[UnknownVT]

Because the spectrum of the black body radiation is defined for all frequencies, one can compute the temperature from the ratio of intensities for any two frequencies. If one is adding a third frequency (color) then one ends up with having 2 temperatures from two adjacent intensity ratios.

If the light source's spectrum is discrete or continous has nothing to do with all of this.

I don't think that is quite right -
Although any light source a color temperature can be calculated -
that is somewhat less useful than a CCT for a light source that approximates to a good/perfect radiating black body -

A CCT for any light source that falls far away from the incandescent black body locus - the Planckian Locus (Wikipedia) - is almost meaningless in terms of any practical usage (LEDs especially colored ones have very narrow bandwidth/frequencies and are almost monochromatic so do not approximate to an incandescent black body therefore would fall very far away from the Planckian locus)

Planckian Locus - ie: the line that traces a incandescent black body radiator at any given CCT (Corrolated Color Temperature)


Take a position way off the locus of approx x=0.425, y=0.3 would have more blue than a light on or above the locus and still have approx the same CCT.


Planckian locus (by CIE)

So looking at this graph one can see this seems a light with more blue but off (below) the locus/line could have a low CCT.

In fact looking at both diagrams since extrapolating those Tc lines which tend to converge -
ultimately it would get to an area off (below) the locus where tiny shifts could result in big differences in CCT/Tc?

So even though in theory one might be able to calculate a CCT for discontinuous spectrum - it is less than useful and can be very misleading - especially one attempts to apply that color temperature to RAW processing

Because image processing software does use three colors indeed (RGB), we get a "temperature" and a "hue" measure. Where a "hue" means that the light didn't have a defined temperature in the first place.

You know I had never thought of it this way -
but that logic does not quite hold water either.

Our eyes see by RGB receptors as well, but when a light source is separated to its color components (like using a prism - that famous "Darkside of the Moon" logo ) one can see that a continous spectrum has many color components. Whereas "white" or any other color made up from separate Red, Green and Blue LEDs do NOT - they separate out to Red, Green and Blue only - the Bayer matrix sensor can detect this and it makes a difference to the photo.

Try it and see for yourself - shoot a magenta scene
1) lit by gel'd/filtered tungsten/halogen
and a separate one
2) lit with just red and blue LEDs

There's a huge difference when one processes it -
why? because the filtered tungsten light has still a lot of color components to work with (albeit a bit low) which the Bayer RGB matrix captures.
Whereas the LED version only has Red and Blue to work with......

For more detailed discussion re: color temperature and how the Planckian Locus is relevant please see: page 2 of this CPF thread

---

Last edited by UnknownVT; 03-19-2010 at 10:42 AM.

[Marc Sabatella]

This is a hard one for me to answer.

I tend to prefer to present my photos as I saw it - be it psychedelic or Warhol'ish and all - it's not that I'm a stickler for documentation - it's just that I feel my photos should reflect the gig as I saw it.

So I tend to shoot JPGs and accept what I get - and mostly I am happy, even if I know full well I could have squeezed more optimal quality out of RAW

OK, that makes sense. Different goals, so different views on what the issues are. I don't usually try to remove *all* color cast, but I try to find a balance between something that looks natural and yet still gives some sense of the color of the light. So my issues surround the extent to which I can achieve that sort of color balance, and yours surround the extent to which the images with the original lighting color fully intact are OK in terms of detail, aberrations, and artifacts.

And I can easily believe that colored LED would be even worse than gels due to the much narrower spectrum I understand they produce. FWIW, I don't have PPL 4, but I just tried PPL 3 on one of my least favorite green images. No dice. It's green and nothing but green, regardless of WB. I had better luck with ACDSee and it's "advanced color" tool, which at least let me shift the green to a different color - but nothing like believable skin tones.

[UnknownVT]

And I can easily believe that colored LED would be even worse than gels due to the much narrower spectrum I understand they produce. FWIW, I don't have PPL 4, but I just tried PPL 3 on one of my least favorite green images. No dice. It's green and nothing but green, regardless of WB. I had better luck with ACDSee and it's "advanced color" tool, which at least let me shift the green to a different color - but nothing like believable skin tones.

That makes sense and that is why I was so surprised at the result I managed to get out of Pentax DCU 4.11 (SilkyPix)
- neither ACR nor LightRoom could do that.

However just on the off-chance I used the "Grey Point Setting" (middle of the White Balance Tab) in Pentax DCU 4.11 (SilkyPix)

and pointed to the bassist "white" shirt which was magenta in the Camera Settings version - and lo-and-behold a more naturally lit shot albeit a bit dark.... (and my jaw dropped)

[Ben_Edict]

Because this notion of color temperature keeps getting repeated, please let me comment about it.

"In the true sense of the definition", only black bodies heated to a given temperature radiate light with a temperature (which is the black body's temperature due to thermodynamic equilibrium).

Because our dear star's "sun" surface is such a black body (heated to ~6000°C), this notion is very convenient.

Because the spectrum of the black body radiation is defined for all frequencies, one can compute the temperature from the ratio of intensities for any two frequencies. If one is adding a third frequency (color) then one ends up with having 2 temperatures from two adjacent intensity ratios.

Because image processing software does use three colors indeed (RGB), we get a "temperature" and a "hue" measure. Where a "hue" means that the light didn't have a defined temperature in the first place.

If the light source's spectrum is discrete or continous has nothing to do with all of this.


E.g., many light sources with a "known" color temperature have a discrete spectrum, like fluorescent lamps.



Further reading: Color temperature - Wikipedia, the free encyclopedia

Falk, if you read the article closely, you'll find, that my definition is correct and that to radiators of discrete spectra only an equivalent to colour temperature, the "correlated colour temperature" is applicable. CCT is only perceived by the human eye to be approximately the same as the colour temp. of a black body radiator!

And that can make a very big difference not only for photographic film or imaging sensors, but even for the perception by the human eye. Even if your emission lines light source produces a brightness maximum ressembling a temperature radiator, it does not guarantee, that all colours can be repoduced, because you may have gaps in the spectrum.

This is the reason, why it is necessary to apply the Colour Rendering Index on top of a stated color temperature.

There is more to a black body/continous light source, than just the possibility to calculate the emission maximum with the help of two point on the spectral response curve. For instance a true black body radiator is strongly following Wien's Displacement Law ("Wiensches Verschiebungsgesetz"), which is not at all applicable to emission line radiators.

Ben

[falconeye]

I don't think that is quite right -
[...]
So even though in theory one might be able to calculate a CCT for discontinuous spectrum - it is less than useful and can be very misleading - especially one attempts to apply that color temperature to RAW processing

Falk, if you read the article closely, you'll find, that my definition is correct and that to radiators of discrete spectra only an equivalent to colour temperature, the "correlated colour temperature" is applicable.

Thanks for getting back on this.

However, I believe we don't yet have reached a common sense of understanding here.

I wrote myself "only black bodies heated to a given temperature radiate light with a temperature". So, clearly, we speak about correlated colour temperature or CCT.

Because the main idea behind all this is so simple, I won't get too technical here. I just want to explain why the question whether a spectrum is discrete or continuous has little importance. What my earlier post tried to explain actually.

So, here we go:

The human eye has receptors with different sensitivity to wavelengths, three kinds (besides luminosity for night vision). These three kinds are called red, green, blue.

In theory, filters could be built which exactly mimic the same sensitivity to wavelengths, for a given sensor. For sake of simplicity of my argument, let me assume that this is the case.

(In case it isn't, sensors and eyes may have different notions of red etc. and a calibration using known color temperatures would be imprecise for non black body radiation (poor color rendition). But that's missing my point!)

My point is that both, camera sensor and eye, don't care how it is stimulated as long as all three kind of recptors are stimulated at all. It simply doesn't matter. And therefore, it doesn't matter if the spectrum is continous or not. The eye doesn't and cannot know. It has no receptors to detect the difference.

Of course, both discrete and continous spectra can be a very bad approximation of Wien's Displacement Law and if it is, there will be large problems with color rendition with real world sensors and Bayer filters. But this isn't an intrinsic property of a discrete spectrum.

[UnknownVT]

I just want to explain why the question whether a spectrum is discrete or continuous has little importance. What my earlier post tried to explain actually.

Maybe I've misunderstood you.

But is there a difference between say real daylight "white"
and "white" made up with equal parts of Red Green and Blue LEDs?

To the eye they both look "white" -
but a prism or measuring instrument that can see the spectrum components will see only red green and blue peaks with the "white of the RGB LEDs -
whereas it will see a continuous spectrum for the daylight with all the colors of the rainbow/spectrum.

This makes a difference in photography and any post processing.

[falconeye]

But is there a difference between say real daylight "white"
and "white" made up with equal parts of Red Green and Blue LEDs?
[...]
This makes a difference in photography and any post processing.

You can make up a false say daylight "white", either by using only three wavelengths (as with your LEDs) or by using a wrong continous spectrum which does still appear white to the human eye.

If you see all the colors of a rainbow or three peaks only doesn't matter. The eye as well as a camera don't contain a beam splitting prism

If a camera has a different spectral sensitivity curve than the eye, then there may be a problem because the so-called color calibration matrix is only calibrated for black body radiation. But this is true for wrong continuous spectra as well. And can be avoided with a better sensor/filter assembly.

So, there simply isn't any fundamental difference between what the eye and what the camera sees.

There is a difference though (as seen by both eye and camera) because a surface may reflect light differently if hit by a wrong spectrum. Because the spectrum of the surface itself may be not smooth at all.


What I wanted to say originally is only this: "The fact that a spectrum is discrete doesn't render the term color temperature less useful than with any other non black body radiation spectrum".

[UnknownVT]

What I wanted to say originally is only this: "The fact that a spectrum is discrete doesn't render the term color temperature less useful than with any other non black body radiation spectrum".

I agree with this -
in fact this is what I was trying to say in my long post with the Planckian locus diagrams.

So, there simply isn't any fundamental difference between what the eye and what the camera sees.

Hopefully I am not quoting you out of context.

This I have to disagree with.

the Kruithof curve shows how most people see "white" under different light levels. In fact the best example is most people regard tungsten lighting as "white" especially at lower illumination levels.

This is pretty obviously not the daylight "white" as we understand it.
If one uses fixed daylight white balance or use daylight balanced slide film - the scene illuminated with tungsten light will appear to be very amber/yellow -

So this is at least one case where the eye sees differently from the camera.

Similarly under varying daylight conditions the CCT (color temperature) can change a fair amount - but our eye/brain adapts and mostly see color scenes as if it were under noonday sun - not many realize that under no direct sun cloudy bright how blue the scene is (ie: much higher color temperature in the 9000-1000K range!) That is until one takes a photo with fixed Daylight white balance or daylight color slide film.

So there is a fundamental and pretty obvious difference between what the eye/brain perceives and what a camera capture.

You can make up a false say daylight "white", either by using only three wavelengths (as with your LEDs) or by using a wrong continuous spectrum which does still appear white to the human eye.

If you see all the colors of a rainbow or three peaks only doesn't matter. The eye as well as a camera don't contain a beam splitting prism

If a camera has a different spectral sensitivity curve than the eye, then there may be a problem because the so-called color calibration matrix is only calibrated for black body radiation. But this is true for wrong continuous spectra as well. And can be avoided with a better sensor/filter assembly.

So, there simply isn't any fundamental difference between what the eye and what the camera sees.

There is a difference though (as seen by both eye and camera) because a surface may reflect light differently if hit by a wrong spectrum. Because the spectrum of the surface itself may be not smooth at all.

I will reiterate the a discontinuous spectrum and a continuous spectrum makes a difference to photography - especially for post processing and adjustments to color balance is concerned.

It may be hard to visualize with "white" light made up of RGB - but the above differences between the eye./brain and camera hopefully give an idea.

Let's go back to my original stated problem of magenta light -
if the magenta light is a filtered/gel'd tungsten/halogen light that light does not have peaks at Red and Blue with a bathtub void in the middle. Its spectrum is some what "continuous" or let's just say much less discontinuous and spiky than of magenta made up with Red and Blue LEDs. In color adjustment for post processing - there is some room to for manipulation/adjustment of color balance because there are the intermediate frequencies captured which allow for this.

Whereas magenta made up of Red and Blue LEDs will have a spectrum that is basically just red and blue peaks - this makes any color balance adjustment very difficult - because the captured image lacks any intermediate frequencies - so there is far less room for manipulation - since those frequencies were not captured in the first place.

I believe white made up of Red Green and Blue LEDs only also poses some problems - but it is obviously not as drastic as the magenta - and mostly I do not regard RGB "white" as a problem since the photo looks OK in the first place - so the argument may be moot.

But my point is that a discontinuous spectrum is fundamentally different to a continuous spectrum - our eye/brain may not detect the difference and the photo may well show the scene as seen by the eye -
but the digital camera is more than capable to recording the "spectrum" and there is a difference when it comes to color balance manipulation - some cases are more obvious than others.

I hope that makes some sense

[falconeye]

This is pretty obviously not the daylight "white" as we understand it.
If one uses fixed daylight white balance or use daylight balanced slide film - the scene illuminated with tungsten light will appear to be very amber/yellow -

So this is at least one case where the eye sees differently from the camera.
[...]
but the digital camera is more than capable to recording the "spectrum" and there is a difference when it comes to color balance manipulation

We still talk different language here ...

When I say eye and camera see the same, you'll introduce the brain and its automatic white balance It's the brain's WB why photographers do WB at all. There would be no need otherwise

In your red/blue LED example, one out of three color ranges is totally missing. Again, not intrinsic to a discrete spectrum.

And no, a digital camera is NOT capable to recording the spectrum. As isn't the eye.

Trust me, I would be able to make up a continous "white" power spectrum which creates more trouble than a "white" spectrum made from three LEDs. I've only had to use a lot of extreme red or blue in the (contiunous) spectrum.

[Ben_Edict]

My point is that both, camera sensor and eye, don't care how it is stimulated as long as all three kind of recptors are stimulated at all. It simply doesn't matter. And therefore, it doesn't matter if the spectrum is continous or not. The eye doesn't and cannot know. It has no receptors to detect the difference.

Of course, both discrete and continous spectra can be a very bad approximation of Wien's Displacement Law and if it is, there will be large problems with color rendition with real world sensors and Bayer filters. But this isn't an intrinsic property of a discrete spectrum.

That is not correct. The human eye is able to adapt and to "fiull" in gaps in the spectrum. After some time, we will even adapt to sodium or mercury vapour lamps. But a digital sensor cannot adapt to fill gaps in the spectrum.White balance can only adjust the overall impression of the image. But if part of the spectrum is missing, the corresponding colopur cannot be reproduced.

Ben

EDIT: I don't question your ability to create a troublesome continous light source, though.

[falconeye]

The human eye is able to adapt and to "fiull" in gaps in the spectrum.

I have to give up at this point. Otherwise, I would have to explain in detail how a cone cell and RGB sensor silicon work.

For a further read, start here:

1. Cone cell spectral sensitivity:

[from wikipedia.com]

2. Sensor site spectral sensitivity:
cf. attachment
[from Kodak, for KAF-40000]

(BTW, the infrared peak around 800nm is removed by IR filters on top of sensors.)

[Ben_Edict]

I have to give up at this point. Otherwise, I would have to explain in detail how a cone cell and RGB sensor silicon work.

For a further read, start here:

How a cone or any receptor work, has not much to do with the problem we discuss. The point is, that the human eye does not only "see" something, but that this received information is being processed by the brain. There is no such thing, as an objective human vision. We only perceive. And the brain is trained by experience (just look, how long it takes to gain that experience during childhood!) to deliever an image, that resembles, what we have seen in other circumstances. So, to a degree, it can also "fill in" (and will try to do so) missing colours, if there is enough information at all, to enable colour vision.

A camera sensor (being it digital or analogue), does not perceive. It records, what's there. If part of the spectrum in the lighting is missing, the sensor cannot record that part.

Colours as such do not exist. Any object that we perceive or record coloured, appears only coloured, because they reflect certain parts of the spectrum and absorb others. If light of a certain spectral range cannot be reflected, because it isn't emitted in the first place, an object will have a severe colour shift (if we assume, it will not reflect strongly monochromatic, in which case it would be black if that certain wavelength is missing from the lighting).

This is the reason, why I mentioned Wien's Displacement Law: White Balance

Ofcourse you can, up to a certain degree reconstruct missing colour information from two other colours. That is exactly how colour tv works. But it is no coincidence, that NTSC is the abbreviation for "Never Twice the Same Colour". The reconstruction of colour has limits and it is pressing the limits to reconstruct colours, that have not been recorded at all, which can be easily the case if the lighting equipment is based on discontinous fittings. If it were otherwise, we wpouldn't need expensive. colour balanced flash lights.

Digital White Balance does take two factors into account: Colour Temeprature and the Green-Magenta shift, introduced by non-thermal radiators. That can aviliate the percieved colour shift by these light sources to a large degree. But ofcourse it cannot bring back colours, which were not recoreded. This is the reason, why often images taken under fluorescent lighting may be overall appear neutral in colour, but lack contrast and colour diffrentiation in some parts, namely those, were the light source performs weak…

Ben

[UnknownVT]

We still talk different language here ...

And no, a digital camera is NOT capable to recording the spectrum. As isn't the eye.

Trust me, I would be able to make up a continous "white" power spectrum which creates more trouble than a "white" spectrum made from three LEDs. I've only had to use a lot of extreme red or blue in the (contiunous) spectrum.

Perhaps we are talking different language.....

Just a last try - there is a measurement called the CRI - Color Rendering Index - which is misunderstood a lot as a quantitative measurement.

However it is useful in this case - take a continuous spectrum white light like noon daylight. This light will have a CRI = 100 (perfect) by definition as the sun sky combination can be regarded as a black body radiator.

Now take the discrete Red Green Blue LEDs which has a discontinuous spectrum distinct peaks at red green and blue - yet appears to be "white" to our eyes. This "white" will have a lower CRI because of the gaps in the spectrum and peaks at red green and blue.

CRI measurement does not need anything like a prism or light splitter/diffractor to separate out the color components -
CRI is measured by color patches illuminated by the light source - see Color Rendering Index (CRI) at Wikipedia
- in fact CIE now recommends the use of a Macbeth chart.

So a camera can "see" the difference between a continuous full spectrum white, and the "white" made up of discrete narrow bandwidth red green and blue LEDs - the lower CRI measurement which is done by measurement of the individual color patches on a Macbeth chart - shows that a camera can detect this, since use of the Macbeth chart is the industry standard for color accuracy in digital cameras.

[falconeye]

So a camera can "see" the difference between a continuous full spectrum white, and the "white" made up of discrete narrow bandwidth red green and blue LEDs

I said I'll give up, so just two cents here

CRI is fine.
Just read my 2nd last sentence in my post #37 which is what you now wrote.

What on earth makes you think that the eye would see more correct colors in the same circumstance? Correct, the brain tries to compensate. As does the photographer's photoshop... I agree, brain still beats photoshop on this, but this is another story

And light with bad CRI is not confined to discrete spectra. Which is all I try to say all the time. I think, I'll shut up now

[Ben_Edict]

I said I'll give up, so just two cents here

CRI is fine.
Just read my 2nd last sentence in my post #37 which is what you now wrote.

What on earth makes you think that the eye would see more correct colors in the same circumstance? Correct, the brain tries to compensate. As does the photographer's photoshop... I agree, brain still beats photoshop on this, but this is another story

And light with bad CRI is not confined to discrete spectra. Which is all I try to say all the time. I think, I'll shut up now

Falk, what you wrote in post no. 37, has already written by me in post no. 34. It is obvious, that we all could get faster to agreement, if we (and I include myself expressedly here) would read, what has been posted before. Many threads could be much shorter and easier to read and provide a better conclusion, then.

Ben

P.S.. One thing, CCT and CRI all do help to improve images, but colours that are not present at all (because that part of the spectrum is missing in the light source) cannot be recalculated properly, at best with a good guess. That's basically all I wanted to say. With a continous spectrum light source, colour correction is much easier and more complete.