Colour rendition of LED lighting

[paulears]

I'm a member of the Guild of TV Cameramen, and much has been spoken on their forum about the methods of measuring the output, colour wise of light sources. Some of the members are very experienced BBC engineering people - the ones that design the camera setups to enable closer match of shots. Very clever people I hold my hat up to. They've also been talking about how LED lighting of the colour mixing types are pretty awful, and they feel that theatrical lighting designers should spend their budgets on equipment with decent colour spectrums so their cameras work properly. I've been explaining our point of view that all we really need of a camera is that the colour we see with our eyes is the same in video one photos. No good lighting things in magenta, if the cameras see blue.

 

 

Lots of too-ing and fro-ing of ideas, so I came up with the idea of using an LED mover, and pointing cameras at it. I grabbed all the things I had around me.

Pentax K100 DSLR

Pentax K5 DSLR

Gopro

Chinese version of the go pro - D4000

Panasonic SD9 - a small consumer handicam ⅓" chip thing my wife uses a lot.

 

I took video clips into Premiere, then saved out individual .jpg frames, same as the two Pentax DSLRs, which were not taking video, just stills.

 

Then I turned them into colour bars in Photoshop - a little more care would have produced better aligned bars, but I ran out of time.

 

The results are very strange - all the cameras have colours they don't see very well - only one camera managed the yellow, and blues and cyans are very variable.

 

I'm not really sure what it proves - perhaps nothing? However, it does show that any carefully planned lighting design can be severely mangled by the cameras.

http://www.limelight.org.uk/bars.jpg

[timsabre]

The cheapest camera did best!

 

I think this just proves that you either design your lighting for the human viewer, or you work with a wide shot on a monitor and design for the camera (and hope all the cameras can be made to look the same)

[J Pearce]

Paul, interested in the methodology of the test. Were the cameras pointed at the lens of the fixture? If so, it could be interesting to repeat by filming the light reflected off a piece of white card (possibly white balanced first?) to see if incident and reflected light give different results.

[Ynot]

Interesting.

 

Question, however...

 

Obviously you could set exposure/aperture accurately and identically on the DSLRs, but what about the others?

Surely colour rendition can (and will) vary significantly with different optical settings on the recording source?

 

 

[paulears]

Test conditions quite simple. Led fixture two metres from white screen. The rgbcmy programmed in, with full or off brightness, plus gold and orange shades just to see what dropping the green did. Cameras alongside the fixture with the image defocused, and no autofocus. Exposure on all set to auto, to simulate typical conditions. It only seemed to get it wrong on white, which was a white led source, not a mixed RGB one, the fixture being rgbw. So pretty much the cameras seeing what was in front of them. One of the odd results was that the K5 which I bought to replace the K100 for stills, in use seems to handle magenta badly, but not that badly in the test.

 

As I said, I'm not really sure what it proves if anything, but it certainly shows that left to their own devices, they mangle colour badly. Perhaps I should repeat it with a tungsten source and gels?

[J Pearce]

A tungsten comparison would certainly be interesting. PAR and gel, match the LED colour by eye, then compare how the camera reads both. Ideally you'd want intensity matched (to remove over-saturation/over exposure as a variable), so maybe a lower power tungsten-halogen fixture.

 

 

For extra interestingness deviate from a white surface to demonstrate RGB insufficiency.

[Judge]

I dont know if it will ever be possible to get accurate colour with LEDs because they do not emit a broad spectrum of light. Only the colour, or mix of primary colours you tell them to and this does not equate to anything that is accurately measurable. Human eyes have had years of evolutionary experience of adapting to light levels and colour ranges and our brains are very good at assimilating information and auto correcting it. Try staring into a red light source for 5 minutes, then look away at a white light and it will look blue because your brain has adjusted your white balance for you. After a few minutes it will reset back to normal.

Cameras have no such brains, only a primitive set of rules that apply sensor sensitivity and processing modifications that are dependent on what you tell them White actually is. Of course some have better AWB than others. And that is the key thing - there is no WB setting for LEDs. Some people say that it should be set the same as for fluorescent but that really off the mark, depending on the kind of fixture. Personally, when shooting stills involving LED I will shoot in RAW using AWB and then correct in post.

With video the best you can do is to use cameras that have the highest possible colour subsampling, white balance from known source keylight and go for it and correct in post.

I really dislike LED as a lightsource, certainly for facelight, and even find it unpleasant for colour washes but of course recognise that its not going to go away because it fits two of the criteria of consumer lightsources very well. It has low energy consumption and has long lifespan (and it will become cheaper) The fact that it makes horrid colour cast, well most people do not know or care so as manufacturers are interested in selling to the masses this is just a compromise that we are stuck with.

As far as professional use goes this is being looked at and is why ETC charge so much for their LED units, as they do look very tungsten-like. Quite how well this translates in terms of WB I do not know. I am guessing that we will start seeing cameras that are better at dealing with an extreme mix of light sources, but they can never deal properly the the nasty pinky light that you get from shonky Chinese LED Parcans.

Paul, I am a bit confused at how you did your test - I would be inclined to maybe do a repeat and this time point the lightsource at a white sheet, set camera to AWB then see how they all cope. It will be a different story, but probably still a disappointing one.

[david.elsbury]

Er, Judge... did you read Pauls post above where he said he pointed the lights at a white screen? And that the cameras were set to AWB?

[Judge]

Err yes, I somehow did totally miss that - thank you.

<replaces foil hat with dunces hat and goes to open a bottle>

 

</done>

[paulears]

The camera people triggered this test because they constantly complain about the theatre style lighting their cameras see badly. They also want a continuous white that has a defined colour temp. I was interested in how some States on stage just look wrong so a simple test looking at a led fixtures output would be like looking at a printed colour chart, and it just doesn't work right at all! Just didn't expect so many differences.

[alistermorton]

Martin Swain from our theatre has a lot of experience in this area, and yes, it's not a simple problem to crack. I had an interesting conversation with him a while back about how LED suffers from the same problems as discharge and fluorescent lighting, but in different parts of the spectrum, due to the way some LED colours are generated (some colours are done with a mix of phosphorescent chemicals, and the spectra aren't continuous).

 

He's occasionally on here as cyclight, but I don't think he's been on for some time.

[p.k.roberts]

There's quite a bit of interesting info relating to this on this EBU site;

 

https://tech.ebu.ch/tlci-2012

[paulears]

Indeed, and I've been talking to Alan Roberts who did those tests about it, but he's quite clear that for accurate reproduction of colours the clear result of those tests is to get better lights, but where we don't quite fall in line is the problem of how colours are processed. The majority of the equipment tested is trying to produce white light, where CRI is clearly an unreliable method of measurement. The results of his tests are really useful, for those wanting tungsten equivalence, but my area of interest is not in white light, or a replacement for tungsten - just in how common saturated light is perceived by cameras. As so many smaller and cash strapped venues are replacing their kit with LED, I'm sure the camera problem is getting worse as more people realise something is adrift - spending a couple of grand per fixture instead of hundred isn't going to be a workable solution. It's spurred on by dance mainly, where costumes will look different on the supplied videos people buy. Pink looking blue, or dark blue looking cyan isn't very satisfactory really.

[GaryNattrass]

Alan Roberts is the man and I respect his testing a great deal, he has even assisted me personally when I made the switch to panasonic P2.

 

I think there is a certain element of cheaper cameras and lights being used but then people still expect them to perform just as pro kit would, the same goes for sound kit to a certain extent.

 

A lot of it is manufacture lead in the quest to sell kit and you see this all the time with specs being manipulated to suit the market and all sorts of claims that are subsequently found to be untrue or massaged for marketing purposes.

[musht]

Anyone catch this last month about whether blue is a new concept:

 

http://uk.businessinsider.com/what-is-blue-and-how-do-we-see-color-2015-2?r=US

 

Colour Rendering Index really dates back to the 1930`s with updates, given the industry plenty of time to work out how to manipulate the numbers...