Erik Kaffehr
Veteran Member
Hi,
Most of the noise in digital imaging is coming from the photon arrival statistics of light.
For a large number of photons, that will follow a probability distribution, like this one:
Calculated for 6250 photons/pixel on average.
Looking at raw data, using RawDigger, we may find these peaks for third left gray patch on color checker, measured on DPReviews Studio Scene shot on the GFX 1000
Enlarging the two green channels, we see these two different 'bell curves'. The numbers on the x axis are 'Data Numbers' the digital representation of the analog pixel voltage.
What we see is in all three diagrams is that the pixel signal from a single intensity patch is distributed in many channels around a central value.
Now, let us look at two patches, with a 1% difference in brightness.
This is a theoretical calculation, where full well capacity is 50000 e- per pixel. Exposure is 3EV below saturation, We have a lot of overlap between the bell curves. I doubt we could tell those tones apart.
Now, that we see that the photon signal varies quite a lot for a single value of brightness, I think it is obvious that it doesn't matter if we describe that bell curve with 16 or 14 bits. The bell curves stay the same.
Scaling the X-axis to 1/4, the curves stay the same.
Now, these figures apply for mid tones. In very dark parts of the image, read noise will come into play.
Best regards
Erik
--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic tends to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles
Most of the noise in digital imaging is coming from the photon arrival statistics of light.
For a large number of photons, that will follow a probability distribution, like this one:
Calculated for 6250 photons/pixel on average.
Looking at raw data, using RawDigger, we may find these peaks for third left gray patch on color checker, measured on DPReviews Studio Scene shot on the GFX 1000
Enlarging the two green channels, we see these two different 'bell curves'. The numbers on the x axis are 'Data Numbers' the digital representation of the analog pixel voltage.
What we see is in all three diagrams is that the pixel signal from a single intensity patch is distributed in many channels around a central value.
Now, let us look at two patches, with a 1% difference in brightness.
This is a theoretical calculation, where full well capacity is 50000 e- per pixel. Exposure is 3EV below saturation, We have a lot of overlap between the bell curves. I doubt we could tell those tones apart.
Now, that we see that the photon signal varies quite a lot for a single value of brightness, I think it is obvious that it doesn't matter if we describe that bell curve with 16 or 14 bits. The bell curves stay the same.
Scaling the X-axis to 1/4, the curves stay the same.
Now, these figures apply for mid tones. In very dark parts of the image, read noise will come into play.
Best regards
Erik
--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic tends to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles
