astrolucida
Offline
• • •
Upload & Sell: Off
|
Specularist wrote:
On the other hand, the low-ISO dynamic range in most CMOS cameras is limited by something other than the sensor. To me that seems like a wasted opportunity, and one that's most likely the result of cynical cost-cutting.
It would be pretty tragic if it turns out that for $50 more (or $10 more?) you can make a camera that takes full advantage of the sensor's dynamic range!
Actually, it's not that simple.
When Canon moved from 12 bit RAW files to 14 bit RAWs, the theoretical dynamic range at low ISOs increased fourfold (2 stops). Or, to put it in another way, the turning point between low and high ISOs was reduced by a factor of 4. A typical case was from ISO 1600 to ISO 400. This ISO is the "unity gain", i.e. where one count in the RAW file corresponds to one detected photon (=electron) in the sensor.
There is absolutely no need to go any higher than the unity gain ISO as you can get the equivalent result in digital postprocessing. However, if you are taking JPEGs, then the camera can do the postprocessing for you and then it makes sense.
A good example is deep sky astrophotography where every image is underexposed, i.e. more exposure will always give you more detail. Hence, in any case you need to pull as much from the exposure as you can get in the postprocessing. Thus, the optimal ISO is selected for the actual exposure. For the current Canon cameras it is in the range of ISO 200 to ISO 400, depending on the model.
However, in practice the amount of noise limits the dynamic range even more than the theoretical range defined by the number of bits. This noise consists mostly of photon noise, which is inherent in the light itself. The lower the light level and the smaller the sensels, the more there is noise. It is possible to reduce this noise by digital noise reduction algorithms but they always affect the other parts of the image as well.
As long as the sensor records signal that is above the read noise, the photon noise is the limiting form of noise, even upto exposure times of several minutes. The hot pixels and thermal dark noise can be removed by using a dark frame, which is how Canon cameras perform the long exposure noise reduction, or you can do that in postprocessing like astrophotographers do, using a master dark frame that is summed from several individual dark frames.
When you take underexposed images using a low ISO and then pull the exposure up in postprocessing, you see read noise. This is caused by the read circuitry for an individual pixel, the analog amplifiers and the analog to digital circuitry. The read noise is largest with low ISOs, so if shadow detail is important to you, increase the ISO upto the unity gain ISO. Also increasing the exposure just to the overexposure limit improves the situation.
Summary: there are several types of noise:
- shot noise, inherent in the signal, increases with low light and small sensels,
- hot pixels and thermal dark noise, can be removed with dark frames,
- read noise, a symptom of too much exposure compensation in post, can be reduced by increasing ISO and exposure time.
All kinds of noise can also be handled in post processing, but it is best to minimize it when capturing the image.
Check the thorough analysis done by Roger Clark at http://clarkvision.com/articles/index.html#part_4 to understand the last bits you can squeeze out of digital cameras. One particularily interesting page is http://clarkvision.com/articles/exposure_latitude-1/index.html as it shows that you can get slightly more than 10 stops of latitude from a 1D II when using RAW. This is in line with tests done by magazines, like the German Color Photo.
The point is that in order to get that 10 stops of dynamic range in use, you have to nail the exposure exactly. If you make an error of two stops from the perfect, you lose two stops from DR. Fortunately you very rarely need the full dynamic range in practical photography, and 6-7 stops is quite enough, giving you 3-4 stops of exposure latitude (i.e. margin or error without it being visible in the final image).
|
