F-Stop Blues

[Daniel Browning]

<span style="color: #888888;"]

It appears that this hasn't been discussed here yet, but I can't imagine getting a more informed interpretation anywhere else...

<span style="color: #888888;"]I've been complaining about it here since the forums first started back in February 2009 and before that I was complaining on other forums. I was first alerted to the issue in 2008 by Peter Ruevski:


<span style="color: #888888;"]http://www.pages.drexel.edu/~par24/rawhistogram/CanonRawScaling/CanonRawScaling.html


<span style="color: #888888;"]I'm sure there were more advanced users who knew what Canon was doing even before 2008.


<span style="color: #888888;"]

<span style="color: #888888;"]In a nutshell it claims that lenses with more aperture than about f/2 are useless on digital because the light comes at such angles that the sensor doesn't record it;


<span style="color: #888888;"]

<span style="color: #888888;"]Well, they don't claim that it's always useless; just that it's often less useful than you'd expect.


<span style="color: #888888;"]

<span style="color: #888888;"]but it does seem that they are on to something here


<span style="color: #888888;"]

<span style="color: #888888;"]Yes, they are. Unfortunately, there is a significant error in DxO's calculations. They seem to have forgotten about the difference between the simple (but wrong) f-number and the *real* (or "effective") f-number.


<span style="color: #888888;"]The f-number we all know and love is f/D. We generally assume that there is a one-to-one relationship between f-number and light intensity. But in reality, that is only a very good approximation -- it becomes less and less accurate as you increase the value of D for a given f. To get the effective f-number, you need to use the more accurate 0.5NA. For narrow f-numbers, it is the same as f/D, but for wide f-numbers, it differs. Using 0.5NA, you can see that an f/1.2 lens has an effective f-number of only f/1.3.


<span style="color: #888888;"]The loss from sensor angle of response is separate from and in addition to the loss from "effective" f-number, but DxO incorrectly assumes they are the one and the same.


<span style="color: #888888;"]In other words, the light losses from sensor AOR that DxO calculated are off by about 30% -- things are slightly better than they put it.


<span style="color: #888888;"]

<span style="color: #888888;"]camera manufactures hide this fact by raising the effictive ISO to offset this loss of light.


<span style="color: #888888;"]

<span style="color: #888888;"]The problem isn't really that they are hiding it. I'm fine with that, actually. The problem is that they are doing such a fantastically stupid job of hiding it. I mean, really, the method they chose is totally braindamaged. They are using digital gain, which accomplishes three things:


<span style="color: #888888;"]

• <span style="color: #888888;"]
• Clips highlights as much as a stop! &gt;
  
  
  <span style="color: #888888;"]
• Increases noise significantly in cases where it would have been otherwise possible to increase exposure.
  
  
  <span style="color: #888888;"]
• Increases quantization error (posterization, "banding", etc.)
  
  
  <span style="color: #888888;"]
• Slows down the camera with unnecessary processing steps (probably inconsequential, but still).
  
  
  <span style="color: #888888;"]

<span style="color: #888888;"]Downsides of doing it the right way, with metadata or exposure:


<span style="color: #888888;"]

• <span style="color: #888888;"]
• None.
  
  
  <span style="color: #888888;"]

<span style="color: #888888;"]So as I see it, there are four separate issues:


<span style="color: #888888;"]

• <span style="color: #888888;"]
• Sensor angle of response
  
  
  <span style="color: #888888;"]
• Compensation for difference between approximate f-number and effective f-number.
  
  
  <span style="color: #888888;"]
• Metadata compensation vs digital gain compensation.
  
  
  <span style="color: #888888;"]
• Ability of the AE meter to compensate without using stop-down metering.
  
  
  <span style="color: #888888;"]