Why dark and light is complicated in photographs
aaronhertzmann.com
aaronhertzmann.com
For digital, the data directly from camera sensors almost always needs some correction, de-mosaicing or massaging to generate an image viewable on a screen. This requires the camera to make what ends up being an aesthetic decision on what the photo looks like. Detail isn’t just how bright or how dark, but also the available gradients in between. This means there are cases where the dynamic range is automatically expanded (instead of clipped) and contrast unnaturally increased in order to have a photo that isn’t just mud.
Ultimately, this means that technical considerations map directly to artistic ones, and there is no objectively correct image from sensor data. The idea that a ‘no filter’ picture conveys some kind of divine truth is a myth.
Your description of reciprocity failure is not quite right. The idea is that if you double the time the shutter is open and also decrease the aperture by one stop, you should not change the amount of light hitting the film (you will change other things, of course). The overall brightness should be the same when you make these “reciprocal” adjustments. This does in fact hold pretty well within a certain range of shutter speeds. Reciprocity failure occurs at longer or smaller speeds, where the reciprocal relationship doesn’t quite work.
It's a game of fitting the captured scene with e.g. 11 stops of dynamic range to an output medium that has maybe 5 or 6 stops of dynamic range in order to trick human eyes with closer to 20 stops of dynamic range into believing what it sees on the print is "real/plausible". That is of course a very subjective process that involves a lot of choices regarding how to shape your curves and compress the available data.
Additionally, human brains do "funny" things with colors and actually re-construct/imagine things that aren't there. Getting the white balance and color grading "right" is another set of steps and what is right depends both on the conditions under which you capture the scene as under which conditions you view the end result.
Most phone cameras and in camera processing software algorithms for all of this that produce a certain look that is pleasing under most circumstances. But those are basically just a hard coded interpretation of the raw data. The point of post processing is having finer control over that process and the ability to use more expensive/better algorithms.
Each rgb channel is a filtered spectrum. But note that the color filters are essentially arbitrary, and only roughly correspond to the color filters of your eyes or your screen.
Also note that some cameras pre-process even raw data, for example with vignetting correction or noise reduction. Some phones even do image fusion and color correction, then re-rawify the data by mosaicing and linearization.
The worst trend being loud music and mumbled dialogue that make it impossible to experience even as a radio drama.
As an examples, I was watching the show '1883' and half the time I cannot understand the dialog!
Now I have the most expensive best new Sony 65" you can buy, so I'd expect the sound to be at least half decent. So it's not my out-of-the-box speakers.
I feel like I'm constantly having to turn the TV up!
Some Googling had people on forums suggesting it's 'muddy mixing' whatever that is, and that shows are now mixed for Surround Sound.
However when I watch a 'normal' TV show like South Park or an average British SitCom, I can understand every word.
The situations is bonkers and its turning me off even bothering to try and watch some shows!
“My photos are a collaboration between me and the algorithm designers and camera manufacturers, and they reflect a combination of aesthetic decisions made by each of us.”
The one downside of taking pictures with such a strongly opinionated technology is that your pictures are going to look like everyone else’s (or increasingly just weird). Ansel Adams developed his darkroom technique over decades - and as the author mentions, applying them took hours of meticulous labor. So his output really did look different from most other peoples’.
You can still differentiate your photographs on the basis on their subject matter, of course. But if everyone is shooting out of the same algorithmic pipeline, making your pictures look better technically is going to be increasingly hard. (And given current limitations, someone knowledgeable in the field today can still usually tell the difference between photographs taken with good lenses carefully deployed and cell phone output. But it’s unclear how long that will be true.)
It is immensely easier to do in light room what previously took hours in a dark room and years to master. We can now spend more time thinking about specific aesthetic choices and less time making test strips of 1 stop intervals on a small strip of the sky to correctly dial in exposure. That’s a process that can take a very long time, involve lots of chemicals, and expose you to some carcinogens.
I love film photography but it is laborious.
Colour transparency film (e.g. Fuji Velvia -- RVP50) shows the same thing as clearly.
You basically can't map real world light into the dynamic range of a typical camera without causing some of this experience, can you?
The question is how you determine how dark shadows should be -- your brain is doing a lot of work to hide from you the tricks it uses to make shadows appear less dark than they might be with a linear response.
Or even how they would look with a non-linear response that is even across the "frame"; your brain is doing localised dodge/burn type work, constantly.
Camera manufacturers have "tastemakers" for this stuff on digital, just as film manufacturers used to have them for film.
Any brain filtering would have to affect the photos as well, even if it was true.
No common image format uses linear response. It would explain this problem if cameras treat them as linear.
Maybe they should just make the cameras take physically correct colors, instead of relying on people, as the typical person will always choose extreme contrast that will make the camera unusable. (and can be easily increased in editing)
Our brain does a perceptual aggregation of multiple frames and inputs. This is not how cameras work.
Also "make cameras take physically correct colors" is impossible unless you're talking about spectral capture, which is orders if magnitudes more complex. If you're using just RGB photosites AND RGB displays, there is no such thing as physically correct colors. Everything will just be a mapping at best, with the best that color science experts can actually provide.
Reality --> eye --> "brain filter"
Reality --> photo --> eye -> "brain filter"
Cameras should only record the colors as accurately as possible. Or if you want to nitpick again, so that the photo stimulates the eye receptors identically to whatever was captured.
And also introduced me to the word "ontogenically". Thanks twice.
Color is incredibly complex. It's easy to say "we should capture it as accurately as possible" but I don't think you fully comprehend the high complexity involved.
Your concept of matching eye receptors is wrong too. Color is perceptual and subjective. Your perception of color is based on your upbringing, your genetics, your environment, your own mental faculties, your mental state etc... What is accurate? Your eyes see some spectral energy, your rods and cones convert those to signals, your brain then adds that into an aggregate set of information that it's constantly infilling and, most importantly, guessing about.
You can't guarantee that multiple people see color the same.
Now even if a camera could hypothetically capture an image accurately to the real world (IMHO only possible with a hypothetical full spectrum sensor), how would you store it? The second you convert it to RGB data it needs a perceptual conversion to the bit depth of the data format. Now even if you have a file format that can efficiently represent this, you'd also need full spectrum displays so that we could beam that exact color to your retinas.
Color science is incredibly complex. You're trying to trivialize it into matching your own narrow perception of color.
For example you might want to go to a beach or a park and shoot throughout the day on a bright day. Put people or objects in the foreground and then shoot them with either the light behind you or in front. Use the incident metering dome to meter the light
(you'll need to look this up, but the broad point of it is you stand in the same light as your subject and point the meter into the light, rather than at your subject)
Once you see what transparency film does in high-contrast situations I think you'll better understand what I'm trying to get across.
Not when the dynamic range of reality is much greater than that of photographs, and your visual system is one of the best visual processors in existence. It’s like reducing a 24-bit image to 16-bit - the image is “good enough” to identify the subject, but it is quite lossy. Photography itself is a lossy process.
The dynamic range is clearly there. And we're not talking about such ridiculous values that the sensor noise becomes prominent.
You can do that correction in that situation because you've looked at the image, you know what it is meant to be, and you can decide on a set of adjustments that produce something that approximates what you want, perceptually.
But without truly extensive scene knowledge, cameras can't do that automatically, and they also can't know what information that is important to the photographer that they'd be affecting if they did.
Cameras have to try to ascertain what would be middle grey in a scene and then apply a general purpose tone curve to an image, but they do not know what is in the scene.
They can't even know for sure if the photo they are being asked to take is properly exposed by any absolute definition, in fact.
[I cut out a lot of this because I don't think it's going to be easy to complete the explanation here]
Gamma is just a single element of a color space. The whole trivializing sRGB into being a 2.2 Gamma versus linear is an over simplification that ignores gamut , white point, bit depth and more.
That's my point. You're trivializing what color is because you clearly have not dealt with it at scale.
You can download an alternate color scheme from Technicolor for Canon cameras that makes the colors appear fairly reasonable.
From what color space to what color space, and what display or print type?
Now you're back to describing a average case perceptual color rather than your initial point of accuracy.
Have you ever shot photographs with a colour transparency film?
Say that your eye is sensitive from light intensity 0 to 100 in some units, but your camera sensor only handles 40 to 60. That means that everything under 40 will be mapped to black, and everything above 60 will be mapped to white.
Only the picture file format should limit what range you can save with any modern camera.
The main limitation isn’t the file format. The main limitation is the sensor. On the lower end, it is noise in different forms that overwhelm the very weak signal from dark areas. On the higher end, the sensors get saturated, that is, the semi-conductor bucket for the charges that is released by the photons get full.
And then the experience of the picture is of course limited by the medium that is used to display it. Even the best screens can’t show even a small fraction of the contrast that the eye experiences outdoors on a sunny day. And don’t mention printed media.
In real life, photographs are printed on paper. The brightness of light reflecting off paper depends not only on the colour of the paper, but on the brightness of the illumination. (Likewise, photographs displayed on a computer monitor depend on the screen's brightness.)
In real life, human brightness perception depends on the brightness of the environment. An LED can look bright in the dark and dim in sunlight, and range dim to medium to bright on a cloudy day without anyone really noticing that the clouds between them and the sun are thicker or thinner.
In real life, there is no 0. There is no 100. Your comment doesn't make any sense.
The narrow band of sensitivity of a film or sensor has to be sort of moved to where it is needed (by controlling how much light gets in or for how long) according to the result the photographer is likely to want from their photo.
Even the most basic of film dead-reckoning methods -- Sunny 16 -- relies on subjective input from the photographer:
https://en.wikipedia.org/wiki/Sunny_16_rule
And it's up against the nature of human perception of light and dark, which as this classic page demonstrates, is complex:
https://scienceinfo.net/video-chessboard-illusion-confuses-p...
There's a fundamental problem with photography: a scene can contain up to 20 EV of dynamic range. Your camera can capture up to 14 EV. But your screen or a print can only depict 5-10 EV.
So something has to give. The camera capture is usually fine, only excluding the brightest highlights like the sun itself, and the darkest shadows, both of which we don't expect to contain detail anyway. But mapping the rest into the (let's say) 8 EV of available dynamic range of the screen is a problem.
Film does it in two ways: it rolls of highlights and shadows smoothly, preserving mid-tone contrast while compressing (and desaturating) extreme tones. Secondly, chemical diffusion enhances local contrast, while reducing global contrast, a bit like the "clarity" slider in Lightroom.
Smart phone cameras do something similar, with HDR image fusion, local contrast, and tone mapping.
Which is why, ultimately, only an edited picture can represent the experience of viewing a natural scene. The full tonal range of the scene can not be displayed. We have to rely on a somewhat artificial compressed rendition instead. But that's not a fault of the camera, or even the screen. But simply the result of showing an emissive reality on an (assumed) reflective medium. To me, it's the same perceptual unreality as projecting a 3D scene onto a 2D medium.
To me, it's what makes photography (and visual arts in general) interesting. They don't show reality, but a depiction of reality. And in that crop, project, compress, lies interpretation and expression.
8bit sRGB can contain 11.6 f-stops of dynamic range, and it is no longer enough for modern screens. There must be a longstanding bug somewhere. It would not be the first time it happened. There was a chroma decoding bug in almost all DVD players. A very similar issue to this, antialiasing in CGI used to be calculated without gamma, until GPUs got fast enough to use it and I suppose that someone started looking into why it performs worse than expected.