“When I use a word,” Humpty Dumpty said, in rather a scornful tone, “it means just what I choose it to mean—neither more nor less.”
“The question is,” said Alice, “whether you can make words mean so many different things.”
“The question is,” said Humpty Dumpty, "which is to be master—that’s all.
13 Likes
It’s not a fundamental thing to understand but it is completely irrelevant to this discussion.
This reconceptualising you’re trying to do is probably great for the software. Thing is that most photographers born since the 70’s probably already conceptualize exposure like you propose even if cameras and software don’t quite help along.
All these photographers also know that you are wrong about there not being underexposed digital pictures. We all have them and we have all binned them. Noise
1 Like
Please humour me with this comparison you alluded too earlier:
Suppose you record audio with amplitude recorded in numbers between 0 and 100. If you record audio with the gain too low you’ll hear more background noise. When someone is whispering and the signal is between 3 and 8 in amplitude it’ll be lost in the noise which is in the range 1-5. So you increase the gain so that the whispering is scaled up and is now between 6 and 15. The person whispering never raises her voice such that numbers over 100 are collapsed into 100/distort etc. You could argue that this is “well-recorded”.
Now switching to the analogous photographic situation where the iso/shutter-speed/aperture is set such that some shadows in a dark room are to be captured and represented with as little noise as possible. The camera is on a tripod so you can have arbitrarily long exposures at any aperture. Suppose you took two photos, differing only by shutter speed. In photo one, with the shorter exposure time, you’d have to increase the exposure in darktable by more than you would (it at all) in photo two. The situation in the room is such that there’s no problem with clipping the highlights, just as in the audio recording nobody was shouting; it’s a dark room light with candles, for example. So by increasing the exposure in software, you’re going to also increase the noise, aren’t you? So in this situation, couldn’t you say that photo one is underexposed, or “less well exposed than photo two given the subject matter and situation”, or some other very similar description, and that signal and noise are getting mingled in a way which a longer exposure would avoid? I appreciate that you can compensate for bringing up dark parts of a photo far more satisfactorily than you can with clipped highlights, but that doesn’t change that one is “better”, “closer to the target” etc, does it?
That’s nonsense. There absolutely exist circumstances where there is zero highlight clipping in a capture (and there is no such thing as shadow clipping). And in those circumstances, there does exist an unambiguous concept of “underexposed” even for raw photos.
2 Likes
Could you please elaborate on this concept. There are a lot of assertions in this thread about there’s no such…, but there are a few of us that don’t have the same thinking, and sadly we don’t have enough knowledge to just say true, true, …
Does it mean that there is somewhere an electronic device designed to capture light (photons) that no matter how few photons are hitting the photosites, there will always be an electrical response that will be translated to a certain R, G or B value? Let’s not mix noise in the recipe. Let’s talk about an ideal circuitry that gives absolutely no electronic noise. We still have the floor sensitivity of an electronic device. Do you mean that it has an infinitely low floor sensitivity? That the sensor will record light no matter how low the luminance of the scene is? If so, then yes, we could definitely say that there’s no such thing as shadow clipping.
But honestly, I think I have missed something, and I would really appreciate a simple explanation about it.
1 Like
Yes, the sensor will always record light no matter how low the luminance of the scene is. The photodiodes will (have a probability to) respond to single photons falling upon them, so no matter how low the light is, if there is any light at all it will respond.
In a real sensor, because there is noise, the amount of signal produced by the light eventually gets swamped by the noise. But that’s not clipping, unless in post-processing you choose to clip the blacks.
Yes, that’s why I said to not consider noise. So ok, I didn’t knew sensors where so sensitive. Thanks!
We need to consider the separate sources of noise. This page has a decent, not-too-technical summary:
1 Like
Well the presence or absence of noise is irrelevant to whether or not shadow clipping exists.
The title of this thread is not correct! It ignores noise and digitization.
Here is a simulation of shot noise (same scene, different exposure times), which demonstrates how an underexposed image looks like:
Hermann-Josef
I suppose dark current correction is only cosmetic then.
THERE IS NO TARGET UNTIL YOU REACH OUTPUT MEDIUM !!! How is that so difficult to understand ? You SO BADLY want ONE exposure setting to be deemed RIGHT ? That makes you feel better ? It brushes your scientism the right way ? “Numbers ! They mean Truth”, is that what it is about ?
Seriously… When you point at the Moon, the fool looks at the finger.
When signal < uncertainty(signal), having a sensor photonic response is not the matter anymore, or signal processing has changed a lot since I was in school. Every pixel whose intensity is below noise variance goes to the garbage bin and gets a 0.
The title of this thread has a follow-up, it’s called a post.
I think I mentioned exposing for the scene exposure range, your example shows exposing to degrade PSNR on purpose.
We’re talking about different things here. You’re talking about zero subtraction. I’m talking about how any nonzero intensity of light, averaged across a sufficient area to eliminate the effects of both shot noise and read noise, will be measured as nonzero after zero subtraction.
That’s being way too pedantic. Simply denote some brightness level above which you don’t care (which is related to the decisions you’ll make when you go to prepare the output medium) and if any of that region is not clipped, then you’re underexposed.
Ideally, noise reduction should take into account the negative values, or else you introduce bias.
Cool down, please 
1 Like
Again… mistaking target with bounds.
Every non-zero intensity of light that is below noise variance/measurement uncertainty is non-data. That’s Metrology 101. There is no discussing this. Besides, negative light makes zero sense.
So I’m commenting on the right thing here, TARGET = middle gray, no?
If so, I get that, the last thing to be done before pushing the image array to the rendition medium is to apply the tone/color transform of the output profile. Until then, the middle gray isn’t in the right place.
If not, need a definition of TARGET…
Indeed.
Yeah, I get that you’re triggered by certain words, but once you’ve stopped shouting and wiped away the slobber from your mouth perhaps you can address what’s wrong with my analogy. Why isn’t the recording/photo with the higher signal/noise ratio better?
Thanks, I’m going back and forth from this to yard work, and my elder dementia keeps me from following well … 
I’ve got to say that my hack software has helped me to see all this in action. When we started discussing ‘linear’, well I just reorganized my toolchain. With regard to the display transform, whichever tool I click the checkbox for is the one that gets piped through the display transform to the screen, and I usually just have a resize as the last tool and checked, so the display transform doesn’t take so long, so having TARGET not be where it’s supposed to be until that happens makes intuitive sense.
I think the point of all this is that the noise floor is a different sort of rubicon than sensor saturation, has different consideration to be applied in placing the data… ??
1 Like
With the presence of noise contributed by sensor and readout, you simply have to account for negative values after black subtraction. They are not negative light values, they are negative electrical values, and cannot be omitted if you are trying to make an unbiased estimate of the actual light value.
1 Like
Yes, let’s keep in civil! No harm in stepping away from the keyboard for a few minutes. Civility it a requisite, not a suggestion.
2 Likes