Spectral emissions from modern light sources (more graphs added)

color

#1

The last few days I have had problems with a “studio shot” (please read a desktop arrangement); a minute blue flower became far too red/violet when shot. I began to suspect the studio light (please read desktop lamp) and started to investigate what spectra our modern light bulbs/tubes really emit.

Using @gwgill’s illumread and a ColorMunki Photo, I produced graphs for every kind of light source that was available in our house. Please note that the y-axis differ, depending on the “wattage” of the light source. The x-axis shows wavelength.

The Natural Group
First, the spectrum that the Sun emits in Southern Sweden, at around noon:


The “wattage” of the next one is very easy: it is 1 cd. Here is the spectrum from a candle:

Incandescent bulb
A good old incandescent light bulb:

A halogen light bulb:

The LED group
A modern IKEA LED bulb:

An IKEA 1000lm LED, named Ryet:

A weaker IKEA LED bulb (which I used for the original blue flower shot):

For comparison, a cheap 24-LED panel:

A recent filament LED:

The Low Energy fluorescent tube group
A Grafilite studio low energy fluorescent tube:

A low energy fluorescent tube, specified at 6K5:

A low energy fluorescent tube, manufactured around 10 years ago:

Luckily, none of them turned out to have any spectral emission gaps.

Have fun!
Claes in Lund, Sweden


Camera spectral sensitivity data
(Mica) #2

This seems like quite the rabbit hole!


#3

Wait, which of them belongs to your desktop lamp?


(Morgan Hardwood) #4

You really need to pop by for a cup of coffee, and bring your monkey friend.

Such hue shifts could result from the matrix optimizer or from the tone reproduction operator in your input profile. If you use RawTherapee, try turning the look table off. Of course assuming your input profile was designed for the light source you used.

I’m curious to hear more.


(Aurélien Pierre) #5

This is most likely a spectral issue. If your light spectrum is not complete and even, a simple RGB correction will never suffice, since you need to “fill the spectrum” between the spikes that show in the graphs, and you could also end-up with metamerism errors, upon which software is completely powerless.

Buy yourself a proper daylight-balanced light source. In photography, there are dozens of things you can spare money on, but the light quality is something that does not suffer trade-offs. Moreover, the whole opensource color digital pipeline relies on the assumption that you are using D illuminants (daylight), so you have no way to properly adapt to other spectral distributions. (For now). You can do all the maths you want, the physics are wrong in the first place. The whole concept of color temperature implies a black body emissivity, so discard anything but sun light and laboratory sources. It’s foolish to correct artificial lighting with a color temperature adjustment.

Plus I’m pretty sure 99 % of the LUT people use around here are doing more harm than good. You can’t do a properly lit exposure of a color chart outside a metrology lab, so every LUT produced in your garage will fail on you and, worse, will make you feel overconfident at the same time. *

The spaces between the spikes you get are just as bad as gaps.


*During winter, in Canada, 4×4 vehicules have higher chances to have accidents due to snow and ice. In Europe, the 3-wheels BMW scooters with hard-tops lead to more severe injuries in average during accidents than regural motorcycles. In France, at night, the lit sections of suburban roads have higher accidents rates than the non-lit ones.

The cause of these 3 statements is people driving safer vehicules or in (seamingly) safer conditions feel more confident, then get more reckless, take more risks, then get more injuries than average.

LUTs and input profiles are exactly 4×4. You feel they are accurate because you get a lot of samples and some magic going on (the less you understand it, the more you trust it, it’s amazing). You feel they will get you out of trouble no matter what. Think again. Work properly and don’t rely on software to clean-up your sh*** for yourself.

Buy yourself a daylight-balanced light.


#6

Do you have any recommendations in that arena?


#7

That’s interesting. Do you mean LUT monitor profiles also belong to the 4x4 realm? (DisplayCal seems to do miracles with my not-so-good monitor, at the expense of not-so-short profiling time)


(Hermann-Josef) #8

@Claes I recommend taking a look at the excellent book by Giorgianni & Madden (2008) to graphically see the influence of the light source.

Hermann-Josef


#9

More graphs added in original post.


(Aurélien Pierre) #10

Philips does fairly good balanced LED. I find them around 16-20$ CA (16 times the price of your regular incandescent stuff, but they are supposed to last about 20 years or so). That’s the kind of stuff you should (at least) have in the room where you do your edits, by the way.

Then, you can see what your local photo dealer has to propose.

I was mainly speaking of the LUTs you do with your camera and your color checker. Such color checkers are impossible to lit evenly and with controlled reflection with general audience means. So you end up with exposure gradients over the chart and the resulting LUT/profile can’t be anything but wrong. Xrite Passeport Color Checkr address that problem with a small chart, easier to lit evenly, but with too few patches to do a proper LUT, you can only use it to do basic averaging adjustments.

It’s a bit less bad for screens LUT because you get the sensors directly on the screen, and the emission is controlled by the soft. But still, there are known issues of the GPU gamma LUTs (VCGT) being dropped randomly (on Windows and Linux), so you have to be careful about that.

So, yes, most of the time, when you apply a LUT on something, you are fooling yourself.


#11

Salut!

Risking to dive into another abyss/quagmire:
but those Philips LEDs you linked to are just
spec’ed at CRI 80. And that is b-a-d.

For those who want to read more about CRI:
https://www.waveformlighting.com/high-cri-led

Have fun!
Claes in Lund, Sweden


(Aurélien Pierre) #12

They are far from perfect, but still better than Ikea’s and low-end competitor, and decent enough. Otherwise, you will have to get proper strobes and flashes, or LED panels, designed for photography (not in the same price range).


#13

@aurelienpierre & @Claes thanks for the responses. I’ve ordered some free sample LED strips. Hoping they will send one of each of the daylight temps. I now remember reading about CRI when I was looking at flashes a few years ago.


#14

Salut, Aurélien!

I am presently measuring CRI of some of the light sources
that I have available.

I will probably be able to show the results within a day or so.


(Andrew) #15

I use an xrite passport and might be the over-confident user (but then there are only so many hours in the day to get to grips with everything photographic…)

You mention gradient across the checker @aurelienpierre, and I can see that would alter brightness relationships, but rightly or wrongly I thought these input profiles were mainly about colour correction. Would that not be independent of any (minor) illumination unevenness?

Also, why do we talk about profiling a camera rather than a camera+lens combo? One lens can impart a different colour “cast” to another, I believe, hence they sell matched sets of video lenses.


(Andrew) #16

Just looking at my flourescent indoor photography bulbs. Androv “daylight full spectrum bulb”. 32w 6000k 95CRI 1800 lumens.

@Claes, good stuff!


(Aurélien Pierre) #17

Luminance is part of colour. Unless you are using a top notch color adaptation model, able to decorrelate hue, saturation, chroma, luma, etc. but there is no such thing in opensource, and the ones available in the litterature always fail one way or the other.

IT8 charts have grey scales all around, so it would be possible to compute a 2D gradient map of the exposure and compensate it with a mesh (that’s a partial differential equation to solve numericaly in 2D, finite element analysis do that quite well), but I don’t know anything about a such thing in opensource. And even so, the numerical solver is not garanteed to converge, and the general audience will not have the knowledge to tweak a solver to get a proper solution, so it’s worth nothing.

And yes, technically, we should profile cameras and lenses, but that becomes quite cumbersome, and will only serve to get more inaccurate profiles.

You better use a simple RGB matrice. That’s not super accurate, but the chart patches get averaged so the possible errors get mitigated.


#18

I have always taken color photos with no particular techniques other than adjusting WB in post-production, now I am trying to familiarize with a ColorChecker Passport (boy, scared by color management I didn’t expect the major difficulties were in opening and closing a ColorChecker instead!)

I use the software that came with it to create camera profiles, but under normal lighting the profile created doesn’t add much to the default one for the camera model, so maybe I think it’s not worth the hassle. But it is in the white balance point that I find the most benefit, and it is just fantastic.

Of course if the light source is not good enough, I think results may be by chance. So I’ll support the invitation to the use of a good light whenever possible.


(Andrew) #19

Here is one of my profiles created with xrite and the supplied software like you @geldo.
Can anyone say whether this is a simple RGB matrix, or a Look-Up Table, or anything else pls?

@geldo, I generally find the reds and blues become a bit more saturated with my own profile.EOS 6D 50mm sunshine June.dcp (1.5 KB)


#20

@RawConvert Yes, your profile seems to give more saturated colors, I don’t know why (mine is Nikon D3100 - 50/1.4)

But I think we are a little off topic here :thinking: