fredmiranda.com
Login

  

  Previous versions of ausemmao's message #10222001 « D4: $6,000, AF @ f/8 and ships February »

  

ausemmao
Offline
Upload & Sell: Off
Re: D4: $6,000, AF @ f/8 and ships February


alundeb wrote:
ausemmao wrote:
But as current technology stands, IQ isn\'t bound by QE, not by a long way. QE + shot noise bound only limits SNR at full well capacity, and yes, there read noise means much less. At every other luma value and especially in shadow areas, all the other sources of noise become more significant as they comprise a greater portion of the reading, until they become the same magnitude. There is absolutely a long way to go there.

This is the reason why you\'d find it hard to tell a D3 from a D3s from a typical ISO 200 image, but wouldn\'t find it hard at all at 12800.


I don\'t know your background, but it is fundamentally wrong that QE/shot noise only influences noise around full well capacity. The QE related photon shot noise is the square root of the the number of captured photons. You will be surprised by how deep into the shadows this noise dominates.

At ISO 6400, mid gray level, the 5DII sensels capture on average about 150 photons. The QE related noise (photon shot noise) is the square root of that, about 12 photons. In comparison, the read noise is equivalent to ~ 3 photons.

4 Ev below mid gray at ISO 6400, the read noise will be about as large as the photon shot noise, and an improvement will be visible. Still, if the read noise is reduced to ~1 photon, the total noise will only be reduced from ~4.2 (sqrt(3*3 + 3*3)) to ~3.2 (sqrt(3*3 + 1*1))


Aside from the 3 e- read noise being the minimum read noise, (it is not constant for any sensor, and tends to rise with ISO), at full well, read noise doesn\'t matter - 2 3 e- vanishes compared to shot noise.
Mid grey, it starts mattering, in the shadow areas it really matters. Even at ISO 3200 you\'re looking at more like 20e- for read noise on a 5D2 (so that right there is 3 stops of DR lost compared to a constant 3e-). That\'s why a D7000 looks unquestionably better than a 5D at 3200, and more so the further you go, and why the D3s looks better than any other DSLR at high ISO. It\'s also why, provided the sensor structures can be reproduced properly, increasing resolution reduces image noise.



Jan 06, 2012 at 10:29 AM





  Previous versions of ausemmao's message #10222001 « D4: $6,000, AF @ f/8 and ships February »