What is the correct WB?

[digitaldog]

This discussion has in the last few posts about setting WB (if respecting the fact the OP is using LR) and had a simple question about subjective renderings:

“There seems to be some perverse human characteristic that likes to make easy things difficult.” -Warren Buffett

[DP]

Yes, and they are completely irrelevant as far as whitepoint is concerned. The only thing we need to know is the whitepoint of the sensor at the time of capture. Nothing to do with the scene lighting. Merely its electronic state.

please be so kind - define what is "electronic state" of the sensor  ? and it given the word "state" you used please explain why'd that "state" can be not related to "scene lighting" "at the time of capture" ?

[DP]

What can happen (but not likely) is that the sensor is calibrated towards D65. The dyes are chosen and electronics set so the sensor behaves as if it has a CCT of ~6504K or thereabouts.

please define what is "calibrated" ? as noted no consumer level cameras have CFA that yield anything close to  UniWB when illuminated by anything close to any daylight w/o using some magentish filters in lenses... hoi polloi like me understand "calibration" for a particular spectrum as getting all raw channel saturated equally when illuminated by (sensor that is, not the scene) that said spectrum... give me your definition of the "calibration" in terms of raw channel multipliers ? because there is nothing else raw converter can get (except SSFs of course).... manufacturer discloses what will be the mulltipliers when sensor is "illuminated" by D65 (or the scene consisting of WB target is "illuminated" by D65)... so define

[32BT]

please be so kind - define what is "electronic state" of the sensor  ? and it given the word "state" you used please explain why'd that "state" can be not related to "scene lighting" "at the time of capture" ?

Electronic state of the sensor simply refers to the fact that the manufacturers can do all kinds of hocus pocus before the data is captured and stuffed into the raw file, which could include electronic adjustments to the sensor so it behaves differently but that we aren't aware of. Given the black box that the capture process really is for a raw data consumer, we simply want it to produce the single wp that corresponds to the state at time of capture.

This whitepoint is about the capture device or capture process. The capture process might change the whitepoint depending on scene lighting but quite obviously, it should not do so in the digital realm. Once the data is captured, they should refrain from further processing and simply communicate the original whitepoint of capture.

Can't explain it any better. Colorscience has a very clear definition of what whitepoint means for input devices. Obfuscating the entire idea with scene referred lighting or other profiling BS isn't typically helping get to a better procedure to actually control white balance (which was more or less what OP was asking for.)

[DP]

If we take the multipliers from your A7 as an example, and we deem equal R and B channel as the likely wp, then the CCT of the A7 is somewhere between 3200K and 6000K according to that list.

I am sorry - what does it mean "we deem equal R and B channel as the likely wp" ? all the pairs of multipliers above produce equal R & B channels  - that is how WB is defined - set of multipliers that make the data from all 3 channels equal, R * wbR = G * wbG = B * wbB (and wbG is assumed to be 1.0 usually in consumer level cameras give what kind CFA are used) ... do you actually mean some illuminant that
w/o using any multiplication produces equal saturation in R and B raw channels w/o any multiplication (so w/o any WB applied) ?  one of such illuminant is already present in the list above - it is "0x7826 WB RGB Levels Fluorescent     : 2228 1024 2256" with the good precision ... so well... we have what sensor "calibrated" for some Fluorescent spectrum then ?

It is that particular setting that we are interested in.

you are interested and you have that - for A7R2: "0x7826 WB RGB Levels Fluorescent: 2228 1024 2256", so what is the next move for a raw shot produced under tungsten incandescent bulb for example ?

Note also that the list provided in the raw data, is very odd from a procedural perspective: the camera obviously doesn't know which lighting it is capturing, and therefore it knows nothing about the temperature, or the required colorresponse.

camera does not know - but manufacturer does not know CFA (and their own typical lenses properties) properties and calculates and/or conducts experiments by illand provides the data that camera's firmware writes...

It could be capturing a sunset with low wp temperature, but requiring a daylight colorresponse. Or it could be capturing an incandescent scene with low temperature, requiring an incandescent colorresponse. The camera doesn't know, shouldn't care, and therefore, shouldn't communicate the info.

that is your personal opinion - the data above are as you can see for well defined cases and the setting that camera calculates as "AUTO" is of course a product of a guess - I'd rather have camera's guess included, because nobody is forcing me to use it.

You could potentially add maybe a spectrophotometer in a camera

I have Seconic C-7000, was a big waste of money ... I ended up not using it other then when creating camera profiles for my use with dcamprof.

[32BT]

please define what is "calibrated" ? as noted no consumer level cameras have CFA that yield anything close to  UniWB when illuminated by anything close to any daylight w/o using some magentish filters in lenses... hoi polloi like me understand "calibration" for a particular spectrum as getting all raw channel saturated equally when illuminated by (sensor that is, not the scene) that said spectrum... give me your definition of the "calibration" in terms of raw channel multipliers ? because there is nothing else raw converter can get (except SSFs of course).... manufacturer discloses what will be the mulltipliers when sensor is "illuminated" by D65 (or the scene consisting of WB target is "illuminated" by D65)... so define

That's exactly what I mean as well. It still doesn't mean the sensor whitepoint actually is D65, it simply means that the sensor would be calibrated to have a CCT that comes close.

I don't need the manufacturer to disclose all multipliers. I want them to tell me the whitepoint that they consider balanced. (Where the multipliers for R and B are equal for example). This may be different for different camera settings, where you could change the electronic state of the sensor under different conditions, but I don't think that is the case for any camera currently. Maybe in case of the old CCDs with precharging, or perhaps for different ISO settings it changes, but I wouldn't know. I don't want to know. I simply want the raw file to communicate the capture whitepoint.

[DP]

Electronic state of the sensor simply refers to the fact that the manufacturers can do all kinds of hocus pocus before the data is captured and stuffed into the raw file, which could include electronic adjustments to the sensor so it behaves differently but that we aren't aware of. Given the black box that the capture process really is for a raw data consumer, we simply want it to produce the single wp that corresponds to the state at time of capture.

still not clear what "Electronic state" exactly is

This whitepoint is about the capture device or capture process. The capture process might change the whitepoint depending on scene lighting but quite obviously, it should not do so in the digital realm. Once the data is captured, they should refrain from further processing and simply communicate the original whitepoint of capture.

what is the "original whitepoint of capture" ? you sound (by using the word "capture") like it might different for each shot then based on different illumination/different scene ? and yet above you were insisting that what you want does not depend on a specific illumination during the shot... what what do you really really want then ? is it something different potentially for each shot or is it something static for a given camera which is related to its CFA properties (like I suggested you want one of the infinite amount of different spectrums that will result in equal R & B raw channels saturation before applying WB) ?

Can't explain it any better.

if you can't then don't

Colorscience

says that WP depends on the illuminant ... so what do we do ? didn't you insist on some WP being indep. from that by introducing some "electronic state" ?

Obfuscating

the only obfuscation was from your side behind "Can't explain it any better." statement.

[DP]

"...hoi polloi like me understand "calibration" for a particular spectrum as getting all raw channel saturated equally when illuminated by (sensor that is, not the scene) that said spectrum...." -- that's exactly what I mean as well.

so illumination producing UniWB then (you saturate R & B not only equally between themselves but both equal to G channel) - and as I noted no real illumination will get that for consumer level cameras in real life situations, and using magenta filters (on light in studio or lenses in the wild) only helps to get close to that... but OK... now I still don't get an idea with one slider controlling 2 multipliers (we don't really need multiplier for the green channel for WB purposes) for R & B channels... also - may raw converted like RPP or as far as I recall Irididient developer (here I might be wrong) will allow you to dial in UniWB because they have controls for raw channel multipliers (for RPP there are no other controls - users always deal with raw channel multipliers) - so he were go - you start with what you want for a given shot... please tell us how do you want to manipulate in UI with 2 multipliers by one control now ?

[32BT]

you are interested and you have that - for A7R2: "0x7826 WB RGB Levels Fluorescent: 2228 1024 2256", so what is the next move for a raw shot produced under tungsten incandescent bulb for example ?

Exactly, that is the question we need to answer from the perspective of a reasonably educated user/photographer!

So, the next move imo would be:

1. User selects incandescent response (from a pop up with profile choices). This changes the ICC profile but not the multipliers.
2. User checks the color considered the colorcast (this should for example show the color of illuminant A relative to D50 initially)
3. User can adjust a slider to remove, increase or even invert the colorcast.




The "color considered colorcast" is the interesting part here. It could initially be the color of the illuminant that is part of the colorresponse profile relative to D50. So, incandescent will initially show a yellow color, Cloudy Daylight would for example show a blue color.

You can then set it via:
1. Temperature and tint
2. A colorwheel or colorpicker
3. derive it from the image (automatically or via the sampler).

So that still gives you all the options you have now, except we no longer need to fiddle with an apparently inconsistent Temperature and tint combo that influence each other and may even change the responseprofile under some definitions.


The procedure is then very clear to the user:

1. Select colorresponse depending on lighting conditions

2. Select the colorcast, e.g. by sampling a graycard

3. Reduce colorcast to taste...


Apply the same logic to the sunset case, and you immediately should see the advantage of this over Temperature & tint as some kind of control over whitebalance. In the latter case it operates completely counter intuitive because it is referring to a traditional indication of filmresponse.

[digitaldog]

Exactly, that is the question we need to answer from the perspective of a reasonably educated user/photographer!

So, the next move imo would be:

1. User selects incandescent response (from a pop up with profile choices). This changes the ICC profile but not the multipliers.
2. User checks the color considered the colorcast (this should for example show the color of illuminant A relative to D50 initially)
3. User can adjust a slider to remove, increase or even invert the colorcast.




The "color considered colorcast" is the interesting part here. It could initially be the color of the illuminant that is part of the colorresponse profile relative to D50. So, incandescent will initially show a yellow color, Cloudy Daylight would for example show a blue color.

You can then set it via:
1. Temperature and tint
2. A colorwheel or colorpicker
3. derive it from the image (automatically or via the sampler).

So that still gives you all the options you have now, except we no longer need to fiddle with an apparently inconsistent Temperature and tint combo that influence each other and may even change the responseprofile under some definitions.


The procedure is then very clear to the user:

1. Select colorresponse depending on lighting conditions

2. Select the colorcast, e.g. by sampling a graycard

3. Reduce colorcast to taste...


Apply the same logic to the sunset case, and you immediately should see the advantage of this over Temperature & tint as some kind of control over whitebalance. In the latter case it operates completely counter intuitive because it is referring to a traditional indication of filmresponse.

A lot of text to end up with where we started: move sliders to taste (your exact term above)!

"While intelligent people can often simplify the complex, a fool is more likely to complicate the simple."  -Gerald W. Grumet

[32BT]

still not clear what "Electronic state" exactly is

Like precharging in the days of CCDs. It may perhaps change the behavior of the separate channels and therefore the multipliers wouldn't be fixed. I presume no cmos camera currently does any analogue adjustments, but then, I don't know either. Nor do I know whether they preprocess some rawdata before it actually is filed. I simply call it the electronic state.

[32BT]

A lot of text to end up with where we started: move sliders to taste (your exact term above)!

"While intelligent people can often simplify the complex, a fool is more likely to complicate the simple."  -Gerald W. Grumet

Ha ha, are you trying to suggest I am a fool? In that case, just say so. I'm nearly grown up now and should be able to handle it. ;-)

[digitaldog]

Ha ha, are you trying to suggest I am a fool? In that case, just say so. I'm nearly grown up now and should be able to handle it. ;-)

I'm not suggesting anything. Other than using your own writings to illustrate we move sliders and the like for all kinds of subjective image rendering as so many others in this thread have stated, using EXISTING tools (specifically LR which IS on topic).
You want to propose using CTT numbers to do this instead, knock yourself out. You want to propose specto-enabled cameras people here can use, build one. Please.
You want to re-code LR to behave as you believe it should handle WB alone, get cracking.
You want to answer the OP's questions about WB in LR, PLEASE begin.
I'll leave you and others another quote, from a fellow who not only understood raw processing but aided actual photographers and our industry, the late Bruce Fraser, with respect to your ideas as yet aimed for the OP:

"You can do all sorts of things that are fiendishly clever, then fall in love with them because they're fiendishly clever, while overlooking the fact that they take a great deal more work to obtain results that stupid people get in half the time. As someone who has created a lot of fiendishly clever but ultimately useless techniques in his day, I'd say this sounds like an example."

[32BT]

I see.

I veered off topic.

My apologies.

In my enthusiasm I got carried away.

[Jack Hogan]

One of the biggest challenges I have doing sunset photography is white balance.

As mentioned, the main issue with sunsets is that one of the main tenets of color science, Color Constancy, tends to fail at low CCTemperatures.  Color Constancy is a perceptual feature of the Human Visual System that makes white look white whether the illuminant is D45 or D75.  With a digital capture, white balance is one easy way to achieve an approximately similar compensation to what goes on in our brains when constancy is in effect.  We effectively normalize all tones so that a particular neural (say that of your gray card) is in fact neutral (e.g. 100,100,100).

However, while that works fairly well in the middle of the range it doesn't near the extremes: for instance every kid knows that white actually looks yellowish to us in the low Ks under, say, candle light.  So in those situations we don't expect a neutral white. If the rendered image is white balanced assuming color constancy, the result often looks unnatural to us, too cold.  It doesn't help that most profiles treat anything below Tungsten as if it were Tungsten and viewing conditions are typically very, very different from taking conditions.

So until someone comes up with a better way of modeling our perception of color in the low temperatures, we are stuck having to creatively/manually modify white balance in conditions near the edges of the Temperature range like sunsets until a good/pleasing result is achieved.  Often that just means increasing the Temperature slider a bit from what a gray card would suggest, as was done below.

Jack

[Guillermo Luijk]

Given how easy is to fool human vision, specially regarding colour tints, trying to find one single correct WB in a sunset is impossible. There is a whole range of correct WB's, the best will be the one you prefer, and you can even change your mind the following day. In brief: your question doesn't have a precise answer.


You'll probably see the furniture in the red square as brown, but it's pure gray.


You'll probably not see the skin of the guy as gray, but it is.

Accurate colour rendering, which is needed in other WB applications such as product, reproduction of pieces of art, costumes,... is a much simpler story. Or at least a story with an accurate answer.

Regards

PS:

[fdisilvestro]

Given how easy is to fool human vision, specially regarding colour tints, trying to find one single correct WB in a sunset is impossible. There is a whole range of correct WB's, the best will be the one you prefer, and you can even change your mind the following day. In brief: your question doesn't have a precise answer.


You'll probably see the furniture in the red square as brown, but it's pure gray.


You'll probably not see the skin of the guy as gray, but it is.

Accurate colour rendering, which is needed in other WB applications such as product, reproduction of pieces of art, costumes,... is a much simpler story. Or at least a story with an accurate answer.

Regards

PS:

The beauty of simultaneous contrast. Is it just easy to fool the human vision or is it for a reason that evolution gave us that?

[D Fuller]

One of the biggest challenges I have doing sunset photography is white balance. To be able to work consistently, I almost always have my camera set to "daylight" (exception will be indoors) - there's no point in using the camera's histograms to compare exposures if the WB is set to "auto" and thus potentially different for each shot!

One big problem here is "memory". I can't remember exactly what it looked like.

Using LR, the choices are:

* 2000K, Tint -24 (WB from clouds)
* 2850K, Tint 0 (auto)
* 4350K, Tint -47 (WB from yachts)
* 5500K, tint +10 (Daylight)

... all of the other "develop" settings are 0.

White Balance, when shooting sunsets, is seldom going to work by picking it from the scene. Here’s why: the “White Balance” picker is a way to measure the color of light falling on a neutral object, and eliminate the color cast so that the object looks neutral. But that is not what you want in a sunset shot. The colors of sunset are changes in the color of the light, not the objects (clouds, yachts, etc.) the lights fall on. You don’t want the yachts to appear white, you want them to reflect the color of the light that’s falling on them from the sunset. So you do not want to balance that color out.

What you are doing already by shooting Daylight Balance is the best starting point I know for reproducing the color of a sunset. It sets the raw developer up to show you how different the sunset light is from midday sun, and that’s exactly what you want. From there, adjust the sliders to taste.