Looking for a rainbow.

[Basic]

I'm looking for a simple way to generate a "rainbow". Or rather, a band with light of varying wavelength, but fairly monochromatic on each spot.

The simplest way I can see is to start with a white "sheet light", generated by a bright light shining through a slit.
Then let this sheet light shine on a (three sided) prism. The slit that generate the sheet light should be parallel with prism axis. The prism should be made of a material where the refraction index varies as much as possible for different wavelengths. To move the "rainbow", I could just rotate the prism.

I'd like the different wavelengths spread as wide as possible. But if it's hard to get a wide spread, I could maybe do with multiple "rainbows" next to each other (generated by mulitple sheet lights shining on the same prism).

Now the questions are:
*) How do I generate a bright sheet light without wasting 99% of the energy from the original light source?

*) Where do I find an optimal prism to split the light?

*) Is there an easier way? Maybe I've missed that there already exist thingys that do this? Maybe some kind of light-show stuff that is used at clubs.

Does anyone here have any ideas?

 

[Simon F]

*) How do I generate a bright sheet light without wasting 99% of the energy from the original light source?

Since you probably have to use an incandescent bulb (in order to generate all wavelengths), could you not control the power sent to the bulb? As the power increases, you'll move the "average" frequency up through the spectrum. Not ideal but it might improve efficiency a little.

*) Is there an easier way? Maybe I've missed that there already exist thingys that do this? Maybe some kind of light-show stuff that is used at clubs.

Do you really need genuine monochromaticity or do you just need it to be perceived that way? Could you get banks of (lots of) LEDS with varying wavelengths and mix the output to match the eye's perception of a true single wavelength light beam?

 

[horvendile]

*) Where do I find an optimal prism to split the light?

If you're not short on money I would talk to Melles Griot. In my experience they are expensive but know their stuff around optics, and they have office in Sweden.

 

[Basic]

Simon:

Yes, I was thinking of using a incandescent bulb for the white light. I assume fluorecent light has a few dominant frequencies. But I think it's enough to generate a light with a fix temperature so I get white light as a human perceive it, maybe with some extra IR.

I wasn't very spedific with the efficiency comment. What I meant was how to make (or get/buy) a simple reflector/lens system that makes most of the light from the bulb shine on the slit, while still having fairly parallel light rays. This is so the sheet light stays "thin" even at a distance from the light source.

And I want real monochromatic light at each spot, not only percieved as that for humans.

But thanks for the ideas. (Do you see any interesting uses for this rainbow light? )


horvendile:
Thanks, I'll check them out. But since it's a hobby project, the "not short on money" part might be a problem.

 

[Simon F]

Simon:

Yes, I was thinking of using a incandescent bulb for the white light. I assume fluorecent light has a few dominant frequencies.

Yes, they have 3 phosphors.

But thanks for the ideas. (Do you see any interesting uses for this rainbow light? )

Err... Testing human vision response.... sampling materials for better/more realistic rendering... building a TV system that can represent indigo/violet <shrug>

 

[Basic]

Number two were the closest: multi spectral photography.

Take lots of pictures of a still object while sweeping a rainbow over it. The hue detected by the camera is transformed to a wavelength, and thus which color channel to store the intensity in. (Note: now there's a lot more color channels than RGB.)
OK, the hue from one image might not be enough. But with the hue from all images, and knowledge about how far the rainbow has moved between the images, there should be enough information.

There's a lot more colors to see than what we humans can sense. I've often wondered how much we're missing.

Problem #2 is to visualize the result... :?


horvendile:
Ouch! $70 for the cheapest little prism that would do it. And that was a 25 mm prism that might be a bit too small.

 

[GameCat]

If you want to do multi spectral photography, wouldn't it be easier to put coloured filters in front of the camera lens? Would be easier to set up as well, since you wouldn't have to block all light except from the rainbow generator.

 

[Basic]

That was the first idea. But I doubt I could find filters with narrow enough bandwidth, and with center frequencies shortly apart in the visulal range. (Plus IR since many cameras actually can receive light in the IR range.) And even if I got a lot of filters, I wouldn't be sure about their filter characteristics. With the rainbow, I could get a good estimate of how monochrome the light is.

I also think it's easier to automate the photography with a video cam and a sweeping rainbow, than with lots of filters.
Shielding of other light wouldn't be a big problem for the stuff I was thinking of, but taking pictures of large scale stuff would of course be hard/impossible.

 

[Simon F]

I also think it's easier to automate the photography with a video cam and a sweeping rainbow,.

Silly question but do B&W cameras respond equally well to all frequencies?

 

[Basic]

I don't know. But by taking a picture of the rainbow shining on a white surface, you could get data for normalization. Then comes the question what a white surface is. But I guess a sandblasted piece of glass would work well (so it gets that "milky" look), since that white is just reflections.

Btw, I'm not thinking of using a B&W camera. I need the hue to classify the pixel to a certain color channel. I'd guess that a B&W camera is fairly good. But I think a color camera might be a bit worse, especially for the wavelengths on the R-G and G-B borders.
Anyway, it should be possible to normalize it.

 

[Simon F]

Basic,
I doubt a colour camera will do what you want.

I was thinking about this and I don't think you need to divide off portions of the rainbow at all. Why not just shine the whole rainbow spectrum on your object and take an image with your B&W camera.

Now MOVE your rainbow by a fraction and then capture another image etc etc. That way you end with N pictures with each section of the target getting a different band of wavelengths (and you use 100% of the light energy). You can then use a program to assemble the various pieces from each image (with interpolation perhaps) to get your final result.

 

[horvendile]

horvendile:
Ouch! $70 for the cheapest little prism that would do it. And that was a 25 mm prism that might be a bit too small.

Yes, well, had I understood that it was a hobby project I probably would have warned you with a bit more emphasis...

 

[Basic]

Simon:
That's exactly what I meant, except for the B&W part.
The problem with a B&W camera is that it's hard to know what multi-spectral color channel a certain pixel should be stored into - all parts of the rainbow looks the same.

I want the color camera to classify what part of the rainbow that lights a certain pixel (from hue). But the value stored in the multi-spectral color channel comes from the intensity from the color camera. So I'd pretty much blend the three color channels to make a B&W camera. Do you mean that there might be "holes" in the spectrum when recorded that way?

Have anyone here used a digital camera to take a picture of an ordinary rainbow? In that case, did the rainbow have a fairly constant intensity?


Hmm..
If the shutter is placed in line with the prism so that it lies in the planes of the colored sheet lights that exit the prism, then it could be possible to know the wavelength at each pixel if the prisms current angle is known. Thus, it might be possible to do it with a B&W camera. But since it's easier to find color cameras, I think I'll still go for that option.


horvendile:
It was still a nice find, even if I won't get something from them.
I noticed that they also sell filter sets. Like a set of eight bandpass filters over the visual wavelengths. Just $1200, what a bargain.

I remember that when I was a kid I had a pair of cheap crap binoculars. It was a fun toy, and it did its job. But it had horrible colour fringes. Just a prism out of that crap plastic would probably be enough for me.

 

[Simon F]

Simon:
That's exactly what I meant, except for the B&W part.
The problem with a B&W camera is that it's hard to know what multi-spectral color channel a certain pixel should be stored into - all parts of the rainbow looks the same.

I assume that you do a calibration pass first!

So I'd pretty much blend the three color channels to make a B&W camera. Do you mean that there might be "holes" in the spectrum when recorded that way?

Yes, that would be my concern.

 

[GameCat]

Also, you probably need a decent (read: very expensive) lens on your camera to get good results. A good lens will give you lots of light (probably important since I doubt your rainbow will be very high intensity) and minimise chromatic aberration.

 

[nelg]

Most B & W film is quite orthochromatic.

 

[horvendile]

Have anyone here used a digital camera to take a picture of an ordinary rainbow? In that case, did the rainbow have a fairly constant intensity?

As a matter of fact, I have done just that! Do you wish to see an example?

 

[Basic]

I assume that you do a calibration pass first!

How?
Maybe just sweeping a white sheet light from the same position that the rainbow is projected from, storing at what light-angle it's visible for each pixel. That, together with data on light angles for different prism angles and wavelengths could be used to find the wavelength on a certain pixel. But it seems a bit more complicated.
Was that the kind of calibration you were thinking about?

Yes, that would be my concern.

If that's the case then I agree that I'd need a B&W camera. I just thought that the different colour channels overlaped.


GameCat:
If the lens is good enough wrt chromatic aberration for normal photography, then it should be good enough for this.
Making the rainbow intense enough would of course be a problem. That's the reason I was worried about getting most of the original light source into the sheet light.


horvendile:
Yes, that would be apreciated. Either here, or by mail. (I'll check the mail account in my profile when I get home.)

And I noticed that Melles Griot had som interesting light sources, and cameras too. I don't dare to think about the price though. They're doing multi-spectral in one more dimension. For each pixel, and each incomming wavelength, check the intensity for each outgoing wavelength. (Accounting for fluorescense.)
http://www.mellesgriot.com/products/forensics/default.asp

 

[Entropy]

When you guys are talking about "camera", are you talking about using film as a capture medium, or CCD/CMOS? Makes quite a bit of difference as far as spectral response goes.

Lens chromatic abberation should be a non-issue.

Edit: I would suggest getting a cheap used or partly broken spectrophotometer. It uses a good lightsource, and turns a prism, blocking the resulting spectrum apart from a narrow slit, resulting in a very narrow band of wavelengths. Dismantle it as needed, and do a frequency sweep. If it's a cheap model, the slit (frequency band) won't be too narrow, giving reasonable output intensity (this is a problem). If you want to judge the overall spectral response of a CCD/CMOS device, diffuse the light - defocussing may be sufficient.

Although - why not simply look at manufacturer data for film or solid state capture media? Lenses should be quite negligeable in comparison, as long as you're in visual or close by.

Edit 2: Something similar to the above can probably be done for free if you know a high-school physics or chemistry teacher. Lightsources, prisms and slits are staples of high-school and undergraduate physics.

 

[Basic]

I'm thinking of a digital camera, or video camera. Do you know if there are any differences in CMOS vs CCD wrt spectral response?

Since it's going to need a lot of pictures, and then end up in a computer for fusion, it doesn't seem practical with film based cameras.