DSLR testing sites like DXOmark and Imaging Resource use HMI and LEDs for color

[ErikKaffehr]

Hi,

According to this article: https://www.imaging-resource.com/TIPS/TESTS/TESTS.HTM, Imaging Resource uses Solux lamps at 15.9V. That is continuous spectrum.

Best regards
Erik

The DXOmark use a LED lightbox, the Imaging Resource HMI bulbs for the "skylight" component in the "sunlit" tests, which IMO renders the results invalid and defies the purpose  of the color accuracy measurement. Any peaky spectrum is a NO-NO in my book no matter the CRI score.
What do you guys think?

[Alexey.Danilchenko]

The DXOmark use a LED lightbox, the Imaging Resource HMI bulbs for the "skylight" component in the "sunlit" tests, which IMO renders the results invalid and defies the purpose  of the color accuracy measurement. Any peaky spectrum is a NO-NO in my book no matter the CRI score.
What do you guys think?

I think that you need to approach this a bit more carefully - unlike Xenon sources for example, just plain reference to LED as a light source says absolutely nothing. Simply because they are all different, use different phosphors and pursue different goals. The quality and chromacity of the LED source is also very dependent on the current passing through it so a very good power supply is a must - i.e. what passes as good enough for kitchen light may not do well for colour applications.

As far as LEDs go - Yuji D50 LEDs are excellent light source for colour related applications (even their VTC 6500K LEDs are very good - slightly less flat but more powerful).

[daicehawk]

I think that you need to approach this a bit more carefully - unlike Xenon sources for example, just plain reference to LED as a light source says absolutely nothing. Simply because they are all different, use different phosphors and pursue different goals. The quality and chromacity of the LED source is also very dependent on the current passing through it so a very good power supply is a must - i.e. what passes as good enough for kitchen light may not do well for colour applications.

As far as LEDs go - Yuji D65 LEDs are excellent light source for colour related applications (even their VTC 6500K LEDs are very good - slightly less flat but more powerful).

There is reference to "Kiyoritsu color lightbox" on the DXOmark's site, no model whatsoever, I doubt they use Yuji (which still has a cyan dip and a green peaklet).

[daicehawk]

Hi,

According to this article: https://www.imaging-resource.com/TIPS/TESTS/TESTS.HTM, Imaging Resource uses Solux lamps at 15.9V. That is continuous spectrum.

Best regards
Erik

https://www.imaging-resource.com/ARTS/TESTS/SUNSOURCE2.HTM

[digitaldog]

since there was nothing there. I assume that's what you mean.

+1
Request for specifics in this thread (like those with respect to accuracy) fall into the same camp it appears.

[Alexey.Danilchenko]

There is reference to "Kiyoritsu color lightbox" on the DXOmark's site, no model whatsoever, I doubt they use Yuji (which still has a cyan dip and a green peaklet).

What's the point of speculating - just ask them for a reference? My point however was that simply dismissing all LEDs is not very wise.

[Jim Kasson]

I think Jim Kasson did present a list of cameras fulfilling the Luther-Ives condition. It is reproduced here:

There were some qualifiers and context. Here is the post where I said that:

https://blog.kasson.com/the-last-word/how-cameras-and-people-see-color/

Jim

[Doug Gray]

Jim,

It occurs that the setup Graeme linked to Spectrally Variable illuminant could be used to get much closer to L/I.

Since it uses a spectral range of computer controlled LEDs to shape the illuminant, this could be used to similar effect as the papers you have linked to. But a big advantage is the LEDs can be instantly changed enabling rapid, successive image shots. Beyond L/I, one could also make reasonably good estimates of the reflectance spectra at each location. This might make preservation of historical/museum artwork images cheaper, faster, and with very good quality.

[Jim Kasson]

Jim,

It occurs that the setup Graeme linked to Spectrally Variable illuminant could be used to get much closer to L/I.

Since it uses a spectral range of computer controlled LEDs to shape the illuminant, this could be used to similar effect as the papers you have linked to. But a big advantage is the LEDs can be instantly changed enabling rapid, successive image shots. Beyond L/I, one could also make reasonably good estimates of the reflectance spectra at each location. This might make preservation of historical/museum artwork images cheaper, faster, and with very good quality.

The LI criterion ensures that the camera sees the scene, lit with an arbitrary illuminant, the same way as the Standard Observer.

Changing the light source isn't going to change that.

If you mean that the objective is that the camera, using a experimentally controlled illuminant, sees the scene the same way as the Standard Observer with some other illuminant, then the light source referenced could help. If you are allowed three exposures, a light source capable of artibrary spectrum could satisfy that objective. But that's not LI.

Or maybe I misunderstand your point.

Jim

[digitaldog]

I'll direct this to the attention of those who may not have read it:


http://dougkerr.net/Pumpkin/articles/Sensor_Colorimetry.pdf



The colorimetric researchers (von)Luther and Ives10 showed mathematically that the outputs of a set of photodetectors will consistently describe the color of the light, regardless of the specific spectrum involved, if (and only if):
The response curves of the three types of photodetectors are linear combinations of the eye cone response functions l, m, and s (there being a couple of other requirements, some pesky stuff about orthogonality and so forth).

[Doug Gray]

The LI criterion ensures that the camera sees the scene, lit with an arbitrary illuminant, the same way as the Standard Observer.

Changing the light source isn't going to change that.

If you mean that the objective is that the camera, using a experimentally controlled illuminant, sees the scene the same way as the Standard Observer with some other illuminant, then the light source referenced could help. If you are allowed three exposures, a light source capable of artibrary spectrum could satisfy that objective. But that's not LI.

Or maybe I misunderstand your point.

Jim

Of course. It's the ability to take multiple exposures with different spectral illuminants which allows one to process the exposures to produce an image much closer to L/I and with flexibility on the illuminant. That is, one could process the exposures and produce images closer to L/I for relatively arbitrary illuminants. That would require maintaining each of the sequential images or, alternately, processing them to extract subsets that have minimal information loss via SVD.

[Jim Kasson]

Of course. It's the ability to take multiple exposures with different spectral illuminants which allows one to process the exposures to produce an image much closer to L/I and with flexibility on the illuminant. That is, one could process the exposures and produce images closer to L/I for relatively arbitrary illuminants. That would require maintaining each of the sequential images or, alternately, processing them to extract subsets that have minimal information loss via SVD.

If I were really going to build a camera that worked this way, I'd skip the CFA and go for a monochromatic sensor, and make three exposures per setup. Alternatively, you could set the light source to as close to single wavelength as you could get and make one set of exposures incrementing lambda by 5 or 10 nm after each one. Then, knowing the camera's native spectral sensitivity, you could compute what the scene would look like with an arbitrary illuminant. This might be practical for digitizing artwork.

Jim

[Jim Kasson]

Of course. It's the ability to take multiple exposures with different spectral illuminants which allows one to process the exposures to produce an image much closer to L/I and with flexibility on the illuminant. That is, one could process the exposures and produce images closer to L/I for relatively arbitrary illuminants. That would require maintaining each of the sequential images or, alternately, processing them to extract subsets that have minimal information loss via SVD.

By "closer to L/I" above, I believe you mean "closer to colorimetric". Right?

Jim

[Doug Gray]

By "closer to L/I" above, I believe you mean "closer to colorimetric". Right?

Jim

That would be accurate as well. If we add additional filters to the standard 3 in a CFA to N, any 3 of the filters will be a certain distance from L/I but using all N filters, in a linear combination, allows a substantial improvement and a much smaller distance away from L/I. That this worked quite well with only a small increase in filters was a principal result of the paper you linked in your blog.

Their multi-spectral approach allows not only better L/I approximation but could provide good approximations with different illuminants. A perfect camera that produces XYZ values, as opposed to spectral values, meeting L/I under one illuminant would only work for that one illuminant.

[joofa]

Alternatively, you could set the light source to as close to single wavelength as you could get and make one set of exposures incrementing lambda by 5 or 10 nm after each one. Then, knowing the camera's native spectral sensitivity, you could compute what the scene would look like with an arbitrary illuminant. This might be practical for digitizing artwork.

The camera that was mentioned earlier, and which is not available at B&H,  did that. Though, the light source was not tempered with. The camera had special filters to scan through a narrow bandwidth (~1nm) in the visual range.

[DP]

The camera had special filters to scan through a narrow bandwidth (~1nm) in the visual range.

ouch, those were that narrow-band pass-through filters ?

[joofa]

ouch, those were that narrow-band pass-through filters ?

Yes, specialized AOTF filters.

[ErikKaffehr]

Hi,

The Bayer pattern could easily be modified for four colours, as it has two green pixels. So, using two different green would be possible. It was tried but never caught on.

Best regards
Erik

That would be accurate as well. If we add additional filters to the standard 3 in a CFA to N, any 3 of the filters will be a certain distance from L/I but using all N filters, in a linear combination, allows a substantial improvement and a much smaller distance away from L/I. That this worked quite well with only a small increase in filters was a principal result of the paper you linked in your blog.

Their multi-spectral approach allows not only better L/I approximation but could provide good approximations with different illuminants. A perfect camera that produces XYZ values, as opposed to spectral values, meeting L/I under one illuminant would only work for that one illuminant.

[Alexey.Danilchenko]

My point however was that simply dismissing all LEDs is not very wise.

FYI if interested, SPD for Yuji VTC 5600K LED (with CSV file as well) - in attachments.

Their D50 series is even better.

[Tim Lookingbill]

FYI if interested, SPD for Yuji VTC 5600K LED (with CSV file as well) - in attachments.

Their D50 series is even better.

Alexey, what do those bulbs cost and where do they sell them? Thanks for posting the info.