Fast lenses, and High ISO

Started Jul 18, 2014 | Discussions thread
Great Bustard Forum Pro • Posts: 42,030
Re: Understanding ISO

Austinian wrote:

Great Bustard wrote:

Austinian wrote:

mosswings wrote:

Good recasting of the subject, GB.

And the reason why sensors aren't getting rapidly better anymore.

Which is bad news for those expecting major sensor improvements in the immediate future, but good news for those fearing their existing gear will soon become obsolete.

Lowering the read noise, however, is important for increasing pixel counts. For example, the read noise at high ISOs for modern sensors is typically in the 2-3 electron range. Let's see the effect this has on various pixel counts over the same area of the photo (I like to use the µphoto, which is a millionth of a photo -- for example, 1 µphoto on a 24 MP sensor is 24 pixels) using a read noise of 2.5 electrons / pixel:

12 MP: 8.7 electrons / µphoto

24 MP: 12.2 electrons / µphoto

48 MP: 17.3 electrons / µphoto

96 MP: 24.5 electrons / µphoto

As we can see, every doubling of pixel count increases the read noise by 41%, assuming that the read noise per pixel is the same. If the read noise pixel were to be half as much for 4x as many pixels, then the read noise µphoto would remain unchanged.

So, there additional improvements in sensor tech with regards to read noise are very important with regards to increasing pixel counts.

Ah! I'd read that for identical technologies, sensor area was the most important factor, but it was implied that it wasn't the only factor. Good to know this is another.

What other significant aspects should digital camera buyers consider, and is such specific sensor information readily available?

The most important factors about the sensor that manufacturers do not tell us are:

  • QE (Quantum Efficiency -- the proportion of light falling on the sensor that is recorded)
  • Read Noise (the additional electronic noise added by the sensor and supporting hardware)
  • CFA (Color Filter Array)
  • Microlens Efficiency

Sensorgen has used DxOMark data to get the QE and read noise for a lot of sensors. However, differences in the dyes in the CFA can result in a greater QE by letting more light in with the result of less luminance noise at the expense of greater color noise (there may also be issues with metamerism, although that's more complex than merely the transmissivity of the dyes).

For example, I compared the Canon 6D at ISO 6400 to the Olympus EM5 at ISO 1600:

Here are sensorgen’s figures for the 6D and EM5:

  • 6D: QE = 50%, read noise @ ISO 6400 = 2 electrons / pixel = 9 electrons / µphoto
  • EM5: QE = 53%, read noise @ ISO 1600 = 2.6 electrons / pixel = 10.4 electrons / µphoto

That's about as identical as it gets, yet the EM5 was cleaner when they should be all but indistinguishable. Assuming that the exposures for the DPR's test scene were exactly two stops apart, I’m thinking the reason for the EM5’s cleaner appearance is due to less color noise. So, I did own conversions and processing to both photos.

As it turned out, they did end up looking all but identical in terms of noise (well, I had to lower the red gamma and overall gamma on the 6D photo a bit to get the colors nearly the same, too).

So, I don't know if that means that color noise isn't a big deal, or that the color noise in this particular example was close enough to where NR (noise reduction) resulted in no significant loss of detail.

Then there's the issue of microlens efficiency (DxOMark's F-stop blues, discussed in more detail here), which can have a decided impact on how much light falling on the sensor makes it into the pixel at wide apertures.

For astrophotographers, or any other photographers that use long exposures, thermal noise is a big deal, so that's another factor to consider which I did not list above.

Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow