Why is same image exposed at 100 ISO noisier than 3200 ISO? [duplicate]

Question

I used to believe that ISO adds noise as a drawback to increasing exposure, but after researching the topic I arrived at the conclusion that it is a misconception. By amplifying the existing signal from the electron well (please correct my terminology if it's wrong), any noise that might be apparent in the final photo will exist there no matter what ISO setting might be chosen. The reason the noise is apparent at high ISOs, is that underexposed areas, which have a lower signal-to-noise ratio, get amplified. Whether this effect is achieved through ISO or through post-processing software should be irrelevant, and the end result should be the same. Meanwhile, the benefit of raising the ISO in this situation should be the greater variance in half-tones.

So I did some tests, and invariably there is actually more noise in same photos taken at 100 ISO and given +5 EV in post, vs same photo taken at 3200 ISO. Here is an example:

According to my understanding, the noise levels should be virtually the same. What is causing noise at 100 ISO? My only guesses are:

• Higher read noise at lower brightness
• Something to do with logarithmic exposure and the way the image is stored
• The camera actually captures more photons when the ISO is amplified

Any ideas? Thank you very much.

Answer 1

First you need to capture RAW to make any sort of comparison in that manner: JPEG is lossy compression based on the premise that it will not waste storage for differences that are comparatively imperceptible (applying a metric of image quality) by human vision. If you are doing image procession afterwards, this premise is violated. You say that you were using RAW but I wanted to point this out anyway. A similar consideration would be the noise reduction algorithms cameras employ at higher ISO: sometimes they are even active in RAW mode to some degree.

Second, your theory only holds for photon noise and in the digital domain. ISO values tend to be implemented in the analog domain, however. Sometimes there is some differentiation when you have intermediate ISO values: the intermediate values are sometimes implemented juggling thresholds and scales in the digital domain while the "proper" powers of 2 are done in the analog circuitry. This means that significant underexposure may lead to significant quantification noise (part of it being caused by an analog noise floor, part simply because of the D/A converter step size).