There is much confusion going on here. The responses were less triggered by your concepts than by the provocative title.
The term “well exposed” in the above sense is not directly related to “no underexposed” from your title. You include prior knowledge that was not mentioned by others.
The fact that there may be no absolute notion of underexposure grounded on PSNR does not mean that underexposure does not exist at all.
I find the concepts and differences in handling exposure in analog and digital photography very interesting. Unfortunately, the fruitful discussion gets lost in a dogmatic dispute about terms.
As has already been said, and I tried to explain it myself, it’s just a misunderstanding with some terms: what I try to get from an image, that has captured a scene, is a certain feeling, something that has drawn my attention. So I choose a combination of speed-aperture (I always shot at my camera base ISO, that is ISO100) that gives me the image I wish. And that look, that image, is my objective (a.k.a. my target). Here is where one misunderstanding has gone for days. I don’t care at all about the scene middle-grey. I don’t care about my shot middle-grey. The only second I care about middle-gray is when my camera suggests a speed-aperture combination. Then I think about that suggestion and apply some compensation to get the look I wish. In a sense, I’m moving my image bounds up and down, but I know I can’t get the whole dynamic range of the scene, so I choose what fits me best. I choose which highlights will be clipped (if any), given the higher or lower importance of the shadows in my image.
I already have a middle-grey card, but I almost never use it.
Given my previous explanation, again there is a misunderstanding. I think we are talking about the same thing, but we are arguing because we don’t use the same words and then it seems that we talk about different things.
So, given that I compensate up or down my camera suggestion, and given that exposure can be understood as the amount of light per unit area reaching an image sensor, then the combination I set for my image (the exposure), that gives me the look I wish, to me is the right-exposure for that shot. And most probably it won’t be what my camera suggested me, but it’s the exposure I need.
And my target (as in my objective) is not a number, is not the bull’s eye, it’s the feeling I have when I look at the image.
I have shot film, and it always has been the same. My camera and its settings are just a tool to get an image, not to get a perfectly exposed middle-grey.
I write these posts usually after the 4th-5th time I repeat the exact same thing in an email because users are lost. They are lost because when you expose them the pipeline after using dumbed-down software for years, they suddenly don’t understand the meaning of the sliders that just appeared. They are lost because, suddenly, the input can be whatever and the output can also be whatever, so if you don’t get how they are connected, then… you can’t use the software. And HDR output is not even there… Imagine in a couple of years, I can already hear the question “why is 100% luminance not white anymore ?”.
Exhume what you will from the bottom of the internet, I will still be the one they mail when they don’t understand why “well-exposed” picture still need an exposure boost in software.
Important concepts are being named by those words. Words influence the way we think, hence the way we understand, hence the way we act. Terms matter.
This is clearly because most people don’t know what a linear raw looks like. I had this issue for a little bit as well tbh until I started to dive into a bit of research. Most “professional software” hides the linear raw.
Lightroom for instance behind the scenes applies not only a tone curve but also a contrast curve, contrast slider, blacks/whites sliders, as well as a exposure compensation. It is absurd. I can guarantee Lightroom users while culling toss out perfectly fine images due to all of these hidden adjustments which force them into trying to compensate with absurd extreme use of highlights and shadows sliders that they never needed to touch to begin with causing halos and other nasty artifacts.
After I learned how to I was able to reverse the lightroom hidden stuff and get much much better results. You know something is wrong with the software after you reverse all the hidden stuff you end up with a auto generated reverse contrast curve, neg contrast slider of nearly -50 and a -1 exposure on your image.
At least capture one gives you a linear raw if you want it but still has halo and artifact issues if you are not careful.
This is why I think people get confused they are use to hidden adjustments software is doing that it should not be doing.
I think it was actually you who introduced “target” , “18%” etc discussions through the filmic interface and various discussions and tutorials? Users who then tried to figure out filmic naturally communicated their questions with the established concepts and vocabulary. I mean it’s literally the names you’ve assigned the main sliders of filmic! If any great number of people email you using those terms it’s because you established them in the software. They are only trying to adapt to the “correct” concepts and words established in your software. I doubt any of them, if their geek/photographer balance is toward the latter, would have used those concepts otherwise.
As pointed out by various posts and links to old material and wikipedia most photographers today conceptualize exposure close to your “new” way of thinking. It’s pretty obvious because taking a few photos you’ll figure it out. I you don’t figure it out yourself any tutorial on the internet will set you straight. Perhaps manufacturers and their gear doesn’t think like that but it hasn’t stopped photographers.
Again no one but you in this discussion conceptualize exposure as a fixed number target. You’ve carried on a 80 post discussion not reading posts but talking to a mental image you’ve formed from other discussions.
“Middle gray” simply represents the mid-tone in the image, the thing you want to be depicted as the “middle tone”. All of the other tones are then splayed left and right from there. What @anon41087856 is saying is that, for a single image, where that is captured in the camera’s range from the “noise carpet” (I’ll use that term, carpets are softer than floors… ) to the hard sensor saturation point is meaningless until the final tone conversion for display or output. Because, that’s where the intended “mid-tone” has to be at a perceived mid-point…
Yes, but he’s arguing a straw man. No one here seems to care about middle grey. I can’t see any believers in “correct exposure” (as related to middle grey) anywhere? The wikipedia definition is the one commonly held.
See also the definitions of over and under exposure. All defined subjectively.
@anon41087856 I sympathize with your plight as a developer and providing support for your software. Generally, users are pretty uninformed and expect miracles…
However, this is a forum for photography enthusiasts who honestly wish to learn and become better photographers. So please, as @Thomas_Do said don’t let “the fruitful discussion gets lost in a dogmatic dispute about terms.” We’ve lost Troy Sobotka and Elle Stone (partly) due to conflicts in finding a reasonable ground for discussion and learning. I sincerely hope this won’t happen with you or anyone else in the community.
I think you are a great teacher in many respects, and I have learned a lot from you. I admire your work and dedication to the development of darktable. I even believe you are absolutely right for the vast majority of the cases. But as a high school teacher myself, please, if people don’t get what you mean and a heated dispute arises with lots of confusion, please consider two things: highlight where people are actually right (reinforce the good things!) and find other ways to correct people’s mistakes instead of simply repeating yourself. I don’t like to learn anymore when someone shouts I am simply wrong or I simply don’t understand.
That being said, onwards to more discussion! Bring it on
The most prevalent definition of “correctness” regarding exposure is the placement of the tone that would be “middle gray” at the center of the human-oriented histogram. That’s to what the spot meter in your camera is calibrated.
We all want the subject to be “just bright enough” in the images we regard, else we complain the photograph is ‘too dark’ or ‘too bright’. Where the mid-tone ends up in the rendition is of supreme importance. We lament about our crushed highlights, noisy shadows, and all that angst is really anchored on where we want the mid-tone to end up.
I think in recent days the thing that help me understand the most about the whole SNR/Clipping/Mid Tone concept was when I started to shoot macro. The main reasons being extremely little light to work with as well as a very low DR to work with. Then by analyzing the linear raw and how to manipulate it with a raw processor. It is a great way to see how middle tones behave because you have total control over everything during the shot and not fighting with nature. This also helped me understand the differences between the preview and the raw thus my camera because the difference is extreme in such a case. Plus the fact of shooting with manual flash thus the meter in the camera does not dictate the “assumed exposure” just helps with the flash setting.
I think the best way to communicate these sorts of concepts to users is to always show a raw histogram that will show users what they’re actually getting off the sensor.
In RawTherapee you can click a (teensy-tiny little) button to show the raw histogram, but in Filmulator I’ve put one in the sequence of mini-histograms that shows what the pipeline is doing to the image data.
It’s the uppermost of the three smaller histograms, and it shows that this image is definitely underexposed by almost two stops relative to optimum.
Averaging noisy data can reduce measurement uncertainty but not improve accuracy. If the noise is white, i.e. the noise density is constant across the frequency spectrum, then averaging N samples will increase the SNR by sqrt(N). Averaging is low pass filtering, hence the high frequency noise is attenuated while the low frequency signal components remain unaffected. The cost you pay is that the high frequency signal components get attenuated too. Applied to image processing this means that spatial details get lost while the noise in uniform areas is reduced. Applying a gaussian blur to the image from a few post above and applying a gaussian blur with radius 10 yields:
Averaging is the key to oversampling Delta Sigma ADCs. They use an 1-Bit ADC and massive oversampling to get 20 Bits or more output resolution. See http://www.ti.com/lit/an/slyt423a/slyt423a.pdf for an explanation.
Information isn’t lost but masked by noise. The low frequency features can be recovered by low pass filtering. Intentionally clipping samples at the noise level removes recoverable information masked by the noise.
Concerning exposure: IMO exposure should be chosen to get the maximum possible SNR without clipping important features of the image. Avoiding motion blur may dictate to sacrifice some SNR. It’s all about finding a good compromise and “good” is highly subjective. I agree with you that in digital we have way more freedom choosing exposure than back in the analog days.
I oppose nothing in that document, it is good, everyone should have a very good read of all what is in there. The thing I am opposing is the statement that is your interpretation which boils down to
And
You can easily test this with simple codes…other people in this thread have given examples of and arguments for it. Your statement is a factually wrong claim. Repeat measurements (given repeatability of the setup) of small signals(measurand) in large random noise, yields the measurand (your signal) with a margin of error (which should be given, even as an estimate, for any measurement anyway).
If a reconstruction yields the measurand with a low enough statistical error, there is a strong argument for its ‘presence’ within the data. The quality of this reconstruction is the topic of whole journals in research.
I do not think so and you could not convince me of it.
This.
And I might add that low pass filtering is a simple concept of filtering. More complex concepts exist as we know from the vast field that is denoising/noise filtering. Low pass filtering serves as a nice example though that everyone can understand and follow.
It’s been very clear that this thread had had a lot of unnecessary hiccups and animosity due to a lack of consensus on terminology.
Aurelien has brought a lot of insight to this community over the last several months, together with a huge amount of work on dt. However, the title of this thread has clearly triggered a lot of people (silently including myself).
So if I go outside right now and take a photo at f/16 and 1/500 seconds and no flash, never mind what the ISO is (it’s about 10:00 pm right now where I sit), it would be legitimate to call me silly or stupid. But what would you call the photo? I would say the sensor wasn’t, well, you know, “exposed” to enough light to make a useful photo. @anon41087856, what would you call it? I’m not asking this in a confrontational sense, I just want to break down the terminology barrier.
There is no such thing as a well-exposed / under-exposed / over-exposed picture in digital. There is only sensor clipping or no sensor clipping, that is all.
My interpretation is that seeing as there is no “correct” exposure, i.e. no one objectively optimal exposure setting for any given scene, there can therefore be no under or over exposure.
An exposure that clips the highlights isn’t “over exposed”, its been exposed to retain more detail in the shadows. An exposure that loses some of that detail in the shadows isn’t “under exposed”, its been exposed to protect the highlights.
The part differentiating between the linear digital capture vs the mapping that is baked into film that follows is critical here, because in order to optimise that baked in mapping on film there is an optimal/target exposure. In digital however you can apply positive or negative gain with little consequence.
Ultimately that appears to be the point here, on film underexposure meant you were burying the scene in the low end of the curve which had all sorts of nasty effects on things like saturation and contrast in your image. It seems as though Aurélien has encountered a lot of people that think having to boost their digital RAW files by two stops means their photo must have been underexposed and so they must be suffering the same detrimental effects, and that is not the case.
Of course if you’ve failed to capture the detail you intended to in either the light or dark areas of the image it seems reasonable enough to call that over or under exposed respectively, but that’s just me.
Agreed, and this is unfortunately where we have somehow been getting into needless conflicts. Not one person in this thread has been saying there is one objectively optimal exposure.
That’s a non sequitur, and is probably at the root of the problem. I think we can generally agree that there can be a reasonably wide range of exposures that are acceptable for one person or another’s artistic objectives (edit for clarity: and a range of exposures, that surround a given person’s ideal, from where they can rescue their photo to get what they originally wanted) . People prove that daily with their different interpretations in PlayRaws. But the fact that reasonable exposure is a (somewhat subjective) range rather than a point does not infer that being well outside of that range is not under or over exposed.
) to the hard sensor saturation point is meaningless until the final tone conversion for display or output. Because, that’s where the intended “mid-tone” has to be at a perceived mid-point…
