Making an image with more than three color channels?

[Fandyus]

As you all are surely aware, there is a way to create white light with just red, green and blue light. But there can be more combinations then that, hence CRI. Methinks, surely this works the same way for color channels too? I think with four you use cyan, magenta, yellow and green. What I would basically like to try is simulate an image a hyperspectral camera could make by merging more than three channels. Say I take a picture of a subject with 8 bandpass filters, the shortest wavelength bandpass is at 340nm, the longest is at 940nm. The rest lay somewhere in between. I now have 8 images but as of now I only know how to merge three. The way I would like to do this is start with a channel that is fully red, then move to a channel that is orange, then yellow, then green etc etc, so that in the end when I stack the channels, the whole spectrum is represented and all colors we can see can be created by mixing the different levels in each channel.

I would like to do this in Photoshop but any other suggestions are appreciated. Thanks.

[Christoph]

Interesting topic... but where to get an "8 bandpass filter"? Is that hypothetical? Midwest Optical has at maximum Triple Bandpass filters...

[dabateman]

1 hour ago, Christoph said:

Interesting topic... but where to get an "8 bandpass filter"? Is that hypothetical? Midwest Optical has at maximum Triple Bandpass filters...

Not in one filter. Fandyus want 8 separate band pass filters and take 8 images and then merge those 8 images into one image using the rainbow. 

I think you need to color each image then stack or merge them together.  Only problem is that yellow and blue make green. As does green. So the colors might be wonky. Which is why most software limits you to the 3.

InfranViewer I think might let you color the individual images. Then you would need to align and stack them elsewhere. 

[Stefano]

As David said, you can try to do that in IrfanView, converting the images to grayscale, and then adjusting the channels to “color” them red, orange, etc. Then you can stack those images using another software. This is how I make my TriColour images.

[enricosavazzi]

2 hours ago, dabateman said:

Not in one filter. Fandyus want 8 separate band pass filters and take 8 images and then merge those 8 images into one image using the rainbow. 

I think you need to color each image then stack or merge them together.  Only problem is that yellow and blue make green. As does green. So the colors might be wonky. Which is why most software limits you to the 3.

InfranViewer I think might let you color the individual images. Then you would need to align and stack them elsewhere. 

I agree. The number of color channels perceived as truly independent of each other cannot exceed the number of primary colors recorded by the visual system, i.e. three in [the large majority of] humans. A given secondary or higher-order color can result from different mixes of two or three primary colors, and the vision system has no way to distinguish among these mixes. A perceived orange, for example, could be a monochromatic 580 nm emission or a mixture of two 700 and 590 nm emissions (or an infinite number of other possible mixes).

 

There are also other, more complex perception problems, e.g. that human red-sensitive cones are also sensitive to violet (around 400 nm), so the perceived violet could be a 400 nm emission or a mix of red and blue. See the figure captioned "Normalized human photoreceptor absorbances for different wavelengths of light" at https://en.wikipedia.org/wiki/Photoreceptor_cell . This has interesting consequences in Bayer sensors with red filters that do not have transmission characteristics exactly matching human red cones sensitivity. These sensors often render certain flowers perceived by sight to be violet as blue instead.

[Stefano]

I’m sure I have already posted this, but I can’t find it.

 

Older sensors see a 405 nm laser (which looks violet to the human eyes, I saw it in person) as blue. Newer sensors correctly display it as violet/purple.

 

 

[Fandyus]

Interesting what you all have said this far. See, I'm just trying to look or a reasonable way to merge more than three images to get color. What if I overlaid the images on one another with equal opacities instead until I'm left with three, would that work well?

[dabateman]

3 hours ago, Fandyus said:

Interesting what you all have said this far. See, I'm just trying to look or a reasonable way to merge more than three images to get color. What if I overlaid the images on one another with equal opacities instead until I'm left with three, would that work well?

Try it. We don't know until we see it to see if you loose that channel information as we with only 3 cones, just blend in the colors anyway. 

Some people can actually see color with rods though. Also extremely rare, but some women may have 4 cones for better red color sensitive. 

[Andy Perrin]

Fandyus, you can do it-- The basic idea is that you imagine three "pseudofilters" (my word), one each for the R, G, and B channels, and you form these pseudofilters by using weighted averages of images taken with the multiple bandpass filters. 

 

Conceptual example:

You start with bandpass filters for (a) 300-320, (b) 320-340, (c) 340-360, 360-380, (d) 380-400, and (3) 400-420nm. You take pictures of the same scene with each filter. Then you do something like this:

 

Adjust each image so has has the same exposure. Then make new channels as follows, where "(a)" means the image taken with filter (a) above:

 

Red pseudofilter = (a) + 2*(b) + (c)

Green pseudofilter = (b) + 2*(c) + (d)

Blue pseudofilter = (c) + 2*(d) + (e)

 

Notice the overlap between each pseudofilter. You want to be able to have e.g. R and G channels active at the same time so you can get yellow. The point of doubling the middle filter is to make an imaginary "peak" at that band of wavelengths. But the exact weights should probably be based on analyzing the spectra of the Bayer filters if you want to get something similar to visible color. The idea extends if you have more images, but the formulas for the weights will be different, and of course you don't have to use just 3 BP filters for each pseudofilter, that was just to show the concept.

 

Doing something like the above has been on my long long (long long) list of "ideas to try" for a few years now... 

 

Also we had a big discussion on these things back here, so definitely read this thread:

https://www.ultravioletphotography.com/content/index.php?/topic/4650-how-to-make-a-tricolour-image-my-method/

 

[Andrea B.]

I'll note that simple straight stacking in Photoshop layers at the same opacities doesn't work for this. Here's an example.

 

I made a 6-stack of Red (next-to-bottom layer), Green, Blue, Magenta, Yellow, Cyan (top layer) each at Normal with 16% opacity (100/6 = 16.66) over a white background bottom layer.

 

A stack made like this has a final color which is always dependent on the order of the layers. That would ruin any kind of Tri-color or Multi-color attempt. Using other opacity settings doesn't change this dependency.

 

Resultant stack with Cyan as Top layer.

 

 

 

Resultant stack with Red moved to be Top layer.

[Andy Perrin]

Andrea, I wouldn't expect that to work anyway, so I'm not surprised. Photoshop is not exactly (ahhhh...) transparent about how all their tools work, and I've noticed this order-dependence you mentioned.

 

(This was not what I was suggesting above, to be clear. I was talking about imaging through bandpass filters and combining the images mathematically, for example with a programming language. Not sure if you were referring to my comment or not.)