Is it possible to (roughly) simulate a tri-color UV/IR image by processing a standard UV/IR image?

[Stefano]

I had this thought for a while. We have already briefly talked about this in the past (below are some links), but I never saw an actual experiment about it (if someone already tried this and posted it on the forum, please link your experiment).

 

So, to put it simply, our cameras have a limited but non-zero ability to distinguish different wavelengths in both UV and IR, which appear as different colors, and this is especially apparent when an image is properly white balanced.

 

In UV, the usual color palette (after a WB) starts with blue at the longest wavelengths (usually near 400 nm, depending on the filter being used), then violet/lavender, then white (at a precise wavelength), then a greenish yellow, and usually around 340 nm there is green, but in real world photos UV-green is almost never seen, unless you have very specific materials such as ruby. If you are interested in a discussion about UV-green objects, read this nice talk members had.

 

In IR (using a 720 nm filter, like Hoya R72 or similar), the palette is quite similar: we start with a yellowish/orange at the 700 nm edge, then yellow, then white, then cyan/blue. I just talked about this here.

 

This means that, both in UV and in IR, the shortest wavelengths appear yellow, and the longest appear blue. But there are really two channels, as Andy showed here. In particular, there isn't the equivalent of a green channel. We have yellow and blue, two complementary colors, they give white when mixed, but that's it. The fact that we have two channels only also explains why there is always a neutral wavelength which appears white. I read on Wikipedia that colorblind people also experience this.

 

BGR images (images in which the red and blue channels have been swapped) map longer wavelengths as red and shorter wavelengths as cyan. This is, to an extent, something similar to a true tri-color image.

 

Regarding true tri-color images, I recently experimented with this technique in IR here, and before me Bernard Foot did the same in both UV and IR. I suggest you to see his work if you are interested in the technique. This images have actual three channels of information, and so it is possible to see red, green and blue objects in the same image, although green seems to be the rarest color.

 

My experiment is to compare a true tri-color image, which was made as described in my initial post, and a simulated version done with more usual methods.

 

Cameras are much less sensitive to wavelengths above 900 nm than to those below, so I needed a light source heavily weighted towards longer wavelengths, and running halogen lamps at low voltages didn't provide enough light. So I used my DIY incandescent lamp (this one), to provide enough light for my purposes. I ran it at about 30 W of power. The resulting images had too much noise and were quite dark, so I took 199 of them, and did a mix of stack/average (sum/50), to have an image 4 times brighter and 50 times less noisy. I did an in-camera white balance before taking the images, and used a Hoya R72 filter.

 

This was the result:

 

Not the best image in the world, but nice enough. The green color cast and the vertical stripes are a result of the noise at low brightnesses.

 

Then I re-white balanced it, swapped the red and blue channels, and increased saturation to the max twice.

 

...and this is the final result:

 

I got blue water and some reds. Then I did the proper tri-color version, and this is how it looks like:

 

This is the visible light reference:

 

For sure, the two images are not identical, but are quite similar. Also, on the Rubik's cube, the green squares became orange and the blue ones became yellow in the tri-color image. I think I can see a slight difference in the simulated tri-color image too:

 

Visible reference:

 

True tri-color:

 

Simulated tri-color:

 

Isn't the square in the corner a bit redder? I may be overseeing stuff, but I have this impression.

 

Conclusion: in IR (probably also in UV, but I only checked in IR), doing a BGR channel swap on a white balanced image can give a clue of what a true tri-color image would look like, although a true tri-color image can only be done using three filters or three light sources and combining the resulting channels properly.

 

Other occasions in which members, including me, talked about this (I may miss something):

 

https://www.ultravio...dpost__p__39254

https://www.ultravio...dpost__p__36801

https://www.ultravio...dpost__p__25092 (the post contains some links).

[dabateman]

If you take a UV photo with a UVB light, like the ExoTerra UVB bulbs and a Broad band filter (like 330WB80 Improved filter). The result looks very similar to merging 313nm, 370nm and 390nm bandpass images. As the light is similar and the cut offs of the filter add to change the results. I saw this when Bernard first started combining images in Gimp and I was following allong with my filters.

[OlDoinyo]

If you want to try to get closer to the reference result, you could try superimposing some hue remappings. I don't know what software could do this in a systematic way, but a crude example exists within the Hue adjustments in Photoshop.

[Stefano]

I think (I should check) that IrfanView allows you to do that. I never tried it and I don't know how it works.

[Stefano]

Actually, water appears yellowish even under normal sunlight (snowy day).

 

Original:

 

BGR:

 

Increased saturation:

 

You can also see a color shift of my lens.

[Stefano]

I just tried the Rubik's cube, the face on the left in the above images, the face with the green and blue squares. There are some reflections which may affect the results a bit (you can see the camera reflected), but I didn't see any difference in the false color, while there is a noticeable difference in tri-color IR images.

 

Original:

 

BGR:

 

Increased saturation twice:

 

Remember, the square in the top left corner appears orange in tri-color IR images, while the others are either white or yellow.

 

I thought maybe a visible reference could have made things clearer, here it is: