Three questions about white balance

[Inkanyezi]

To achieve this, the lighting at the source would need to be as close as possible to an absolute consistent Colour Temperature. This could be achieved using a (well designed) Photographic Studio and (well maintained) Studio Flash.

This subject to some is a bit touchy, mainly because "colour temperature" is an evasive property, and per se, does not exist in any gas discharge lamp, as electronic flash.

The colour temperatures of daylight and flash are "perceived" colour temperature, but do not inherently cling to the K scale, as they are not emitted from a black body.

The black body radiation defines the K scale, as a black body will radiate that colour temperature when heated to the same Kelvin temperature. However, a characteristic of black body radiation is a richness in red and infrared, while the blue and violet as well as ultra-violet are not so pronounced. Between those extremes, black body radiation has a smooth spectral curve, with no spikes or notches, falling evenly toward the higher frequencies.

Sunlight is not emitted from a black body. Any star, Sun as well, is a large nuclear reactor, mainly combining hydrogen into helium. When the very jagged spectrum of the sunlight hits the atmosphere, it is largely dispersed, and much of its blue radiation is spread over the entire globe, while the longer rays hit the surface of Earth more directly. The daylight spectral curve does not have the smoothness of the black body radiation. Neither has the studio flash.

In essense, this means, that if the studio flash has a spectral curve similar to that of daylight, it is possible to get similar tonal reproduction using a flash as with daylight. However with true K radiators, as incandescent bulbs, it is impossible to get similar colour rendition. It can be accomplished with a parametric filter to change the spectral curve, as neodymium lamps do, even though those lamps have a rather low colour rendition index. The metameric response may then be similar to daylight, when white balancing for the lower K value of the lamp.

So without filtering, the metameric response from daylight and K radiators (incandescent lamps) cannot be similar, are not equal; never.

The lighting industry has coped with this problem by adopting two different K scales, one for black bodies, up to 4999 K, and another from 5000 K and above. The one for values from 5000 and above is called "daylight".

And of course an electronic flash, by design, is consistent. At least as long as its energy is consistent, the discharge from the capacitors.

We might appreciate, that skin tones of a person is never seen in the light of an electronic flash. The electronic flash has its own properties, and the metameric response will be different from natural daylight, although it may be consistent. It is however different from any way that we may see skin tones naturally.

It boils down to the simple fact that there is no "dead on" perceived skin tone, even if there may be a "dead on" white balance, as measured from a white reference. The white reference or measurement directly from the light source, can adjust our three channel processing of the RAW data to the colour of the light source, but the metameric response, the tri-stimuli response in our eyes, the skin tones, will depend on how the light is reflected from that particular skin. It will be different, with any different light source.

[William W]

Absolutely.

WW

[Manfred M]

I don't totally agree with your view of black body writeup, Urban, but that being said, the last time I formally looked at the stuff was 35 - 40 years ago, so I've forgotten a lot of what I knew.

While it is true that a black body radiator is a theoretical construct and no "real" material completely emulates one. However, real materials come very, very close over a limited temperature range. Heat a bar of steel to melting point and the glow of the liquid metal acts like a black body radiator. Unfortunately, real materials do not have an infinite melting point and will eventually ionize and no longer show the characteristics of a black body radiator. Temperature measurements in the refining / metal production industries as well as in measuring combustion processes (for instance temperature monitoring in combustion chambers of thermal power generation) use optical measurement techniques track quite close to what one would expect from a theoretical black body radiator.

An electronic flash should provide relatively stable colour temperature output AT A GIVEN POWER LEVEL, but will show characteristics above or below the theoretical 5500K value under normal operations. Add to this the impact of aging flash tubes and the colour temperature output will be come deficient at the blue end and richer at the yellow / red end of the spectrum. Add a light modifier, and the effective colour temperature can change even more. Even with a decent set of lights, set at the same output level and "new" flash tubes I have seen output variations in the ±50K to ±100K.

http://paulcbuff.com/e640.php

The other issue is that the sunlight in a vacuum tracks quite well against a theoretical black body radiator, but once you involve the atmosphere and the interactions between the gases (particularly nitrogen), water vapour and dust particles, we get a mixed lighting situation. North light on a clear cloudless day can give colour temperature readings in excess of 10000K, whereas readings to the south are going to be closer to what we refer to as "normal" daylight.

[xpatUSA]

In support of the foregoing posts esp. Bill, Urban, Manfred, I agree with the consensus that it is impossible to a) set a single, correct WB in-camera for just about any scene you look at and b) even believe general color temperature values for this or that kind of lamp. No wonder cameras struggle to get even close!

Found this today:

http://www.lrc.rpi.edu/programs/nlpi.../whatisCCT.asp

A very clear explanation, I thought, and it had this excellent illustration (apart from the bad grammar at top right):



The interesting part is the two 3000K CCT lamps A and B. I hope that your tint slider has plenty of adjustment

The above is in CIELUV space; for those more comfortable with CIE xyY, here y'are:


.

[xpatUSA]

Drifting slightly OT . . .

I may have asked this in a previous CiC life but, apropos of this thread, what do y'all think is the absolute bang-on totally-correct WB setting for a colored light-source when you want the image of that source to have the correct color?

For example, if you shoot a luminous watch close up in total darkness and want the lume color to be correct.

[Inkanyezi]

Found this today:

http://www.lrc.rpi.edu/programs/nlpi.../whatisCCT.asp

A very clear explanation, I thought, and it had this excellent illustration (apart from the bad grammar at top right):

The site indeed is a very good source of information about light sources.

There are several pages, that can be accessed simply by clicking "next" and "previoius" arrows.
Much of the info is summary in one page, but can have better coverage in another, so for example the difference between light sources below 5000 K and from 5000 K and up are not specified in the page http://www.lrc.rpi.edu/programs/nlpi...eringIndex.asp although they are briefly mentioned:

"For CCTs less than 5000 K, the reference illuminants used in the CRI calculation procedure are the SPDs of blackbody radiators; for CCTs above 5000 K, imaginary SPDs calculated from a mathematical model of daylight are used. These reference sources were selected to approximate incandescent lamps and daylight, respectively."
(My boldfacing)

The difference between blackbody radiators and "daylight" is the spectral distribution, where the blackbody has a smooth curve that is high at low frequencies and falling toward high frequencies, while "daylight" has two "humps" close to each end, with a "valley" between. "Daylight" is considerably stronger in the blue to violet range than the blackbody and weaker in green. Different references are used for the respective colour temperatures. Below 5000 K the reference is blackbody, and from 5000 K and up, the reference is "daylight", an imaginary entity, a construct from a mathematical model. Lamp manufacturers still use not only calculation, but also subjective evaluation for assessing CRI. Test samples, maps of colour patches illuminated by the standard reference and the test source respectively, are observed and compared.

In the camera, as well as in software for RAW processing, the two different spectral curve models are used when setting WB as a K value.

No energy saving lamp, CFL lamp or LED lamp has the properties of the black body radiator. They usually have a CRI below 90. This is very evident if mixing CRI 99 incandescent lamps with FL or LED sources. However, when not mixing light sources, rendition of colours might be more appealing with modern energy saving lamps than with incandescent lamps in spite of their lower CRI, because their spectral distribution may be closer to that of daylight. Therefore, mixing light sources should be avoided. Using only one type mostly produces very good colour rendition in images when WB is set with a reference card.

The K scale can be trusted for WB setting with incandescent sources, but should not be used for other types of lamps.

[William W]

. . .An electronic flash should provide relatively stable colour temperature output AT A GIVEN POWER LEVEL, but will show characteristics above or below the theoretical 5500K value under normal operations. etc.

I concur.


In my commentary on this thread I have taken the view more toward the PRACTICAL side of Photography rather than the Theoretical side of Physics of Light.

As mentioned before my main thrust was to convey that 'correct' or 'dead on' Colour Balance is not easy to get and more importantly not really necessary.

The mention of Black Body Radiators only forces that point of view, and from a different and more basic premise.

I tried terribly to be careful in choosing my words lest the whole theory of Colour and Light were to come into the conversation and I am glad this quote was used as the introduction to those topics.

My bold now for emphasis:

"To achieve this, ['dead on' white balance] the lighting at the source would need to be as close as possible to an absolute consistent Colour Temperature. This could be achieved using a (well designed) Photographic Studio and (well maintained) Studio Flash."


WW

[William W]

. .The K scale can be trusted for WB setting with incandescent sources, but should not be used for other types of lamps.

Here's where advice would differ.

I take a more practical approach and, whilst I concur that the K Scale (for example on the camera's WB settings) can't be used to indicate an exact reference of the Colour Temperature of many individual light sources or a mixed lighting source, on the other hand it can be a valuable PRACTICAL TOOL to USE as a setting when DOING Photography.

WW

[TonyW]

In support of the foregoing posts esp. Bill, Urban, Manfred, I agree with the consensus that it is impossible to a) set a single, correct WB in-camera for just about any scene you look at and b) even believe general color temperature values for this or that kind of lamp. No wonder cameras struggle to get even close!

Found this today:

http://www.lrc.rpi.edu/programs/nlpi.../whatisCCT.asp

A very clear explanation, I thought, and it had this excellent illustration (apart from the bad grammar at top right):



The interesting part is the two 3000K CCT lamps A and B. I hope that your tint slider has plenty of adjustment

The above is in CIELUV space; for those more comfortable with CIE xyY, here y'are:


.

We are getting into the nitty-gritty here which I like, but I still have some problem with the definition of iso-temperature lines and CCT. In the article, they seem to be defined circularly. That is, the iso-temperature lines are points in the chromaticity diagram with the same temperature and the temperature of a point off the black body locus is defined by where the iso-emperature line hits the black body locus. Perhaps I have missed something here or I need to read further.

On a slightly different point, we know that it is possible that light sources with power spectra different from black body radiation can be seen as a point in the chromaticity diagram along the black body locus (and their colour temperature is then clearly defined). This is an example of metamerism. But what we see in an image is mostly not the light source but its reflection off the subject of the image which would not just depend on the temperature of the light source. This means that to replicate exactly what we see directly in a scene we would need to apply a different white balance in different parts of the image (which supports the idea that there is no absolutely correct white balance). Fortunately, in general we do not aim for this.

[xpatUSA]

We are getting into the nitty-gritty here which I like, but I still have some problem with the definition of iso-temperature lines and CCT. In the article, they seem to be defined circularly. That is, the iso-temperature lines are points in the chromaticity diagram with the same temperature and the temperature of a point off the black body locus is defined by where the iso-temperature line hits the black body locus. Perhaps I have missed something here or I need to read further.

I'll give it a try, Tony. Have a look at this 1960 CIE diagram and ignore, for the moment, the isotherms and the CIE standard illuminants:



Now, invent a lamp, any old lamp which will be the chicken as opposed to the egg. What is the correlated color temperature of the lamp? We can't answer, because we don't know the actual color of the lamp. But we could measure the color or calculate it from it's spectral emission curve and thereby determine it's color in CIELUV space as u,v = say 0.2,0.3 (chosen to make it easy) = call it T23 for Tony's lamp #23. If we now mark that u,v point T23 on the diagram, we can then find the point on the Planckian curve where a line from T23 meets the curve at exactly 90 degs (normal) to the curve. The theoretical black-body temperature at that point on the curve is the correlated color temperature (CCT) of our hypothetical lamp, about 8000K.

In other words, the lamp color (u,v chromaticity) determines the CCT, rather than the CCT determining the lamp color. What we should learn from that, too, is that 5500K on a lamp package does not tell us what color the lamp is (for example, check out points D55 and E in the diagram above - they are different colors).

[Inkanyezi]

(for example, check out points D55 and E in the diagram above - they are different colors)

It should be no surprise, considering that D55 is above 5000 K, so it should not lie very close to the black body curve, as it does not have the spectral distribution of a black body.

It boils down to what we perceive as "white" light. "White" can miss several lines in the spectrum, which is very evident if we look at the cheapest types of white LED, those we find in dead cheap pocket lamps or bicycle lights. Those are usually a blue LED, with yellow fluorescence, emitting blue and yellow, which we perceive as white when we look at the lamp. When the lamp is mounted as headlight on a bicycle, and it shines on the rear reflectors of a car, they do not light up with red light, as there is no red in the spectrum of those cheap diodes. For very evident reasons, the CRI of such a lamp is very low, below 50, but nevertheless, when we look at the lamp or illuminate a white paper with it, it looks white. If we shine it on a blue paper and a red paper, the blue one will be blue, and the red one will be black. A yellow paper however will be yellow in that light.

Metamerism is our three-colour image, as perceived by our eyes, and mimicked by the camera sensor and computer display by the three colours that can display virtually all colours in the spectrum. However, any colour in nature, reflected off a surface, if it is not fluorescing, can only reflect the colours present in the light that shines upon it. To some degree, pigments may twist the light waves, so that the light bounced off the surface will be different in colour than the light shining upon it, fluorescence. We find it in clothing, vehicle colours and many other places, and particularly in clothing it is annoying that it is difficult to find a metameric match when taking pictures of fashion. That vividly green coat may look brown or violet in the image, and the violet scarf might get a dull blue colour in our image. A wider gamut of colours can help, a wider colour space, if we also provide the right type of illumination. Most people know that a garment will not look the same in indoors lighting as outside, and it is common to take it outside to see what it will look like in daylight. The light that shines upon an object can define the colour we perceive when looking at it.

The colour space of most computer screens cannot display some of the colours we may perceive in nature. Violet tones, in particular if saturated, are not possible to reproduce in the sRGB colour space of the computer screen. If we don't drive any channel to clipping, we can still have tonality, but the colour will not be the one we would perceive when looking directly at for example a rhododendron flower in daylight. This is a situation we have to cope with, and to make the best of it, we can learn to handle saturated colours carefully, in order to retain as much as possible of their metameric rendition.

[xpatUSA]

Originally Posted by xpatUSA . . . (for example, check out points D55 and E in the diagram above - they are different colors)

It should be no surprise, considering that D55 is above 5000 K, so it should not lie very close to the black body curve, as it does not have the spectral distribution of a black body.

That was not my point and no surprise was intended.

[Inkanyezi]

That was not my point and no surprise was intended.

I know, as it was declared in the post already:

ignore, for the moment, the isotherms and the CIE standard illuminants

However, it has some bearing on the thread topic and its original questions, even though it might have been more relevant in the recently capsized thread on problems with red: Editing Red

Violet colours, a mix of the red and blue channels, are the most difficult to contain in the very limited colour space of our monitors, while red actually does not pose a great problem, if we understand how the camera measures light and the fact that colour is ignored by the light meter as well as the combined histograms and statistics in computer programs, which are intended to display distribution of tones and exposure.

I will use the image at http://naturfoto.ifokus.se/discussio...iscussions-1#4 to show the problem.

If we look at the combined histogram for all colours, it clearly states a maximum value of 251, which in theory should avoid clipping. But there is a hitch. The number that is presented is a statistical number, which considers all colours, pretending that the green and blue channel values should be present in any pixel and integrated into the red channel value to calculate a mean value. Hence there is a possibility that any channel would reach clipping without the histogram or the numbers showing anything odd. After all, the mean value of 255+0+0 is only 85.



The red channel histogram however displays something completely different. We can see that red is heavily clipped, that a lot of red pixels hit the ceiling of the limited space that can be displayed. The "all colours" histogram did not reveal this, and maybe the creator of the image would not bother about those clipped pixels, as we all have artistic freedom to interpret reality. But this red channel histogram clearly identifies that many red pixels are at the 255 level, even though the three channel histogram purported that nothing was above 251.



There is a similar problem with any saturated colour that is brightly illuminated, and sometimes exposure must be held back, also when the colour itself is outside the colour space, in order to retain tonality, by not clipping any channel. The most problematic colour in this respect is violet, but also yellow can be suspected to the same behaviour. When looking at real RAW histograms, you may clearly see, whether it will be possible to retain tonality in very bright saturated colours, and unless you understand how to measure light at them, bracketing might be the only way to preserve their tonality.

[xpatUSA]

Violet colours, a mix of the red and blue channels, are the most difficult to contain in the very limited colour space of our monitors, while red actually does not pose a great problem, if we understand how the camera measures light and the fact that colour is ignored by the light meter as well as the combined histograms and statistics in computer programs, which are intended to display distribution of tones and exposure.

Quite so, see my post here:

One Way to fix a Flower Shot

I will use the image at http://naturfoto.ifokus.se/discussio...iscussions-1#4 to show the problem.

The link is not working today

If we look at the combined histogram for all colours, it clearly states a maximum value of 251, which in theory should avoid clipping. But there is a hitch. The number that is presented is a statistical number, which considers all colours, pretending that the green and blue channel values should be present in any pixel and integrated into the red channel value to calculate a mean value. Hence there is a possibility that any channel would reach clipping without the histogram or the numbers showing anything odd. After all, the mean value of 255+0+0 is only 85.



The red channel histogram however displays something completely different. We can see that red is heavily clipped, that a lot of red pixels hit the ceiling of the limited space that can be displayed. The "all colours" histogram did not reveal this, and maybe the creator of the image would not bother about those clipped pixels, as we all have artistic freedom to interpret reality. But this red channel histogram clearly identifies that many red pixels are at the 255 level, even though the three channel histogram purported that nothing was above 251.



There is a similar problem with any saturated colour that is brightly illuminated, and sometimes exposure must be held back, also when the colour itself is outside the colour space, in order to retain tonality, by not clipping any channel. The most problematic colour in this respect is violet, but also yellow can be suspected to the same behaviour.

Is it also your belief that a reduction in exposure reduces saturation, or have I misunderstood?

Yellow flower discussion here: https://www.cambridgeincolour.com/forums/thread46509.htm#post520251
.

[Manfred M]

While I enjoy the technical discussion, getting my feet back on the ground for a moment...

1. Most photographers I know really don't care about all the technical details, as all they want is a decent looking shot;

2. Most shots outside of a studio have multiple light sources, either direct or indirect, so will not have a single white balance point;

3. Unless I am shooting in a studio situation and am taking using test targets with known colour values, I won't know the actual colour temperature of the light hitting my subject anyways. I am assuming my flash tubes are not too old and have gone a bit yellow. My eyes aren't good enough to get things much closer than a couple of hundred K anyways, without a target. On a good day, my mid-range studio lights are probably running ±50K, between shots, best case. If I start changing the power level, this will change even more.

4. Not being a commercial photographer, dealing with a client's logo, I shoot for a consistent look and a "good" look. Even when I shoot a target with a model, I will tend to warm up the shot a touch. Neutral WB people shots usually don't look as good as ones that have been warmed up a touch.

Bottom line; close enough is usually good enough. People do spend too much time worry about this rather than getting good images...

[xpatUSA]

While I enjoy the technical discussion, getting my feet back on the ground for a moment...

1. Most photographers I know really don't care about all the technical details, as all they want is a decent looking shot;

2. Most shots outside of a studio have multiple light sources, either direct or indirect, so will not have a single white balance point;

3. Unless I am shooting in a studio situation and am taking using test targets with known colour values, I won't know the actual colour temperature of the light hitting my subject anyways. I am assuming my flash tubes are not too old and have gone a bit yellow. My eyes aren't good enough to get things much closer than a couple of hundred K anyways, without a target. On a good day, my mid-range studio lights are probably running ±50K, between shots, best case. If I start changing the power level, this will change even more.

4. Not being a commercial photographer, dealing with a client's logo, I shoot for a consistent look and a "good" look. Even when I shoot a target with a model, I will tend to warm up the shot a touch. Neutral WB people shots usually don't look as good as ones that have been warmed up a touch.

Bottom line; close enough is usually good enough. People do spend too much time worry about this rather than getting good images...

Thank you for your cogent point, Manfred.

Taking the hint and leaving the technical discussion . . .

[Inkanyezi]

Is it also your belief that a reduction in exposure reduces saturation, or have I misunderstood?

No, reduction of exposure does not reduce saturation, but it reduces - just that - exposure. It is not uncommon in violet that the red channel is clipping, or the blue, or sometimes both.

Likewise for yellow, the red channel sometimes clips while the two other do not reach maximum value. Green is the channel that most seldom reaches clipping.

Saturation, as I see it, is absence of other colours than the saturated colour, regardless of luminosity. Particularly for the red channel, and sometimes for the blue, exposure must be held back, compensated with minus, from what the meter suggests. We also have the special cases of mixed colours as violet and yellow, which influence the light meter when saturated in a rather similar way, so that they tend to read as a lower value than their actual luminance as the camera will see it, although not as low as a saturated red.

It might be noted that the tri-colour matrix of the media we use cannot produce a saturated yellow or a saturated violet, as they are always displayed as a mix of two colours, and thus by definition are not saturated but presented as tri-stimuli.

The link not working might be that you are not a member on the site. Direct link to the image: http://naturfoto.ifokus.se/u2/497c5c...t/dsc-2074.jpg It belongs to Malin Hultman http://www.malinhultman.weebly.com

[Max von MeiselMaus]

Drifting slightly OT . . .

I may have asked this in a previous CiC life but, apropos of this thread, what do y'all think is the absolute bang-on totally-correct WB setting for a colored light-source when you want the image of that source to have the correct color?

For example, if you shoot a luminous watch close up in total darkness and want the lume color to be correct.

Interesting as the technical discussion is, it doesn't answer practical questions like these. I am with Manfred that the technical stuff is useful to be aware of, but it is how to apply it in real life situations that is the problem.

So, with this one. My thoughts. If you set the WB to the colour of the luminous glow, it would appear grey. No good. Will you have any other lights in the room or will the only illumination be the luminous paint? If the latter, I wonder if setting it to (notional) daylight would give a more accurate colour. If there are any other lights, you would of course have to set for them.

I would be interested to see that shot. Sounds like an intriguing exercise.

[xpatUSA]

Interesting as the technical discussion is, it doesn't answer practical questions like these. I am with Manfred that the technical stuff is useful to be aware of, but it is how to apply it in real life situations - that is the problem.

Here's my real life situation from some years ago. How did I measure the brightness in cd/m^2 of a watch's luminous material, armed only with a digital camera? Just the material, not it's surroundings.

So, with this one. My thoughts. If you set the WB to the colour of the luminous glow, it would appear grey. No good.

Correct.

Will you have any other lights in the room or will the only illumination be the luminous paint?

Yes, as I said, "if you shoot a luminous watch close up in total darkness".

If the latter, I wonder if setting it to (notional) daylight would give a more accurate colour.

What is (notional) daylight?

If there are any other lights, you would of course have to set for them.

Of course? Are you sure?

I would be interested to see that shot. Sounds like an intriguing exercise.

Here's an old post from my blog with such a shot:

http://tcktek.blogspot.com/2011/06/exposing-lume.html

6500K gets mentioned twice in there, and even a tint value. I now believe both to incorrect for a couple of reasons

[Manfred M]

Shoot a tiny gray card illuminated by the watch alone? Use that to do a custom white balance...