[falconeye]

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.

I now get an idea about the source of our confusion:

You: continuous spectrum = spectrum where all wavelengths are present, at least to some extent.

Me: continous spectrum = spectrum without spikes aka non-discrete spectrum.


If I reread your posts with your above definition, I tend to agree. However, this isn't the definition of continuous spectra.

And most importantly: you would consider a discrete spectrum to be continous if it only doesn't leave big holes. Absence of holes doesn't render a spectrum continuous, though.

To pinpoint the level of confusion:

A discrete spectrum with red,green,blue LEDs would be more continuous (in your terms) than a continuous spectrum (in my terms) entirely leaving out the green range.

Maybe next time, we should be more careful when using the term "continuous spectrum". Not everybofy may know or use the term as it is normally used in physics.

[Ben_Edict]

I now get an idea about the source of our confusion:

You: continuous spectrum = spectrum where all wavelengths are present, at least to some extent.

Me: continous spectrum = spectrum without spikes aka non-discrete spectrum.


If I reread your posts with your above definition, I tend to agree. However, this isn't the definition of continuous spectra.

And most importantly: you would consider a discrete spectrum to be continous if it only doesn't leave big holes. Absence of holes doesn't render a spectrum continuous, though.

To pinpoint the level of confusion:

A discrete spectrum with red,green,blue LEDs would be more continuous (in your terms) than a continuous spectrum (in my terms) entirely leaving out the green range.

Maybe next time, we should be more careful when using the term "continuous spectrum". Not everybofy may know or use the term as it is normally used in physics.

May be, that is the problem. I am used to the physicist's terminology and also have written a book about colour and colour theory in the distant past…

A line spectrum can "appear" to the eye as being continous, but will be revealed more or less easily, depending on how narrow or broad the lines are. LEDs traditionally have very narrow emission lines. If you have a look at that graph from wikipedia (File:Red-YellowGreen-Blue LED spectra.png - Wikipedia, the free encyclopedia), you know what I mean. This spectrum may be perceived as "white", but a sensor cannot perceive, just record. It will record the combined light of this light source as white (correct WB provided), if it illuminates a white (grey, black) sheet of paper. But if you illuminate a coloured surface, certain colour will be emphasized (those corresponding with the peak output of the LEDs) and other colours will be completely absent (those in the spctrum's valleys).

After all, this white LED only covers app. 100nm of the whole visible spectrum in reality!

I think, we can easily agree, how this will be really problematic to correct by the camera or even in post-processing.

Ofcourse there are "full-spectrum" white LEDs available, but these are not used in the light sources under discussion, as the output is too low. For instance http://www.pbase.com/image/33858156 will deliver also a perceived white light, but its spectrum has a strong peak in the blue. This peak also reveals the non-thermal origin of the light. But obviously this light source could be more easily corrected, than the white composite LEDs.

Ben

[UnknownVT]

A line spectrum can "appear" to the eye as being continous, but will be revealed more or less easily, depending on how narrow or broad the lines are. LEDs traditionally have very narrow emission lines. If you have a look at that graph from wikipedia (File:Red-YellowGreen-Blue LED spectra.png - Wikipedia, the free encyclopedia), you know what I mean. This spectrum may be perceived as "white", but a sensor cannot perceive, just record. It will record the combined light of this light source as white (correct WB provided), if it illuminates a white (grey, black) sheet of paper. But if you illuminate a coloured surface, certain colour will be emphasized (those corresponding with the peak output of the LEDs) and other colours will be completely absent (those in the spctrum's valleys).

After all, this white LED only covers app. 100nm of the whole visible spectrum in reality!

I think, we can easily agree, how this will be really problematic to correct by the camera or even in post-processing.

This is also what I've tried to say in the measurement for CRI - discrete red Green and Blue LEDs may make up a light that appears "white" to the eye -
BUT the camera can capture that "peaky" spectrum and the troughs of the in-between wavelengths/frequencies.



CRI measurement using a Macbeth chart is a good example that a camera can and does capture and differentiate a full continuous spectrum white as in noon daylight, and the "white" made up of discrete red green and blue LEDs.

This is shown up in post-processing where certain colors are more difficult to adjust due to the very low instances of those in between wavelengths/frequencies.

Hopefully this is not in such a foreign language .

[falconeye]

This spectrum may be perceived as "white", but a sensor cannot perceive, just record. It will record the combined light of this light source as white (correct WB provided), if it illuminates a white (grey, black) sheet of paper. But if you illuminate a coloured surface, certain colour will be emphasized (those corresponding with the peak output of the LEDs) and other colours will be completely absent (those in the spctrum's valleys).

Ben, it appears we both use the same terminology then

But something else comes to my attention: What you write above is of course correct. But I stumble upon your "perceive" vs. "record".

In the example you are just giving ("narrow white" giving false coloured surfaces), both the human eye AND the camera will "see" the same wrong colors.

Also, when you write "certain colour will be emphasized" you actually mean "certain wavelength will be emphasized".

Colour is a three-dimensional vector of recorded/perceived luminosities for the three distinct receptor types of an eye/camera. A single wavelength can be given a color (a weight of its two neighboring receptor channels) but not every color corresponds to a wavelength, like white. Multiple wavelengths then are a spectrum, not a color. And of course, there is an infinite number of spectra which correspond to any given single color.

Of course, you know all this already, no doubt about. But it is difficult to discuss if one isn't very careful in wording.


So what all boils down to:

Both the eye and a camera convert a spectrum (a function) to a color (a 3D vector). The conversion operators (the spectral sensitivities I quoted above) may be slightly different but this doesn't matter in our discussion. The point is, after the conversion, it doesn't matter anymore how the spectrum looked like. And both, eye and camera sensor, only transmit the converted result, not the spectrum. A camera doesn't record a spectrum!

The construction of a difference here between camera and eye must fail.


Ben, are you sure you understand what I'm trying to say? Or are you just trying to explain what you think I may not understand?

---

Last edited by falconeye; 03-22-2010 at 05:54 PM.

[falconeye]

CRI measurement using a Macbeth chart is a good example that a camera can and does capture and differentiate a full continuous spectrum white as in noon daylight, and the "white" made up of discrete red green and blue LEDs.

This is shown up in post-processing where certain colors are more difficult to adjust due to the very low instances of those in between wavelengths/frequencies.

UnknownVT, I don't doubt that, never did, never said otherwise and fully understand the fact.

As a camera can distiguish between correct and bad continous spectra for white as well.

As does the eye.

A discrete spectrum just is particulary bad (for both, eye and camera). But not automatically worse than a continuous spectrum.

[UnknownVT]

As a camera can distiguish between correct and bad continous spectra for white as well.
As does the eye.
A discrete spectrum just is particulary bad (for both, eye and camera). But not automatically worse than a continuous spectrum.

Hmmm... I agree with that -
and also have not changed my take on anything I posted in this thread -
I wonder what made it seem as we were talking about different things or languages then?

[falconeye]

I wonder what made it seem as we were talking about different things or languages then?

It all started when I commented Ben's phrase:

"In the true sense of the definition, these do not even have any colour temperature, as they do not have a contious light spectrum."

I struggled to explain why the fact that a spectrum is discrete doesn't mean it can't have a color temperature any less than a continuous spectrum can.

Ben's phrase simply wasn't correct. That's all I ever wanted to say.

[Ben_Edict]

It all started when I commented Ben's phrase:

"In the true sense of the definition, these do not even have any colour temperature, as they do not have a contious light spectrum."

I struggled to explain why the fact that a spectrum is discrete doesn't mean it can't have a color temperature any less than a continuous spectrum can.

Ben's phrase simply wasn't correct. That's all I ever wanted to say.

Sorry, Falk to disagree again. A colour temperature is an inherent property of a continous spectrum, strongly related to the temperature of the radiating body.

A line radiator has by definition no colour temperature. The most you can say, it has properties, equivalent to a colour temperatur, the "correlated colour temperature.

If we look at the example of the composite white light LED, you may calculate this CCT from the emission peaks. But that's all.

Ben

[falconeye]

Sorry, Falk to disagree again. A colour temperature is an inherent property of a continous spectrum, strongly related to the temperature of the radiating body.

A line radiator has by definition no colour temperature. The most you can say, it has properties, equivalent to a colour temperatur, the "correlated colour temperature.

Ben, I knew you would say this

Ben, in the strong sense, only light from a heated black body has a temperature.

When you say "a colour temperature is an inherent property of a continous spectrum" you forget to say "if it is from a black body" and thereby you imply "and so a discrete spectrum doesn't have a color temperature".

But your ommission and implication are just wrong.

You either add the ommission "if it is from a black body" and then, both a continuous spectrum from another source and a discrete spectrum don't have a color temperature.

Or you don't and end up woth the correlated color temparture (CCT) all spectra do have, discrete spectra included.

Turn it whatever way you want, your statement as you make it is incorrect. It is not the continuous/discrete property of a spectrum which does it have a/no temperature (it rather is from being/not being black body radiation).

But now, I am really repeating myself and will stop doing so in this thread.

[Ben_Edict]

Ben, it appears we both use the same terminology then

But something else comes to my attention: What you write above is of course correct. But I stumble upon your "perceive" vs. "record".

In the example you are just giving ("narrow white" giving false coloured surfaces), both the human eye AND the camera will "see" the same wrong colors.

Also, when you write "certain colour will be emphasized" you actually mean "certain wavelength will be emphasized".

Colour is a three-dimensional vector of recorded/perceived luminosities for the three distinct receptor types of an eye/camera. A single wavelength can be given a color (a weight of its two neighboring receptor channels) but not every color corresponds to a wavelength, like white. Multiple wavelengths then are a spectrum, not a color. And of course, there is an infinite number of spectra which correspond to any given single color.

Of course, you know all this already, no doubt about. But it is difficult to discuss if one isn't very careful in wording.


So what all boils down to:

Both the eye and a camera convert a spectrum (a function) to a color (a 3D vector). The conversion operators (the spectral sensitivities I quoted above) may be slightly different but this doesn't matter in our discussion. The point is, after the conversion, it doesn't matter anymore how the spectrum looked like. And both, eye and camera sensor, only transmit the converted result, not the spectrum. A camera doesn't record a spectrum!

The construction of a difference here between camera and eye must fail.

Ben, are you sure you understand what I'm trying to say? Or are you just trying to explain what you think I may not understand?

I agree with most of what you write. May be my trial to summarize, what I mean was too short and again leads to misunderstanding.

When I am talking about "colour", I mean colour, not wavenlength or spectrum, simply because colour is, what we are talking about in photography (leaving BW aside).

I think, that the difference between colour perception, which is a product of the physiological properties of the eye and post-processing (based on experience and applied fuzzy logic) by our brain, is very different from colour recording with a camera sensor.

The human perception enables us to "automatically" correct colours under most lighting conditions, to meet, what the colours would look like under daylight conditions, simypl, because we "know" by experience, how these colours should look. We can also fill in colour gaps, if the light source is a strong emission line source (sodium/mercury vapour).

A sensor/film, will only record those colours, that are actually present in the scene. So, if we deal with a continous light source, will enable the sensor to record all the colours, present in the scene (represented by the wavelengthes, reflected by the surfaces, we perceive as coloured). If the colour temperature of that continous light source is higher or lower than the conventional 5600K, white balance will shift the spectral sensitivity curve to match the colour temperature and thus provide a final image, that will be corrected to the 5600K convention (if we choose not otherwise).

If the subject is illuminated by a light source with only some strong emission lines (as is the case with those LED PARs under discussion in this thread), many coloured surfaces will not receive the wavelengthes representative of their surface colour and thus cannot reflect them. They will appear in dull and different colours, than it would be the case under norm light.

White balance tries to correct this to a certain degree, by adding the Green-Magenta shift correction to the simple colour temperature shift. But with only few emission lines available, this is not possible, or at least not to the full extend. The result may be an image, that can be perceived as having the correct whites and blacks, but nevertheless certain colours will be missing or off.

Ofcourse there are some post-processing technologies available, that make use of fuzzy logic or AI algorythms to reconstruct these missing or off colours. You can use face detection and then apply a correction algorythm to correct for skin tones automatically, as some software seems to do.

My point only is, that this is today not possible to the full extend and continous light sources are much better (aks easier to correct, if correction is necessary at all) than line radiators. That may change someday. But still: If we consider, for instance, subjects with strong pure colours. How should any algorithm reconstruct the colour, as perceived under norm light, when the subject's colours fall completely into the emission gaps? No way, to do that correctly.

This whole, lengthy discussion (from my side) only served the purpose to illustrate, that emission line based light sources pose much more severe problem for colour balancing and correction, than conventional temperature radiators do, and that obviously, we have to deal with two very different lighting concepts.

Ben

[falconeye]

My point only is, that this is today not possible to the full extend and continous light sources are much better

Ok, what you seem to mean here is this: Typical continous light sources have a spectrum with better color rendition than typical discrete light sources.

I agree. The confusion is that you didn't say it. You actually said that a continous spectrum is better than a discrete one by definition. Which simply isn't true. And you may never have meant it, actually.


And again, as for the eye outperforming a camera. It doesn't. If you would take a 360° camera shot (w/o any white balance actually) and fully immerse the viewer into the image (using some fancy 3D glasses or better yet, a CAVE), the brain would identically adapt to the wrong colors as if it would have been immersed into the wrong light spectrum in the first place. There is nothing the camera does worse than the eye (except if it uses filters with inappropriate spectral sensitivity curves).

If light renders bad colors, it does so both for eye and camera.

[reeftool]

LED lights are turning up everywhere. They have almost completely taken over the automotive market. With very long life and low power draw, they are now available as replacements for standard lighting. They have a noticeable color difference in the light. While I don't shoot concerts, I have read through the thread with interest because I think all of us are going to be dealing with this issue very soon in our photos everywhere, not just concerts or holiday lights.

[falconeye]

LED lights are turning up everywhere [...] They have a noticeable color difference in the light

Yeah, I've purchased a liquid-cooled white led light bulb to replace a broken bulb in my cave. And I must say, my brain just refuses to do a proper white balance

[UnknownVT]

LED lights are turning up everywhere. They have almost completely taken over the automotive market. With very long life and low power draw, they are now available as replacements for standard lighting. They have a noticeable color difference in the light. While I don't shoot concerts, I have read through the thread with interest because I think all of us are going to be dealing with this issue very soon in our photos everywhere, not just concerts or holiday lights.

Thanks for that input - it is a very good point.

White LEDs are mostly blue chips and color adjusted/balanced with yellow phosphors (LED lamps Wikipedia) and to our eyes the light is "white" - the more efficient acceptable tint of white LEDs not surprisingly tend to be the cooler/bluer whites ~6500K (although to my eyes they seem to be higher - this is just a guesstimate) - my other interest is flashlights (hence the cross-references to CPF) where white LEDs have now started to become dominant because LED flashlights are now much brighter than regular incandescent to the point where today on average they can be twice as bright as the once considered ultra bright xenon bulb powered by lithium flashlight like the legendary SureFires (please see: Puny LED flashlights (Not!) + COLOR RENDITION Comparison )

also please see Really bright LED streets lights in town over at CPF for another application of LEDs

Anyway back on topic - white LEDs do not pose quite as obvious problem as using discrete red, green and blue LEDs to make "white" because of all the foregoing discussion - since discrete RGB LEDs produce in reality a spectrum that is very peaky with deep troughs in between.

Although having said that white LEDs have uneven spectrum with spikes - so that its Color Rendering Index (CRI) is pretty poor at best about CRI=70. This is even worse than fluorescent lights which on average are about CRI=82 - and most people know how troublesome fluorescent are to photography. (Point of interest the CCT of fluorescent lighting is also adjusted/balanced with phosphors.)

White LEDs are being reconsidered in terms of traditional CRI and new ways of measuring the CRI for LEDs are being proposed - the reason is that white LEDs appear to be a bit better than their poor overall CRI rating.

LED lighting need not be the harsh cool-blue light - one used to associate with fluorescent lighting - which someone very aptly called "morgue white" ....

Have people noticed that fluorescent lighting doesn't seem to that unpleasant anymore - with the popularity of energy saving of CFL (compact fluorescent lighting) spirals -
one can get CFLs at 2700K (soft white), 6500K (daylight) and more recently 5000K (sunlight) and even 4100K (cool white) now from GE - which means they are now in the mainstream. They do the CCT balancing with phosphors

Similarly white LEDs need not be the cool-blue'ish white of the typical 5mm domed LEDs - because of the use of phosphors one can get warmer tinted "white" LEDs - currently gaining some popularity are the "Neutral White" LEDs in the range of about 4100K and there are the warm LEDs that tend to imitate tungsten lighting at a range around 2700K.

In that way to use an old saying "the future looks bright".

But getting back to photography the poorer CRI and uneven spiky spectrum of white LEDs will pose some problems for more critical color balancing - although I am guessing to the majority things aren't going to be too bad - as the camera does capture more or less what we see - it's only the more extreme lighting as in the current stage lighting using only Red Green and Blue LEDs that we are seeing the more obvious problems - I am guessing that eventually we may see some true White LEDs installed for stages - hopefully they'll be in the "Neutral White" tinted range of about 4100-5000K rather than then harsher cool white of 6500K.

There is a lot of development and research work on white LEDs for commercial and domestic lighting where color rendition etc are of importance - so when LED lighting comes into the mainstream things have to somewhat acceptable - otherwise there the inertia will not be overcome and the majority will not accept them.

For example discrete red, green and blue LEDs can get almost any tint of white and probably much more economically than any balanced white LEDs - so why have these not been even suggested for the home or commercial lighting - or even street lights - it's because they give pretty horrible color rendition (CRI) due to the very peaky uneven spectrum with deep troughs - we might see the light as white - but that combination probably will give us headaches pretty quickly.

So I've managed to come full circle -
there is a (big) difference between white with a more even spectrum like in noon daylight - and the "white" made up of discrete RGB LEDs.

[kxr4trids]

Rather tangentally, has anyone else viewed black and white chromes projected in a dark room and "seen" color in them? Or looked at a black and white print, and "known" what color a particular object was.

While film/sensor has difficulty reproducing a color that isn't there, I'm wondering if it matters to the final viewing anyway ... So I can look at a red car under a HPS streetlight and it will be gray, or I can photograph it and it will be gray. But when you reintroduce the human element and I perceive the photograph, won't any processing/interpolation that I would have done "live" also happen when I view the reproduction?