White Balance variations with different RAW converters

[Manfred M]

Manfred I know where you are coming from but I don't think you can blame human vision. It is more a result of convention and terminology. I assume you have done a bit of welding etc and your vision enabled you to easily appreciate that a white hot plasma cutter was a little warmer than the dull red used when brazing.

Not unusual for terms and words to have different meanings in every day conversation compared to it's specific meaning when used in a specialist area such as law, medicine, science etc.

We covered this subject briefly (and not totally satisfactorily) during a colour correction course I took last winter at a local community college. The prof spoke about the "physics" of colour temperature and how that differs from the psychological view that we have about associating the red / yellow end of the spectrum with warmth and the violet / blue end with coolness.

The upshot of it all was that we need to understand both the black body radiation concept of colour temperature so that we can produce an colour corrected image and then possible play with that correction to impact the viewer's feelings about the image. Adding a tiny bit of a red / yellow bias in a portrait usually results in a more pleasing image. Adding slightly more blue to a fall scene suggests approaching winter.

[Manfred M]

Manfred,
I think it was that answer that didn't satisfy me. And still doesn't.

1. Physics: Colour temperature = the temperature of a black body emitter.

2. Psychology: Red / yellow overtones suggests warmth ; blue overtones suggest coolness.

No one ever suggested our brains will process these two interrelated concepts identically.

[george013]

1. Physics: Colour temperature = the temperature of a black body emitter.

2. Psychology: Red / yellow overtones suggests warmth ; blue overtones suggest coolness.

No one ever suggested our brains will process these two interrelated concepts identically.

Manfred,
There is an expression in Dutch and I found it in English: white-hot. As Paul mentioned, hotter than red. or livid.
I don't care about the psychology. You may call it whatever you want. But a fact is that the hot and cold site in colortemperature are oppositional in what happens in photography.

I'm pretty sure now the colortemperature as used in photography is a corrective temperature. A lightsource of 3000K needs to be corrected to 6000K, or something like that. Less blue and/or more red. Or the other way, To much correction gives a more red feeling.

It still leaves the question unanswered why different converters give such a different color when using a fixed colortemperature.

George

[Manfred M]

Manfred,
There is an expression in Dutch and I found it in English: white-hot. As Paul mentioned, hotter than red. or livid.
I don't care about the psychology. You may call it whatever you want. But a fact is that the hot and cold site in colortemperature are oppositional in what happens in photography.

I'm pretty sure now the colortemperature as used in photography is a corrective temperature. A lightsource of 3000K needs to be corrected to 6000K, or something like that. Less blue and/or more red. Or the other way, To much correction gives a more red feeling.

It still leaves the question unanswered why different converters give such a different color when using a fixed colortemperature.

George

George -the reason that they are different is that they use different algorithms to calculate the appropriate colour temperature. I did a more controlled test than you did using three well regarded raw converters.

Here is the test shot:



I used a Paul C Buff Einstein 640 studio flash, with a strip box diffuser. This flash is rated to fire at 5600K ±50K

I manually metered the light with a Sekonic L-358 flash meter. The important number that you can read on the meter is where shows 100%; this means that 100% of the light being metered (and recorded) comes from the flash and there is NO ambient light influence in the image.

Camera settings: Colour temperature set to "flash", ISO 100, 1/125th sec at f/11. The ColorChecker Passport is included in the image as a baseline reference as well as giving me a proper colour correction target. In the test I got a colour temperature reading of the raw file that was opened by the software and then I used the color sampler tool to fine tune the white balance.

1. DxO Optics Pro 10 - reported colour temperature of 5996K as shot and as 5179K after correction.

2. Adobe Camera Raw - reported colour temperature of 6150K as shot and as 5000K after correction.

3. Phase One Capture One 9 - reported colour temperature of 6817K as shot and as 5371K after correction.

So as I wrote before, the different raw converters use different algorithms to calculate the colour temperature of the image and would therefore be expects to produce slightly different results.

By the way, we say "red hot" and "white hot" in English.

[george013]

George -the reason that they are different is that they use different algorithms to calculate the appropriate colour temperature. I did a more controlled test than you did using three well regarded raw converters.

Here is the test shot:



I used a Paul C Buff Einstein 640 studio flash, with a strip box diffuser. This flash is rated to fire at 5600K ±50K

I manually metered the light with a Sekonic L-358 flash meter. The important number that you can read on the meter is where shows 100%; this means that 100% of the light being metered (and recorded) comes from the flash and there is NO ambient light influence in the image.

Camera settings: Colour temperature set to "flash", ISO 100, 1/125th sec at f/11. The ColorChecker Passport is included in the image as a baseline reference as well as giving me a proper colour correction target. In the test I got a colour temperature reading of the raw file that was opened by the software and then I used the color sampler tool to fine tune the white balance.

1. DxO Optics Pro 10 - reported colour temperature of 5996K as shot and as 5179K after correction.

2. Adobe Camera Raw - reported colour temperature of 6150K as shot and as 5000K after correction.

3. Phase One Caoture One 9 - reported colour temperature of 6817K as shot and as 5371K after correction.

So as I wrote before, the different raw converters use different algorithms to calculate the colour temperature of the image and would therefore be expects to produce slightly different results.

By the way, we say "red hot" and "white hot" in English.

You missed what I was doing. I didn't let the converter calculate a temperature and its interpretation of the image. I just said to the converter: go to 6000K. If you let the converter do the job, it will have to find a reference point first. And that won't be the same.

Let me put it different. If your converter can give a reported color temperature of 6817K and after correction of 5371K, then I think those figures are mathematical based.

George

[revi]

In fact, the raw converter has only one way to change the white balance for an image:
change multipliers for 2 of red, blue, and green, relative to the third.
All other ways of indicating a colour temperature will have to be converted to this system...
(and that works, as in all cases you have a system with two relative coordinates:
reed/green + blue/gree, K + tint, ...)

The problem for a generic program is that the filters used for the red, green and blue channels
can have different absorbtion characteristics for different cameras, so a given standard white
balance would translate to different multiplier for red and blue (wrt. green, fixed at 1).
Iirc, the EXIF data contain a correction matrix for this, but that might not be sufficient (or that
matrix is in the makerdata part, which is rarely documented). So differences between converters
aren't all that surprising.

What surprises me in Manfred's experiment, is that the three converters report different colour
temperatures "as shot". Unless Manfred set the program to use its own automatic white balance
routine, not using the camera value (recorded in EXIF).

And as already noted, in George's initial series of images, there are other differences than just
white balance (as evidenced by the clouds, which have a rather different contrast over the series).

I know that at least Darktable (a converter with only Linux and MacOS versions) applies a tone
curve to the image after demosaicing, which is rather different depending on the camera brand
(default is to pick the brand shown in EXIF, but you can select any of the predefined curves, or
roll your own). So that's another variable in the process with a rather large effect on the end
result.

[george013]

In fact, the raw converter has only one way to change the white balance for an image:
change multipliers for 2 of red, blue, and green, relative to the third.
All other ways of indicating a colour temperature will have to be converted to this system...
(and that works, as in all cases you have a system with two relative coordinates:
reed/green + blue/gree, K + tint, ...)

The problem for a generic program is that the filters used for the red, green and blue channels
can have different absorbtion characteristics for different cameras, so a given standard white
balance would translate to different multiplier for red and blue (wrt. green, fixed at 1).
Iirc, the EXIF data contain a correction matrix for this, but that might not be sufficient (or that
matrix is in the makerdata part, which is rarely documented). So differences between converters
aren't all that surprising.

What surprises me in Manfred's experiment, is that the three converters report different colour
temperatures "as shot". Unless Manfred set the program to use its own automatic white balance
routine, not using the camera value (recorded in EXIF).

And as already noted, in George's initial series of images, there are other differences than just
white balance (as evidenced by the clouds, which have a rather different contrast over the series).

I know that at least Darktable (a converter with only Linux and MacOS versions) applies a tone
curve to the image after demosaicing, which is rather different depending on the camera brand
(default is to pick the brand shown in EXIF, but you can select any of the predefined curves, or
roll your own). So that's another variable in the process with a rather large effect on the end
result.

At the moment I can only speak of CaptureNx. Setting the WB by colortemperature only changes the R and B channel, the G channel stays as it is. Setting the WB by graypoint also let the G channel untuoched.
Using the gray eydropper does average the 3 channels.

Can you tell me how the colortemperature is calculated? Knowing that would bring me much further.


George

[Manfred M]

What surprises me in Manfred's experiment, is that the three converters report different colour
temperatures "as shot". Unless Manfred set the program to use its own automatic white balance
routine, not using the camera value (recorded in EXIF).

It surprises me a lot too. I used the flash colour temperature preset on the camera, which is 5400K, according to the camera manual. This is fairly close to the 5600K rating of the flash.

I'm running an upgrade on the CC suite right now, so can't open the Adobe software right now to check a few things.

[pnodrog]

It surprises me a lot too. I used the flash colour temperature preset on the camera, which is 5400K, according to the camera manual. This is fairly close to the 5600K rating of the flash.

I'm running an upgrade on the CC suite right now, so can't open the Adobe software right now to check a few things.

In a thread some time ago (I think it was Inkanyezi - Urban Domeij) pointed out and I thoroughly agreed that a camera can not determine the colour temperature accurately by evaluating a scene. It is the incident and ambient lighting that needs to be evaluated. The same goes for software trying to guess (yes guess) what the colour temperature was at the time the photograph was taken in an attempt to make a WB correction. It is not surprising that the various algorithms the software uses produces different results.

P.S. Yes it was Inkanyezi and THIS is one of several threads that have discussed it. Some are on WB there are other threads on colour temperature.

[Manfred M]

In a thread some time ago (I think it was Inkanyezi - Urban Domeij) pointed out and I thoroughly agreed that a camera can not determine the colour temperature accurately by evaluating a scene. It is the incident and ambient lighting that needs to be evaluated. The same goes for software trying to guess (yes guess) what the colour temperature was at the time the photograph was taken in an attempt to make a WB correction. It is not surprising that the various algorithms the software uses produces different results.

P.S. Yes it was Inkanyezi and THIS is one of several threads that have discussed it. Some are on WB there are other threads on colour temperature.

If you read both my and George's posts, we are not dealing with AWB at all and neither of us were using that setting; but are using presets. So while I understand and generally agree with what people wrote in the thread you have linked to, I don't think the link is particularly relevant to the issue being discussed.

In my case I set the camera WB to "flash" setting which is 5400K on the D800. I shot a flash that is rated at 5600K ±50K and ensured that ambient light would not impact the shot, so while not a perfect match, in the scheme of things pretty close. What seems somewhat bizarre is that all three raw converters I tested seem to have developed a mind of their own, when it comes to selecting the colour temperature of the shot. The "as shot" readings are nowhere near the 5400K - 5600K range. I would have expected the preset colour temperature to come across with the metadata and establish a starting point for the raw conversion. This is not happening and the raw converters are selecting some rather unusual colour temperature settings.

[dje]

What seems somewhat bizarre is that all three raw converters I tested seem to have developed a mind of their own, when it comes to selecting the colour temperature of the shot. The "as shot" readings are nowhere near the 5400K - 5600K range. I would have expected the preset colour temperature to come across with the metadata and establish a starting point for the raw conversion. This is not happening and the raw converters are selecting some rather unusual colour temperature settings.

Manfred this had me intrigued and I just had to do some research into it !

I had a look at the metadata (Manufacturers Notes section) for one of my NEF files and the only setting I could find in relation to As Shot WB was a tag with red and blue multipliers eg R 1.7 B 1.4 G1 1.0 G2 1.0. I could find no reference to temp or tint. I came to the conclusion that the temp and tint displayed by the software must be an estimate calculated by the software itself, based on the multiplier values.

A bit of googling then discovered this quote from Eric Chan of Adobe here

"There is metadata in the raw file that indicates the white balance gains (per-channel scale factors) for the in-camera white balance settings at the time of capture. ACR reads this metadata and translates it (per the Adobe-built camera profile) to temperature and tint values which are displayed in the ACR user interface."

Presumably other raw converters do something similar.

This is a revelation for me.

Dave

[pnodrog]

If you read both my and George's posts, we are not dealing with AWB at all and neither of us were using that setting; but are using presets. So while I understand and generally agree with what people wrote in the thread you have linked to, I don't think the link is particularly relevant to the issue being discussed.

In my case I set the camera WB to "flash" setting which is 5400K on the D800. I shot a flash that is rated at 5600K ±50K and ensured that ambient light would not impact the shot, so while not a perfect match, in the scheme of things pretty close. What seems somewhat bizarre is that all three raw converters I tested seem to have developed a mind of their own, when it comes to selecting the colour temperature of the shot. The "as shot" readings are nowhere near the 5400K - 5600K range. I would have expected the preset colour temperature to come across with the metadata and establish a starting point for the raw conversion. This is not happening and the raw converters are selecting some rather unusual colour temperature settings.

I realise that the thread was about AWD. However the first step AWB is for the colour temperature to be determined and this cannot be done by looking at scene using a camera or by software analysing an image regardless of its file type. Obviously if the scene has good sized areas of neutral colours the chances of a satisfactory result is greater. However in all cases the camera or software has to make assumptions.

The metadata indication of a flash being used should offer a very good clue to the software about the colour temperature that existed for at least some of the scene. However the vast majority of photographs are taken without flash so I do not know how much weight the software will give to the flash information.

[xpatUSA]

. . . However the first step AWB is for the colour temperature to be determined and this cannot be done by looking at scene using a camera or by software analysing an image regardless of its file type. Obviously if the scene has good sized areas of neutral colours the chances of a satisfactory result is greater. However in all cases the camera or software has to make assumptions.

Indeed, AWB is almost by definition an estimate. However, there are a good few algorithms out there that do a passable job on the "average" image - and a really bad job on an image with large areas of just one color.

As to what the WB actually is - it is represented as the color temperature of a scene illuminant (which can be anything) and it's distance (tint) from the classic blackbody temperature line as shown on the good old 1931 CIE xyY gamut diagram, or better shown on the CIE Luv diagram.

It is rare that Real World lighting ever has the same x,y or u,v as any standard camera WB setting. It is perhaps unreasonable to expect that all converters should give the same colors from a single image.

Also, some camera do more than just use 3 multipliers. The Sigma SD14 for example passes a full 3x3 matrix (three channel color mixing) to the raw converter to do with as it will.

[george013]

I found some interesting info on the net http://photo.stackexchange.com/quest...of-a-raw-image
Reading that, the color temperature doesn't come from the camera or the RAW-file. It's a calculated number.

George

[dje]

I found some interesting info on the net http://photo.stackexchange.com/quest...of-a-raw-image
Reading that, the color temperature doesn't come from the camera or the RAW-file. It's a calculated number.

George

Yes George that's what I am saying in post 31 above.

Dave

[george013]

Yes George that's what I am saying in post 31 above.

Dave

I just woke up. Can't find smileys.

George