It's all about photon counting statistics a.k.a. Poisson statistics a.k.a. shot noise: http://en.wikipedia.org/wiki/Shot_noise
If we ignore the mysteriuos details of de-Bayering (let's pretend all cameras are like the Leica M Monochrom), and that we are in a situation where photon counting statistics are the dominant source of noise (which is what we should be talking about, since we are concerned with the fundamental question of noise in more vs. fewer pixels, not other noise sources that will vary from one sensor design to another), then all that matters is how many photons end up each pixel of the final output image. Consider this simple case: you want to order an 8 x 12 print from Mpix, which they will print at 250 dpi, for a final output image with 6 MP, and we don't do any noise reduction. If you take the image with a 6 MP sensor (kind of like a K100D Monochrom, but with a modern sensor), each sensor pixel/photosite will translate directly to an output pixel, so input or sensor image noise = final image noise. If you take it with a 24 MP sensor (K-3 Monochrom), each photosite will on average get 1/4 as many photons as the K100D's photosites. Poisson statistics tell us that the noise goes as the square root of the number of photons, so each of these pixels will have a signal-to-noise ratio (SNR) that's only half of the SNR of the K-3 pixels. But when you average together groups of 4 pixels from the K-3, the SNR of the aggregated pixels will increase by the square root of 4, which is 2. 1/2 * 2 = 1; like I said it all comes out in the wash. On Tue, Aug 5, 2014 at 5:38 PM, steve harley <[email protected]> wrote: > on 2014-08-05 13:50 Bryan Jacoby wrote >> >> I think this idea of bigger/fewer pixels leading directly, as in >> through the very basic physics of photon noise, to lower noise is >> wrong-headed.I couldn't care less what the signal-to-noise ratio of > >> _pixels in my sensor_ is. What I care about is the SNR of pixels in >> the output image, whether that be an image displayed on a screen or >> the dots made by a printer. > > > i have pondered this too, and i suppose the question is whether one could > average the pixels on a 24 Mp sensor to get as clean a 12 Mp image as from a > 12 Mp sensor; i suspect there are multiple factors beyond the number of > photons hitting a photosite that make the relationship non-linear (so that > lower Mp would net lower noise even after averaging) > > but since in general we'd expect the 24 Mp sensor, in bright enough light, > to capture much more detail with a comfortably low noise floor, i think we > have to choose between low-ISO detail and high-ISO SNR > > > > -- > PDML Pentax-Discuss Mail List > [email protected] > http://pdml.net/mailman/listinfo/pdml_pdml.net > to UNSUBSCRIBE from the PDML, please visit the link directly above and > follow the directions. -- PDML Pentax-Discuss Mail List [email protected] http://pdml.net/mailman/listinfo/pdml_pdml.net to UNSUBSCRIBE from the PDML, please visit the link directly above and follow the directions.
