More noise is very low light regardless of ISO

[funkytwig]

Hi, did not know what to put up as a subject for this so the one I used is not 100% correct but does comunicate the basic idea.

I did some low light tests with my GH3 and there seemed to be a little noise at the top 3 ISOs. I then did the tests again but this time had a little more light (the previous tests were done with very low light).

With these new test I get no noise on any of the settings. My conclusion is it is not just the ISO that affects noise but the amount of light. I.e. if you have very little light (for a given ISO) there is a highter propensity for noise than if you try the same test with more brightly lit scene. Therefore for true ISO noise tests you must shoot with various light levels as well as ISOs. This means that the test that I often see where different ISOs are tried in daylight only tell some of the story.

This is something I suspected (with my 550D) but now I have done more exhaustive tests with my GH3 I am becoming increasingly convinced it is the case.

What do people think?

Regards,
Ben

 

[Geoff Gilson]

What you describe there holds true for any digital camera I can think of, but not quite in the way you describe it.
Exposures with the same signal gain (in this case ISO value), will produce a measurably similar amount of baseline noise, regardless of the signal content.
Encoding formats aside, this holds true for both still images end video clips, because both modes sample the sensor in much the same way, signal-wise.


When you compare two images of the same scene, both taken with the same ISO setting, the image taken with better lighting will show less noise because it has a superior dynamic range (the numeric difference between the darkest and lightest parts of the image).

If the darker-exposed image is lifted to match the exposure of the other (brighter) image, it will look noisier because the overall noise is being amplified.

Also a lack of exposure/range in an image's colour components can show up as noise in dark areas, because there are gaps in the data (for instance, when you shoot a street scene lit only by orange light and there's no blue in the image, so the blue-filtered areas of the sensor don't receive a good exposure.)


Or think of it in another way. If you're listening to a vinyl record, you hear the surface noise and crackle a lot more keenly when the music is quiet, or at the gaps between tracks.
If the recording is too quiet for your liking, you can turn up the volume (gain) but there'll be more hiss.
When the recording is loud enough in the first place, you really just notice the music when it's playing. It's exactly the same for sound and images - a full signal with good dynamic range should sound cleaner to the ear or look cleaner to the eye. (That is, when played back correctly.)

 

[funkytwig]

If the darker-exposed image is lifted to match the exposure of the other (brighter) image, it will look noisier because the overall noise is being amplified.

By lifted do you mean use higher ISO?

Or think of it in another way. If you're listening to a vinyl record, you hear the surface noise and crackle a lot more keenly when the music is quiet, or at the gaps between tracks.
If the recording is too quiet for your liking, you can turn up the volume (gain) but there'll be more hiss.
When the recording is loud enough in the first place, you really just notice the music when it's playing. It's exactly the same for sound and images - a full signal with good dynamic range should sound cleaner to the ear or look cleaner to the eye. (That is, when played back correctly.)

I dream of vinyl noise.

Regards,
Ben

 

[Geoff Gilson]

By lifted do you mean use higher ISO?

Sorry for the confusion. By lifted, I mean exposure corrected in post.

Even out the differences in both examples, and the clip/image that was originally darker should look worse for it.
A still image will look noisier, and a video clip will look noisier / blockier and the colour will start to break up.

I dream of vinyl noise.

Electrical noise would be more appropriate, but surface noise is a clearer example. :3