WB Shift unravelled - An experimental investigation

[Rightsaidfred]

Hi all

So I always wanted to understand WB shift.

Why?

Number one, many jpeg recipes use a special WB shift setting.

Number two, if I do a custom WB with a white card, colors are each and every time too warm for my taste. I want to cool down colors, and the WB shift function is the tool of choice for this. But I never really understood the logic behind WB shift. Buzzwords are color temperature (which is well defined) and 'tint' which I did not really understand.

A recent thread eventually ignited the spark in me to invest a little time into this First I did a bit of web search and posted my conclusion there as a reply. Now I did an experiment. Heck - I was not fully right, and I want to present a better version here for discussion.

The most comprehensive non-professional explanation I could find is a discussion on DPReview, 'What is 'tint' in white balance? '.

The WB shift selector consists of two axes, a yellow/amber...blue axis ('B') and a green...red/magenta axis ('R').

Now comes my experiment.

My target was to mimic a change of the color temperature setting (Kelvin, K) with the WB shift function.

[ATTACH alt="The normal image (of course not a great image - only for testing purpose), shot in daylight on a cloudy day with my trusty X-T20, XF 23/2, f/4, WB set at 6300 K, no WB shift selected. You see here the "Waldkirche" in Planegg, Germany, a fascinating church building established in 1925/26."]3206492[/ATTACH]
The normal image (of course not a great image - only for testing purpose), shot in daylight on a cloudy day with my trusty X-T20, XF 23/2, f/4, WB set at 6300 K, no WB shift selected. You see here the "Waldkirche" in Planegg, Germany, a fascinating church building established in 1925/26.


Developed with X-Raw Studio from the same raw, WB set at 10000 K (the maximum my X-T20 allows for), no WB shift selected. Of course, all the colors get warmer.


Developed with X-Raw Studio from the same raw, WB set at 6300 K - and I tried to mimic 10000 K with the WB shift. What I found the best fit is R7 B-8.


Developed with X-Raw Studio from the same raw, WB set at 3700 K, no WB shift selected. Of course, we get cooler colors all over.


Developed with X-Raw Studio from the samw raw, WB set at 6300 K - and I tried to mimic the 3700 K colors with the WB shift function. Best fit was R-7 B9.

The fits are not perfect but pretty good. I also played with values in between to confirm (5000 K and 8200 K, photos not included here). And I also confirmed my findings with another, very differently lit scene (artificial light, not included here). The results are consistent.

My conclusion

So what I conclude and am putting up for discussion here is that a color temperature (K) setting change is not equivalent with simply wandering along the vertical B axis in the Fujifilm WB shift selector.

In fact, a lower color temperature setting (to get cooler colors) can be mimicked with the WB shift selector by decreasing reds and increasing blues at the same time by similar amounts.

Vice versa, higher color temperatures (to get warmer colors) can be mimicked by increasing reads and decreasing blues by similar amounts at the same time.

And every other R/B combination in the WB Shift pattern additionally changes the 'tint' of the image. So the most extreme 'tint' shifts are top right and bottom left in the WB shift selector pattern.


My conclusion: a change of the color temperature (Kelvin) setting is approximately equivalent to wandering along the white line drawn in the WB shift selector. The area outside the white line can be regarded as an additional 'tint' change.

Now I am curious about other opinions and findings

Note: It is well a known fact that the WB shift setting can be stored with custom presets only with the newer camera models.

Regards,

Martin

--
SmugMug - https://martinlang.smugmug.com
500px - https://500px.com/martinlangphotography
Insta - https://www.instagram.com/martin.lang.photography

 

[Yawlen Z]

Largely if not entirely correct I would think. And still important to stress that similar doesn't mean equal. I had bookmarked this thread that you link to before, and will reread parts of it now while distilling the wording to a more expansive reply – suffixed by a question. Anyway, ever since I got the hang of doing this shifting quick when poking a lens out, I didn't bother much with custom wb anylonger, or grey/white cards and such totally valid means. Currently, (as long as it's a good way to go) I like to pick a Kelvin value to the cool side of fine day, and warm that back on a shift. Or towards rendering 'muddy' interior ambience. And over time you do develop an instinct to set these incremental notches, and judge them in review.

 

[Erik Baumgartner]

As a RAW only shooter, I don't worry too much about getting the WB perfect in-camera, "in the ballpark" is generally good enough as I'll typically tweak it a bit in post anyway. I find Fuji's AWB to usually be pretty reliable except in "the woods" where it consistently skews a bit cool and very magenta (easy to fix with a RAW file). I hear the new X-H2 has A.I. assisted AWB, let's hope that works better. I'd hate to be a lpeg shooter hassling with the WB in the woods. I almost never set a "fixed" or preset WB, or use WB Shift, but a jpeg shooter would probably want to (WB is one of the primary reasons I shoot RAW).

I should mention for those unaware, a WB Shift setting won't affect the "as shot" AWB metadata in the RAW file with the older cameras - it will in my newer X100V. but not the older X-T2/20. So for those who want to bias the AWB one way or the other (Fujis AWB can run a bit cool, IMO), it won't affect the RAW files in some cameras.

In situations where I really want the color to be accurate (especially in difficult mixed lighting), I'll occasionally a put a neutral card in the frame and take a few test shots with the card positioned under the different light sources so I'll have easier time sorting out the color in post, but I usually just wing it.

As the Temp/Tint values in Lightroom/ACR (and I assume, other editors) don't actually correspond to actual Kelvin values and can vary significantly from one camera model to another (very different color with the same Temp/Tint values), I star with the "as shot" AWB values and go from there, or use the WB dropper on something neutral in the frame. I have a good idea where the Temp/Tint settings ought to be in most light (Tint is almost never at zero, even in in good natural light). As I usually shoot multiple frames in the same light, the AWB will usually look 'just right" on one of them, and I can then just paste those Temp/Tint values to all the other images shot in the same light.

 

[Yawlen Z]

I'll jot the most pressing part of my question right here now then. Did you manage to acquire any exact clue as to how an AWB measurement goes, off the sensor plane? What lightquality sampling method, what computation? I've never been able to spot a decent explainer on that.
And would the camera's wb-choice reflect and incorporate a 'shifted' quality in the environmental lighting it just measured, as opposed to merely moving up/down the Kelvin range?

As I don't always do this fussing and tinkering so perfectly well, of course I'll record raw files too. It's more I'd presume that most everyone would want the best reference jpeg you can get to right away – as that one is really about the desired render – and swiftly compare that to pre/after-shot perception.

 

[Mark S Abeln]

I'll jot the most pressing part of my question right here now then. Did you manage to acquire any exact clue as to how an AWB measurement goes, off the sensor plane? What lightquality sampling method, what computation? I've never been able to spot a decent explainer on that.

Automatic white balance is estimated from the captured sensor data, and there are a multitude of various algorithms that have been developed.

This actually is one of the earliest problems in artificial intelligence, dating from the late 1950s and early 1960, as an explanation of the physiological phenomenon of ”color constancy”.

One early method is the “gray world” algorithm which basically just averages all of the colors in an image and then subtracts that color from the entire image. It usually doesn’t work very well.

Another early algorithm, “white world”, is to select the brightest patch of color in an image and subtract that color from the entire image. Again, this sometimes works but not always.

Another method takes the brightest bits of each color channel and equalizes those values: this sometimes works very well, especially if the darkest parts of each channel also are equalized, which is commonly found in modern software as “auto levels”.

A more subtle method, “color by correlation”, often found in modern apps as the automatic white balance, is to only consider a restricted set of possible illuminants, and select the one which gives the most plausible results. This method usually gives good if not excellent results.

One researcher described automatic white balance as an underdetermined problem, insolvable in the general case. This technical article explains some of the problems:

Choose Identity Provider

courses.cs.washington.edu

 

[Erik Baumgartner]

Thanks for the information, very interesting. I didn’t know any of those particulars. Do you know what our phones are doing? The AWB in my iPhone is significantly better than with any camera that I’ve ever owned.

 

[bastibe]

Thanks for the information, very interesting. I didn’t know any of those particulars. Do you know what our phones are doing? The AWB in my iPhone is significantly better than with any camera that I’ve ever owned.

Your phone varies its white balance across the frame. Shadows get warmed, sunlight is made cooler, even artificial lighting is compensated somewhat. Depending on the phone, this actually breaks down rather frequently. But at first glance, it corresponds better to our perception of constant colors than a uniform white balance does.

 

[Rightsaidfred]

As a RAW only shooter, I don't worry too much about getting the WB perfect in-camera, "in the ballpark" is generally good enough as I'll typically tweak it a bit in post anyway.

Fair point.

I am a jpeg and raw shooter. If possible, I avoid post processing. If not, I do it mostly with darktable on my Linux machine (see My darktable workflow to provide a decent baseline // Updated (darktable 4.0)).

WB is the setting that I personally find most difficult to optimize in my raw development workflow. I have a large and excellent monitor - but I always get tired eyes quickly when trying to optimize WB with a raw image from a bad starting point. I then somehow loose the orientation of what is a good reference, what appears most natural to the human eye. What I once found particularly bad was portrairs in front of some cheap Christmas tree LED illumination - I just could not get it right afterwards (well, certainly these LEDs had a silly color spectrum), it also was high ISO, bottom line, I wasted a lot of time for nothing. Probably a white card would have helped.

In short, it is hard and stressful for me to estimate the best WB setting parameters for a given image if the starting point is bad. So having as good as possible a WB starting point is the best I can ever do to facilitate my post processing.

I find Fuji's AWB to usually be pretty reliable

Yes I also find it really good. It matches my personal taste relatively well.

except in "the woods" where it consistently skews a bit cool and very magenta (easy to fix with a RAW file).

Yes. My last series was in the woods. Since I know this is difficult in terms of WB, this time, I took a white card with me and cooled down the colors with the WB shift function in camera (X-T4). Very happy with the outcome.

And, Erik, 'easy' is relative, depends on practice and skills. You certainly have both

I hear the new X-H2 has A.I. assisted AWB, let's hope that works better.

Ah interesting.

I'd hate to be a lpeg shooter hassling with the WB in the woods. I almost never set a "fixed" or preset WB, or use WB Shift, but a jpeg shooter would probably want to (WB is one of the primary reasons I shoot RAW).

See above.

I should mention for those unaware, a WB Shift setting won't affect the "as shot" AWB metadata in the RAW file with the older cameras - it will in my newer X100V. but not the older X-T2/20.

That's good to know! Wasn't aware of this behavior of the older cameras, although I own one, next to my X-T4.

So for those who want to bias the AWB one way or the other (Fujis AWB can run a bit cool, IMO),

Haha I like it cool No kidding. I said, custom WB always produces too warm colors for my taste. Adjusting this is one of my applications for WB shift.

...

In situations where I really want the color to be accurate (especially in difficult mixed lighting), I'll occasionally a put a neutral card in the frame and take a few test shots with the card positioned under the different light sources so I'll have easier time sorting out the color in post, but I usually just wing it.

Yes but it is effort, it is hassle, it takes time, you need to think of it, ... Who really does it with family photos?

...I star with the "as shot" AWB values and go from there, or use the WB dropper on something neutral in the frame.

....if there is....

Often, there is no neutral white/gray area in an image to pick.

I have a good idea where the Temp/Tint settings ought to be in most light (Tint is almost never at zero, even in in good natural light).

So I guess, AWB will not only change the color temperature but also adjust the tint. But why do we often only speak about Kelvin when it comes to WB in photography? I guess because the topic is so difficult to understand, and there are different buzzwords and parameters (color temperature, tint, hue, chroma, etc.).

As I usually shoot multiple frames in the same light, the AWB will usually look 'just right" on one of them, and I can then just paste those Temp/Tint values to all the other images shot in the same light.

I do the same - a professional photo series must have all the same look. I take the best one, maybe adapt, and copy/paste with darktable (modern workflow, scene referred). Here again - the better the starting point the easier.

In my last series, as I mentioned above, I used a custom WB (white card) and adjusted towards cooler temperatures with the WB shift function (X-T4). I did not fully understand what I did, I just slid up the B axis quite a bit, but it was not bad and I could simply take it over for my raw development. Since it was manual WB, I did not even have to copy/paste - it was the same in the whole series. Just need to think about it when changing the surrounding.

Based upon my little experiment yesterday, I understand the WB shift function at least somewhat better. Today I know that with simply increasing the blues, I not only reduced the color temperature of my series, but I additionally changed the tint since my adjustment did not follow the "white line". For good or for bad - anyways, I like the outcome. For me, it is always important to understand and manage what I'm doing.

Regards,

Martin

--
SmugMug - https://martinlang.smugmug.com
500px - https://500px.com/martinlangphotography
Insta - https://www.instagram.com/martin.lang.photography

 

[Rightsaidfred]

... (snipped)

Very interesting considerations and interesting read.

Thanks!

Cheers,

Martin

 

[Barbarus]

I think it's worth adding that Fuji's AutoWB algorithms seem to be unusually reliable. I'm a raw shooter too after Nikon DSLRs ruined too many of my artificial light JPGs, but with Fuji, it's fairly rare that I need to tweak white balance, let alone worry about custom WB while shooting.

 

[Jack Hogan]

Nice experiment Rightsaidfred, it would have been nice to see a gray card in there.

I am not familiar with X-Raw Studio nor its WB Shift function but based on your examples it looks like it modifies the white balance multipliers (if not ignore this post). If so you may be able to get some tones to match by playing with Shift at the wrong Correlated Color Temperature - but only at the expense of many others.

The reason is that the multipliers are an important first part of the overall transform from raw to output color space data - but they are followed by several other components of the transform that are a CCT dependent compromise: crucially and typically white-balanced-raw to XYZ followed by Chromatic Adaptation.

Good raw converters (manufacturers' are typically good) normally have several built-in transforms fine tuned to provide the best overall compromise at the estimated CCT. They are good over increasingly wide ranges (i.e. say +/-10% of CCT).

When shifting white balance multipliers at the wrong CCT, say 3700K when in fact it is closer to D65, one can perhaps get an acceptable compromise in matching the chosen dominant tones - but the Raw Converter will unknowingly select the incorrect XYZ and CA transforms, those for 3700K as opposed to those for D65. The overall compromise color transform may end up producing the eyeballed tones decently but at the expense of many others.

This is true even when the white balance multipliers are 'perfect', taken off a gray card. If the multipliers are perfect but the CCT is materially incorrect, the Raw Converter will use the wrong transform potentially producing suboptimal tone compromises, what you may describe as too warm in the OP. That's why in critical situations it is important to have a good estimate of CCT, as better described in the series of articles around here

https://www.strollswithmydog.com/on-white-balance/

Many photographers learn this intuitively early on, but that's why we do it.

Jack

 

[sir_c]

Especially with those cheap modern LEDs with their funky color spectrums it is hard to nail it. What often would be the 'mathematically correct' WB is often the least pleasing to look at. The bluish ones properly mess up skin tones and/or the ambient lighting.

 

[Yawlen Z]

Hooray for the puzzling intricacy of these research papers and blogposts. I may need to hop offline awhile to digest it all, but somehow paragraph headlines like "Roll Your Own Transform" and "Choose Your Poison" resound right up my alley.

Just to chime in, another science afflicted link that I found in the 'Sensors' cubicle at Swiss MDPI – on a subject real relevant to some. And OpenAccess at that. There's no mention of X-Trans in there, but hey Rumour says that in the smartphone camera biz there's a pack of CFAs running wild. With haywire RAW fur on them.

https://www.mdpi.com/1424-8220/19/14/3215

 

[Rightsaidfred]

Nice experiment Rightsaidfred, it would have been nice to see a gray card in there.

That's an absolutely fair point.

From this discussion I learned to more often place a gray or white card into one image in a series. Particularly when it comes to shooting in a forest or artifical light or other challenging conditions.

Not only for custom WB - really take photos with the card placed in the scene to have this information for post processing.

I have a Novoflex white card but might get the genuine WhiBal pocket card you mentioned in your article, the small one.

A color checker is the even more professional solution but not sure whether I'll accept the effort. I am only a hobbyist.

I am not familiar with X-Raw Studio

This is just operating a Fujifilm camera from a computer. The camera needs to be connected and transforms the raws.

nor its WB Shift function

Think every manufacturer offers such a function.

I wanted to better understand what it does.

but based on your examples it looks like it modifies the white balance multipliers (if not ignore this post).

Unable to confirm or contradict.

If so you may be able to get some tones to match by playing with Shift at the wrong Correlated Color Temperature - but only at the expense of many others.

The reason is that the multipliers are an important first part of the overall transform from raw to output color space data - but they are followed by several other components of the transform that are a CCT dependent compromise: crucially and typically white-balanced-raw to XYZ followed by Chromatic Adaptation.

Good raw converters (manufacturers' are typically good) normally have several built-in transforms fine tuned to provide the best overall compromise at the estimated CCT. They are good over increasingly wide ranges (i.e. say +/-10% of CCT).

I use darktable for post processing.

Sometimes in camera raw development (directly or by means of X-Raw Studio).

When shifting white balance multipliers at the wrong CCT, say 3700K when in fact it is closer to D65, one can perhaps get an acceptable compromise in matching the chosen dominant tones - but the Raw Converter will unknowingly select the incorrect XYZ and CA transforms, those for 3700K as opposed to those for D65. The overall compromise color transform may end up producing the eyeballed tones decently but at the expense of many others.

This is true even when the white balance multipliers are 'perfect', taken off a gray card. If the multipliers are perfect but the CCT is materially incorrect, the Raw Converter will use the wrong transform potentially producing suboptimal tone compromises, what you may describe as too warm in the OP. That's why in critical situations it is important to have a good estimate of CCT, as better described in the series of articles around here

https://www.strollswithmydog.com/on-white-balance/

Thank you very much. Great read. Need to read it multiple times

Many photographers learn this intuitively early on, but that's why we do it.

Jack

Thanks again Jack.

Regards,

Martin

--
SmugMug - https://martinlang.smugmug.com
500px - https://500px.com/martinlangphotography
Insta - https://www.instagram.com/martin.lang.photography

 

[Easy Lee]

My target was to mimic a change of the color temperature setting (Kelvin, K) with the WB shift function.

[ATTACH alt="The normal image (of course not a great image - only for testing purpose), shot in daylight on a cloudy day with my trusty X-T20, XF 23/2, f/4, WB set at 6300 K, no WB shift selected. You see here the "Waldkirche" in Planegg, Germany, a fascinating church building established in 1925/26."]3206492[/ATTACH]
The normal image (of course not a great image - only for testing purpose), shot in daylight on a cloudy day with my trusty X-T20, XF 23/2, f/4, WB set at 6300 K, no WB shift selected. You see here the "Waldkirche" in Planegg, Germany, a fascinating church building established in 1925/26.


Developed with X-Raw Studio from the same raw, WB set at 10000 K (the maximum my X-T20 allows for), no WB shift selected. Of course, all the colors get warmer.


Developed with X-Raw Studio from the same raw, WB set at 6300 K - and I tried to mimic 10000 K with the WB shift. What I found the best fit is R7 B-8.


Developed with X-Raw Studio from the same raw, WB set at 3700 K, no WB shift selected. Of course, we get cooler colors all over.


Developed with X-Raw Studio from the samw raw, WB set at 6300 K - and I tried to mimic the 3700 K colors with the WB shift function. Best fit was R-7 B9.

The fits are not perfect but pretty good. I also played with values in between to confirm (5000 K and 8200 K, photos not included here). And I also confirmed my findings with another, very differently lit scene (artificial light, not included here). The results are consistent.

My conclusion

So what I conclude and am putting up for discussion here is that a color temperature (K) setting change is not equivalent with simply wandering along the vertical B axis in the Fujifilm WB shift selector.

In fact, a lower color temperature setting (to get cooler colors) can be mimicked with the WB shift selector by decreasing reds and increasing blues at the same time by similar amounts.

Vice versa, higher color temperatures (to get warmer colors) can be mimicked by increasing reads and decreasing blues by similar amounts at the same time.

And every other R/B combination in the WB Shift pattern additionally changes the 'tint' of the image. So the most extreme 'tint' shifts are top right and bottom left in the WB shift selector pattern.


My conclusion: a change of the color temperature (Kelvin) setting is approximately equivalent to wandering along the white line drawn in the WB shift selector. The area outside the white line can be regarded as an additional 'tint' change.

Sorry to add to this old thread ... but I was just going to start a new thread about the exact same topic. So I found this one and thought it made more sense to add my thoughts here.

I think the OPs key finding has not been properly discussed here. It is that Fujifilms white balance matrix is just about 45° CCW shifted from most other WB matrixes such as in Lightroom or Capture One or in my Canon camera.

In most cases you get a Kelvin (or "Temperature") axis which you can adjust to make your image look warmer or cooler (usually presented as yellow-blue axis) and a "Tint" axis to add or remove a green or magenta color cast. So it would be intuitive to think that in Fujifilm cameras the blue-yellow axis in the white balance tool corresponds to color temperature K and the green-red axis corresponds to tint.

This is what I thought for a long time. But since I've started working with Capture One I realised this is not the case by looking at the way C1 translates Fujis WB values. So I did this little experiment which brought the same conclusion that the OP also reached. I took a few indoor shots at K=3000 with varying WB shifts and looked at how C1 translated this into its own Kelvin and Tint values.

K3000, red 0, blue 0 -> 3143 Kelvin, tint -0.9
K3000, red +1, blue 0 -> 3197 Kelvin, tint 0.8
K3000, red +2, blue 0 -> 3265 Kelvin, tint 2.9

So increasing the red in the Fuji camera not only increases tint in C1 but also color temperature.

K3000, red 0, blue -1 -> 3194 Kelvin, tint -2.3
K3000, red 0, blue -2 -> 3254 Kelvin, tint -3.8

Decreasing the blue in Fuji does not only increase color temperatur in C1 as expected but also shifts tint towards green.

K3000, red+2, blue -2 -> 3385 Kelvin, tint -0.1

Increasing red and decreasing blue at equal amounts leads to an increase in color temperature in C1 with no color tint.

I wonder why Fujifilm has chosen this kind of white balance matrix. I think it is somewhat misleading. What looks like a yellow shift in Fujis WB tool is actually more of a green-yellow shift.

 

[Yawlen Z]

A useful exercise this, thanks for that. Even if I won't be dipping into CaptureOne software for a while, it's nice to get an idea on how it goes about the "translation" of this sort of thing. As the Fuji shift matrix graph is the first one that I learned to work with, and I don't pack a whole lot of experience with other makes, that still is my preferred way to handle wb shifts. Where the colors that I'd address as "mixed" ones are corner targets, graphically. Most of any individual's preference appears to stem from an acquired viewpoint difference in colorspace perception, maybe?