Originally posted by Igor123 I thought that was the intent in that test? - to underexpose the iso 100 shot and then push it to the same exposure and also get the same amount of light and noise, and then using different isos to see if the camera preprocesses the iso 100 and 800 shots differently (which it seems to do a bit?)
I'm not sure which test exactly you are talking about.
In general, yes, there are tests that compare pushed low ISO shots with high ISO shots. The idea behind such tests to examine how "ISO invariant" or "ISO less" the sensor is. It is generally a desirable property of a sensor if a pushed low ISO shots does not show more noise than a high ISO shot. Such "ISO invariant" sensors allow shooting with highlight margins (thus avoiding overexposure) and/or high dynamic range shots (which will see their shadows pushed in post-production), without paying the price of increased noise levels.
All I wanted to point out with my statement is that by pushing shots in post-production, one is increasing the effective ISO setting. In other words, you cannot underexpose a shot by three stops, push it in post by three stops and then say "
look at at all that noise and mushy detail, it was only an ISO 800 shot", since what you are looking at is essentially an ISO 6400 shot. This is similar to how cropping changes focal length. One cannot shoot at 50mm, then do an extreme crop, enlarge the crop to normal viewing size and then say "
Look at the perspective compression this 50mm delivers. By cropping, one increases the effective focal length.
So if anyone is talking about the noise in an ISO 800 (or so) shot, they need to be explicit about what kind of pushing in post-processing was necessary to achieve the exposure level shown.
Originally posted by stevebrot Sorry, but I am not following you here. Are you saying that noise depends, at least partially, on the amount of light striking the sensor and that underexposure (less light) creates noise?
Yes, except that "
partially" should be "
mostly".
As photoptimist already elaborated upon, light is stochastic in nature. Light isn't "analogue" in the sense that you can dim it down to infinitesimally small levels while maintaining a "smooth" signal. Light is "discrete" as it interacts with matter in the form of individual photons. The lower the light levels, the more that stochastic nature of light comes to the fore. At very low light levels, the photon count becomes so low that there is not a chance of any even exposures across the sensor -- even if you had perfectly even illumination -- as the photons create a random and thus very noise tapestry of isolated exposure events. Even at very high light levels light behaves in the same random manner but as the signal to noise ratio (SNR) is much higher, the noise isn't as visible.
With modern Sony sensors, the read out noise is so low that you are mostly seeing the stochastic nature of light, not the sensor-contributed noise.
The reason why high-ISO settings provoke more "photon noise" (aka "shot noise") even in well-lit scenes is because high-ISO settings imply other settings that limit light entry. You typically have high shutter speeds and/or high f-ratios. If you didn't limit light entry in some way, the high-ISO shot would be hopelessly overexposed. So in low-light scenes, you see the stochastic nature of light directly and high ISO settings (or pushing in post) just magnifies the noise. In well-lit scenes, which would have a good SNR if recorded directly, high-ISO settings artificially limit the light levels reaching the sensor due to the light-limiting shutter speed / aperture settings, thus causing bad SNR, i.e., visible "photon noise".