterness\" per pixel is 1.08
For the whole sensor, 1.3 times more photons can be collected.
Take this with a grain of salt, it is only an estimate, and I apologize for the tech language.
let\'s be optimistic and say there was some texturing and call it 1.4x
well the APS-H surface area is 1.56x larger (I think) so that is almost giving everyone their desire for an APS-H sensor then plus this one takes EF-S and has normal walk-around zooms made for it And since the 1D2 photon collection stunk this should be better for both DR and SNR than the 1D2 APS-H and even the 1D3 was not quite as efficient per area as the 50D I don\'t beleive so let us now bump the 1.4x to 1.45x and compare to 1.56, pretty close
of course this is optimism placed on top of rough calculations....
and again I did take some optimistic liberties above!
(anyway my initial crazy estimates from looking at the jpgs that it might go half to a FF in terms of SNR appear to have beem a bit much; it didn\'t really make sense that it could do that much better really, just caught up in teh initial release enthusiasm)
Sep 03, 2009 at 12:28 AM
Previous versions of skibum5's message #7488364 « 7D noise vs 50D numbers »
