ISO invariance understanding it - just checking something

Started 2 months ago | Discussions thread
Brian Kimball
Brian Kimball Regular Member • Posts: 485
Re: ISO invariance understanding it - just checking something

CESA wrote:

DN perfornace

If we look at that plot from any of them but in particular the A73, analysing from the ISO point of view it means that from ISO100 to ISO500 it is ISO invariant and the from ISO640 all the way up it is also ISO invariant?

More or less.

If so, I have a couple of questions here:

1. How do we correlate the exposure slide and the highlights and shadow sliders with this in terms of noise? I will explain:

I'll go down this rabbit hole, with one huge caveat: understand that post processing sliders are not designed to recreate physics or mimic their physical namesakes.  They are designed to produce pleasing adjustments to images, period.  For example, ACR/LR's exposure slider is not linear and does not impact shadows & highlights in the same amount it impacts midtones.  There's a built-in curve there that you can't see or modify.  Also, ACR/LR's shadows and highlights are not blunt instruments- they are adaptive to the content of the image.  Thus they provide different results for the same slider positions on different images.  Also interesting to note that a certain amount of highlight recovery is automatically performed in ACR/LR, even if the highlights slider is at 0.

a) If I underexpose, by say 2 stops, a picture using ISO100 in post to get a correct exposure I would bring the exposure up by 2 stops without losing details and introducing noise?

Correct, but you ARE increasing noise.  Just no more noise than if you had shot at ISO 400 with the same aperture & shutter speed.

b) This would be equivalent to have shot at ISO 400 if I had aimed to correct exposure using same shutter speed?

I'm having trouble parsing the "aimed to correct exposure using same shutter speed" part- if you mean in all of these examples, aperture & shutter speed don't change, then yes, it would be roughly equivalent.

c) So by doing this one can protect highlights.


2. Now another scenario would be: I increase the ISO to say ISO400 in order to have an higher shutter speed, almost like shooting to the right of the right.

Increasing shutter speed and increasing ISO proportionally cancel each other out with respect to image lightness, so this is not "shooting to the right of right."  This is shooting at your metered exposure.

This would still be in the first range of ISO invariant section (assuming that what I said above is correct).

a) So this means that in post I can bring down the exposure 2 stops (equivalent of have shot at ISO100) and get the correct exposure having protected the highlights?

Increasing ISO above your meter's recommendation does not protect highlights, it threatens them.

At a fixed aperture and shutter speed, if neither ISO 100 nor ISO 400 clip highlights, and it's ISO 100 that provides the correct image lightness, then just shoot ISO 100.  Shooting at ISO 400 and dropping image lightness by 2 stops in post gives you no benefit, gives you a source file with more read noise, and drastically overcomplicates things.

This would mean I would be introducing noise in the shadows right?

Shooting at an unnecessarily high ISO that results in an overexposed image requiring darkening in post increases noise in your captured source file.  Yes.  Decreasing the lightness of those shadows by 2 stops in post will make the noise less apparent in your final exported image, but that noise is unnecessarily & permanently recorded in your source file, for no benefit.

b. Again I guess that this would have been equivalent to have shot at ISO100 without losing quality and adding noise right?


3. Now in this scenario I Am curious to know. Say that you shot at ISO400 and have underexposed by 1 stop. So we are at the edge of that ISO invariant interval. Done this to increase a lot the shutter speed.

a) Now in post I would need to increase the exposure by 1 stop. Would this mean that we would jump to the next range of ISO invariant interval?

If you mean, "I should have shot at ISO 640 and lightened my image by 1/3 of a stop in post, instead of shooting at 400 and lightening my image by a full stop in post"... yes, that's a reasonable conclusion if you're trying to eke out every last bit of performance.

Would we expect more noise? Probably a different colour artefacts only?

Numerically more noise in the ISO 400 shot lightened by 1 stop.  How that noise increase manifests visually is up for you to test.

b) So would you also jump to the next interval if you had shot at ISO100 and increased the exposure by 3 stops?


4. I guess that looking at the plot we are better off shooting at ISO640 instead of ISO400 looking at the ISO performance no?

ISO performance

Technically, yes.

Now another question related to point 4 is. If we shoot at ISO640 and if we bring down the exposure to the correct place as long as we decrease it by 1 stop at least we are good because we will not end up on ISO400 which has a worst performance. Is this thought correct?

If I understand you correctly, you're looking at the above PDR graph and thinking "as long as I avoid the dip between 400 & 640, I'm good noise-wise, regardless of what I do in post."  That's not really a correct reading of the PDR graph, as it does not show the effect that lightening has on visible noise in post.  A graph of dynamic range vs ISO is effected by read noise, but is not in and of itself a graph of read noise performance.  It's also a graph of the inescapable downward trajectory of DR as increasing ISO raises the noise floor and pushes highlights past clipping.

To be more specific, you seem to be wanting to compare

  • noise at ISO 640
  • noise at ISO 400 + 2/3 stop of lightening in post
  • noise at ISO 320 + 1 stop of lightening in post

320 & 400 should be roughly equivalent.  640 should be (technically) better than both.

Another question is regarding the highlight and shadow recovery. I know these can add noise however how do they corralate with the ISO perforance and DN read? The same analysis in terms of increasing or decreasing the exposure and ISO invariance and dn read noise applies? Do you know what I mean?

You might want to take this question over to the PS&T forum.

Hope all this helps. 👍

Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow