various light sources on a colour checker card

Recently in the dedomed Nichia 219 thread I experimented with a colour checker card to get an idea of how colours appear with different types of leds. The colour checker is not exactly meant for this kind of application, but I believe it is a nice way to see not just the tint of a light source but also what it does to the different colours.

I found it was difficult to get it right, had to digitally correct the images to match what I see in real, and was still not pleased with the results. Not only I did not get the colours right, but the dynamic range of computer monitors is not sufficient to show everything that is going on in real life (the brightest colour tones have much more 'colour depth' in real life, clear differences between leds in real life show less on screen.

I have now tried a different camera (a Sony DSC W200) and that went much better. Although the dynamic limitations of computer images and monitors are still valid, at least the colour tints on screen matched the real tint well without digital corrections. I find it a very good camera myself, Zeiss lens etc., but it is certainly not comparable to a good reflex camera with big sensor, and it does not produce RAW images (I do not own such a nice camera :-( ). Nevertheless I was so pleased with the results that I want to show them to you, I added some more light sources above the ones from the dedomed Nichia219 thread.

One thing that you must be aware of when viewing these colour charts is that LCD-screens (most of you will use one) are very sensitive to viewing angle: a slightly different angle will already influence not only brightness, but also colour tints. Two identical charts placed on a different hight on your screen will look different. (The oldfashioned cathode ray tube based monitors do not suffer from this)

Some other considerations...

All setting on the camera were fixed, exept exposure, I needed to vary that to compensate for the differences in brightness of my light sources (mostly flashlights). I did an experiment to check the influence of different exposures of the camera: I made two images with the same flashlight, one on a 25% PWM setting, one a 100% setting. The automatic expose of the camera did not fully compensate that, the 25% picture is just a bit dimmer. I then digitally corrected that (by just adjusting the digital 'exposure' setting of the Mac Preview application), and the the 100% image was very precisely reproduced:

Nevertheless in the following charts I did not do any digital corrections, these are the images just as they came from the camera. One exeption: the 9x5mm-led flashlight was so dim that I had to digitally adjust exposure a bit to make it comparable to the other charts.

The dark frame that contain the colour tiles also show a varies tone, depending on the light source, that influences the whole appearance of a given chart. Especially the daylight charts show a lighter coloured frame (because the angle of the lightsource compared to the chart was smaller). I guess it would have been best if I had made the frame the same black in every chart. I'm sorry that I did not do that, so you will have try to look at the tiles while ignoring the frame. EDIT: in the meantime I took the effort of making the frames black, in post #2 are the same charts with black frames , and that really helps looking at them :-)

EDIT2: I improved the daylight pictures that had a haze over them, and added the MT-G2 5000K and 2700K in post #16.

With all their limitations, these charts gave me already some insights I did not have before. One is that the Nichia219 has a warmer tint compared to daylight than I thought it would have, even direct sunlight is quite 'cool white'. Another insight is that some of us here (including me) are really really very spoilt tint snobs: as you can see, all lights, including the 9x5mm chinese generic el cheapo led light, gives off a white that shows all colours ok, what are we whining about? ;-) That said, this chart is not the one that shows subtle skin tones and other sensitive tint differences of nature, perhaps to compare those high CRI light sources a different colour chart that shows that would be better.

Please do not watch the charts before reading the above, the information they contain is limited and I probably have not even mentioned every flaw of them:

Here's a smaller version of the same chart, it compares quicker and because the charts are closer together, screen viewing angle is less of a problem:

I started a lonely thread again :-)

In the OP I mentioned that the viewing angle of a LCD monitor is of great influence on the colour tint. But i found today that that is true for my Apple Mac monitor but on the HP-monitor at work the colours are much less influenced by the viewing angle. So apparently it varies.

Further I took the effort to blacken the frames of the colour charts, and it helps a lot in interpreting the charts. So I incude it below. The frames of the charts do not look so tidy anymore, this is caused by the rather primitive way I used to make them black, sorry about that. But the colours of the tiles are unaffected. On the whole I think the charts have really improved:

And again a smaller version:

Ill keep you with company. :) Although, I can be quite critical at times.. :p I occasionally make posts/threads with limited feedback too. It happens with several people from time to time. Post a picture of a squirrel and you are more likely to get more response. Its a less complex subject.

I quite recently sent you a larger mail on this subject again. ;)

Ill get back to you on your latest experiment, until then. Free bump and company in this thread.. :)

This is great information. I’d love to see how the 90+ CRI XM-L2 and Cree bulbs stack up if you have them.

I just discovered a little trick that will go into my reviews. You can post side by side Imgur albums. This allows whatever to be compared side by side. I think that’d be perfect for this thread. Here’s an example:

You have to post change the “image format” to “simple post editor”.

Added it:

http://imgur.com/a/Xdh2y/embed

http://imgur.com/a/Xdh2y/embed

Nice work djozz. The XML-2 0D is the surprise package to me.

It’s funny how close the Nichia918 and the XM-L2 6A1 are. Study the pink squares in all the charts closely. I know in real life, the 6A1 has a very warm tint compared the the 918, but their color rendering on the chart is very, very close.

I just came across this thread and having done complicated setups for usable and demonstrable results i can totally relate to how much thought and work went into this setup.
These results will be very useful to me in guiding future purchases, so a big thank you from Bort for covering something that is hard to find elsewhere.

Not a criticism but a question, i am a bit surprised that the daylight pictures look washed out, perhaps because there are so many more lumens in daylight then a flashlight, it looks a lot like the pictures are calibrated to the nichia 219, is this the case?

Also what current outputs are the other LEDs, since tint changes based on drive current (i assume CRI may also vary but i don’t know by how much)

Do you take requests on future tests (if your willing to do any)?

Interesting that the “XP-G high CRI” image differs so much from the daylight images. That appears to show that not even Cree has these things well worked out and under control. It is clearly on the warm side. I like warm light for other reasons but don’t think light so different from daylight can really be called optimal for color rendering.

Thanks guys, it feels less lonely here now ;-) , thanks RaceR for the bump, I'm looking forward to more tint discussions.

Thanks leafty for the side-by-side albums, that is a neat way to compare the charts :-) . I use flickr for my photo's, I will try to find out if they have that option too.

Thanks for the thorough reply Bort :-)

Yes, I find it very difficult to to do it right, I'm still not finished experimenting how it should be done and what is too bad is that the best you can do will still not be very good because of the colour/dynamic limitations of camera's and screens. Because they do not show the colour intensity of real life, I even do not think that these charts are very adequate for a well weighted decision on which led to choose, but I guess it is better than not having them :-)

Yes, the daylight pictures do look a bit washed out, and I'm not really sure if that is just how daylight really is; unlike with the led-charts I can not compare the image on the computer screen side-by-side to the chart illuminated by day-light (my computer is not outside, light through glass from windows give a green tint over it, and it would be too much light on the computer screen anyway when outside). I did the small experiment with the two different exposures with the Nichia219 light, the camera gave close enough results but that was only a factor 4 in illumination. The daylight on the chart produced probably more than 50klux on the chart, the led lights somewhere between 1 and 5klux, that is a much bigger difference. So I am not done getting the daylight images right. I may try a neutral grey filter in front of the camera to get exposeres comparable to the leds. Thinking in that direction, if you reproduce the daylight colours on your screen, as I did with these charts, it is not a natural experience because real daylight is much much brighter, this dimmer version will probably be interpreted differently by your brain.

Most leds are driven reasonably hard (out of the top of my head, the xml2 0D: 700mA, Nichia219: 1.4A, Nichia dedomed: 1.4A, xml2 5B1: 3A, xml2 6A1 1.4A, xpg highCRI: 1A-ish, xpg2 3C: 1.2A) And yes, it will have an influence in tint, but when the current is not extremely low or high I hope it is not too much, at least I guess the difference will be not so great as the differences between the diffrent types of leds.

Yeah sure, I'l do requests, once the set-up is thought out it is not much work to produce these charts. I do not have every type of led in the world at hand though ;-)

very nice work djozz…
apparently nichia color cards don’t look that good.
in my eye and laptop screen, XPG 3C produce most neutral color.

+1

These images are mostly an interesting test. But to me, they show no information that can be used to help with a buying decision in real life use. More the opposite can be seen based on some comments here so far. People questioning if Cree knows what "high CRI is". Suddenly Nichia 219 does not look as good as some normally see it, or as its show in other pictures.. And 100 CRI daylight looks washed out..

This is all just the white balance settings in the camera, it is not adjusted to each emitter. Im guessing its set slightly above 5000K? Which means the further away the emitters are from this temperature, the more "off" they tend to look, high CRI ornot. This is just the nature of the white balance settings.

Djozz, I think the WB corrected versions I sent you show the differences better. But still, far from great, because there is no way to perfectly show tints and the effect of it to others via the internet.

Basically, if you want to compare tints. Auto WB on a white wall and various lights with various emitters on at the same time gives you a decent comparison. But only a comparsion between the tint shown, and that comparison does not tell anything about CRI, and it can also be slightly misguiding. :p Despite that, I think its one of the better ways show color differences between two emitters though. But it the viewer who is familiar with some of the tints will have an advantage in translating how they would look in real life.

If you want to compare CRI based on pictures. Make sure to have the right WB for each emitter, and I am not only talking about the correct Kelvin At least that is what I think is the best way to do it. But Im no expert.

If people want to compare temperatures, read the specs and learn how they look in real life. WB can make any temperature look just how you want it to be.

This comparison with a fixed WB of around 4900-5400K (?) is halfway a CRI comparison and halfway a tint and temperature comparison. And as you have said yourself it has several minor flaws/limitations. I will not try to to it better, but I will have to point out one flaw later. I think you did a good job, and redoing the pictures in an even better way will still have limited value to the viewer. Because what is really the correct setting anyway?

Just to show some examples to people who are less into the subject (again, not that Im an expert):

The "clock" picture I have seen several times here, It was originally posted on CPF I believe and then shown here at BLF? (to that someone, hi there, Im borrowing your pictures, thanks! :) )

Nichia 219 looks great. Mouse over for XM-L3C, which looks crap (especially because the WB, White Balance is off)

Now, lets see the same pictures corrected with the same automatic WB tool. Basically, the correct WB for BOTH pictures.

Notice how similar they are now! This is just both pictures done with a correct WB. Suddenly there is close to zero difference in how colors appear.

Does any of these pictures compare to what people see in real life. I feel that the corrected version shows the 3C better than it is. But the first comparison is also way off. Either way, none of the pictures are perfectly comparable to real life..

Lets take a look at some other pictures. These ware not supposed to be a CRI comparison. It is just a few shots out of a little beam comparison I did some time back. But I can use them to make a point.

UF M5 with de-domed CW XP-G vs UF M5 domed CW XP-G (basically same emitter)

I think the picture above shows fairly good the typical "slightly messed" up tint you can get with de-doming. Basically, its overly green-ish and yellow. I would say color rendering seems worse with the de-domed emitter. Based on the picture example above. But some might say that grass looks nice and green with the de-domed emitter. Which is true, but still, its off..

UF M5 with de-domed XP-G2 vs another UF M5 with Nichia 119 (WW 92 CRI)

WOW, yes. obviously WW Nichia 119 looks overly orange in the tint. IN THE PICTURE (it is warm in real life though) You almost gotta wonder if its "high CRI" emitter, right? Everything has a orange look. Maybe Nichia got the "high CRI thing off too" just like Cree??? Fact is, this is again just the nature of the camera setting vs the emitters temperature.

Now, lets take the same pictures as above, but with a corrected WB on both.

De-domed CW XP-G vs Nichia 119 on mouse over

As you can see, suddenly both of them looks perfectly white.. Not really any noticeable CRI difference either based on what I can tell. But none of the pictures above are anywhere near to what the eyes see. Both are obviously very neutral white looking without any hint of warmth or cool blue looking color or any distinct colors at all.

Another example with the pictures above, again different camera settings. This time, kelvin is set to 4900K for the dedomed XP-G. And 3600K for the Nichia 119. But the the WB tint (if that is a word) is not corrected like the picture above. Only the temperature is roughly correct.

I think the example above is one of the better in terms of CRI. But the WW 119 looks slightly less warm than it is. This technique (/setting) did not hide any of the overly greenyellow tint on the dedomed XP-G though, which is more real compared to the nice white beam above.

These three last sets above just shows the same pictures with different settings. Which is correct? All of them are, but at the same time, none of the pictures can really explain how these tints appear in real life. Which is the best buy. They can just give an indication. And that indication can easily be judged wrong by the viewer.


Lets take a look at a set of the pictures that djozz took, only WB corrected.

XP-G 3C (70 CRI) vs XM-L2 5B1 (80+ CRI) on mouseover

Extremely similar yes. But the the main reason I did a mouseover above is to see easily see the different light intensity through the two different pictures above. Just to point out one thing. The 3C is obvilusy much darker in the lower right corner compared to lower left. This makes the colors change too. When comparing subtle differences like we do in this test, this is a flaw that changes the appearance of the colors. You easily see how much darker the colors are on the 5B1. Is it the tint or CRI or is it the light intensity? Mostly light intensity I would say.

The effect above is impossible to avoid when shining a beam directly at the card. You would have to do a "ceiling bounce" or something.

Blacking out the dark part around the color tiles kinda helps with color perception (its a good reason why a cinema screen have a black frame), but it also hides the difference in light intensity throughout the pictures. Basically, it looks better, but it hides the flaws and kinda makes the comparison less correct by hiding the light intensity difference.

Just to compare what difference in light intensity does.

This is the XP-G 3C picture (with color profile change and corrected WB) , vs XP-G 3C with the same treatment + a simple manual exposure change on some of the picture. The purpose is to show how light intensity changes colors.

The mouseover is now more even in terms of light intensity, but its still much darker on the right side, so the colors through the picture are still off.

Edit: Sorry about the long post,spelling, language limitations etc.. :p

Thank you djozz. This is a very deep thread that I will have to read later when I can focus on it. If you ever get a chance, please add the popular MT-G2 emitter to the above. I think there is only one bin/tint currently being used by BLF members.

I have a feeling the MT-G2 will be perceived as one of the best looking emitters with the white balance setting that is being used. :)

I included a wallshot of the MT-G2 compared with some other emitters in the "tint mixing thread" several days ago. Might be interesting for people who is into this topic..

RaceR86, thanks for the critical comments on the charts and for the big contribution to this thread: in information and discussion , it makes me rethink everything and leads to more checking on what method would be best.

I agree, and already stated above that these charts have limited use. But if the pictures of the different leds, made by the exact same method, are placed next to each other, as above, they do give an idea about the differences between the leds, so I do think that the charts are of -limited- use for led-choosing.

I was hoping that these colour charts would also show the difference between low and high CRI leds, but like you I don't see that very well, some very low CRI leds show the colours surprisingly well (no one mentioned the good performance of the generic chinese 5mm led yet ;-) ), and the high CRI ones do not look outstanding, that is not what we experience. But -unlike you- I don't think that has anything to do with white balance settings, but rather with the inability of camera's to capture everything that your eyes capture, and with the limitation of the computer screen to show it like it is in real life.

The white balance corrections that you did, do help to more clearly see the failure of these charts to distinguish between high and low CRI leds

I have done a few experiments, one to show that using just the 'sunny daylight' white balance setting that I used throughout the pictures does show on screen the colours fairly close to how they appear in real life. Unfortunately to show that, I had to make a picture of that. So instead of looking at a comparison of real life to what appears on screen, you are looking at a picture of real life compared to a picture of a picture . It does show however that the colour reproduction is not so bad, and also that it is far from perfect (the tint in the grey series of tiles is off). The first picture shows on the left the colour checker card illuminated by the MT-G2 5000K (@2A) ( @ ImA4wheelr : I will add these to the chart collection above :-) ) and on the right onscreen an image of the colour checker illuminated by that same led. Not that bad a reproduction I think, although you can see onscreen loss of dynamic range and tint change.

Next is an image with on the left the colour checker card illuminated by the MT-G2 2700K 90CRI, on the right on screen still the image of the colour checker illuminated with the MT-G2 5000K.

And last an image with on the left the colour checker card illuminated by the MT-G2 2700K 90CRI, and on the right onscreen an image of the colour checker illuminated by that same led.

I show these comparisons to make clear the point that you must not change the white balance when photographing the light of different leds, rather you must choose a white balance setting that reproduces real life on screen the best and stick to that; if the white balance is good for reproducing the light of a cool white led, it will also be good for reproducing the the light of a warm white led. And leave the adaptation to the different colour temperatures to your brain.

Two more pictures, that have a rather sobering effect on my aspirations to reproduce real life, it shows that more than the colour tiles, the grey tiles show the deviations that the camera produces the best: two flashlight beams on screen next to the same beams in real life. I did choose two leds of which I knew that they do not reproduce well on screen (left my 3200K xp-g2 mystery led, right Nichia 219), but still the difference is awful, more awful than you can see in this picture made of it:

and if you could see real life: The mystery XP-G2 is much more red-ish than the Nichia (which is already slightly on the red side of white), but it photographs as very yellow (which it is certainly not).

So this post is about direct reflections off the colour checker card. I am still learning and I am sorry to say that this is one thing that was not fine-tuned yet. Ideally the tiles on the colour checker cards should show no direct reflections, but in practice they do a bit, this causes a haze over the picture with the tint of the direct lightsource. The surface of the tiles is matt but not perfectly so.

Bort rightly suspected that haze in the daylight pictures, those were made with illumination from all sorts of angles. I now redid the daylight pictures, but now I illuminated the colour checker through a hole in a cardboard box, so that the illumination angle caused far less direct reflection to the camera. And a big difference it made , sunny daylight is much closer to the Nichia 219 now, and further away from the cool whites.

The artificial light sources do not show much haze because the light was already shined on the card at a low angle (~20 deg. to the surface of the card). (RaceR86 pointed at the frame to show there was a lot of difference in illumination of the tiles from one to the other side of the card. But although there will be a small illumination difference over the card, the grey-to-black variation that shows in the frame -that has a rough but shiny finish- is mainly caused by a varying illumination angle from one side of the card to the other, not so much a difference in illumination intensity. The tiles themselves are very matt and do not suffer from that much).

So here is again another update of the charts, with two better daylight pictures (what strikes is that direct sunlight does not differ much from a sun-less lightly overcast situation). There has not been any heavy overcast rainy days since last week, I may do a picture then too. In the new picture are also the two MT-G2 leds added.

I looked into Flickr to see if I could make the albums like the ones that leaftye kindly made for this thread, but I did not find a way to do that so sadly no side-by-side albums for now :-(

...here am I floating round my tin can....

I just keep talking. I did a late-afternoon shot with a grey sky, and I must say that there is hardly ant difference from the other two daylight shots. Apart perhaps from sunsets and sundowns, daylight may vary enormous in intensity, it seems to hardly vary in colour spectrum, or at least in the way the colour checker card is recorded.

Djozz, so what do you get out of all this?

I must say your daylight pictures improved a lot! Great work! Those pictures now makes sense compared to the first set...

The last late daylight afternoon picture does not "pop" as much in the colors though, but temperature looks quite similar to the other daylight pictures. Any reason why?


I disagree. But I don't really see what your are trying to show with your examples. You don't get that "washed out" warm look with the proper camera WB setting. Im not saying warm should look white. But a camera would never use a daylight WB with say incandescent indoor lighting. Simply because the picture would look overly warm and kinda washed out in the colors, at least more than it really is. (Which I think the picture I quoted shows)

If you think your statement is right, it should work the opposite way too. Set your WB to "tungsten light" (should be around 3200K.) Redo test on a couple of different lights.. I bet you will see that the WW lights suddenly look much better, and the CW lights will look way more washed out blue.

Reason being the LEDs with a temperature closer to the WB setting will easier look better compared to when the WB is far off. There is no way to make all LEDs look "real life on screen" with the same fixed WB.

Im not saying there is anything wrong with the setting you are using (although that would depend on what you are trying to show people). Im just saying that the setting you are using will easier make lights around the 5200K range look better in terms of colors. Or to say it in another way. With that setting, the 3000K lights will look more "off" compared to real life.

What I think this comparison is useful for is to show people that 3000K 90 CRI cant compete with say, 4500K 90 CRI light. The warm light just doesnt have the same ability to render the various colors as good due to the extra amount of red/orange compared to a NW high cri light. This should not be a huge shock though, because warm is supposed to make colors look warm... Basically put a slight orange look on them. The problem is that your white balance setting makes the WW lights appear warmer and worse than they are.. It enhances the effect.

If you are using a fixed WB setting, 4500K would be a better compromise for all these images IMO, but all LEDs would still not look "real life" with that fixed setting... Espesially not the emitters that are 1500K away from that temperature...

Im no expert, and this is just my opinion/thoughts based on my knowledge and experience (which may not be good enough).

... hey, at least you got some company in this thread.. :) ^^

:beer:

Thanks for the company RaceR86, I just now notice your post, shame on me . About your comments on white balance: I think I still disagree, but I am not dead sure, and I'm no expert on colour capturing, just hobbying about (where are the hard core experts?! I guess on another forum..)

But because I still think I believe in using the daylight setting, and for the sake of keeping comparing the leds the way I started it, I'll stick to the daylight setting once more to show three different daylight situations. I have been waiting for more than a week for the friendly sunny autumn weather to finally stop and today some really foul weather arrived over the Netherlands :evil: :

800 milimeters of rain fell over Holland today, so I could do the one daylight situation I had not covered yet, dark grey overcast with rain:

So what do three different daylight situations look like next to each other? (left direct mid-day sunshine, middle light overcast with no sun, right heavily overcast with rain :

I digitally matched the exposure of the middle and the right chart to the left chart (I believe this does not affect the tint of the colours, just, well, the exposure), same three charts just corrected for the same exposure:

What I already noticed in a former post is just getting clearer: very different daylight situations vary in total light 'output' but do not at all lead to much different colour tints. Yes, a bit less red when it is raining heavily compared to direct sun-light, but not the variations you see with different leds, it is much closer than I thought.

1

2