Composite UV imaging using multiple filters

[bvf]

One of the question marks raised over the method I have been using is that there is not much overlap between my three filters and so I'm losing nuances of colours, because items in the subject with a narrow spectral spread would have to be either red or green (say) but could not be yellow. Your method should be able to address this - but probably not with the 3-colour approach I used on your three images,

 

So something occurred to me after I did the quick processing of your shots. You have 7 images. It might be worth including some of them in two of the colour bands, e.g. Blue = 310, 320, 330; Green = 330, 340, 350; Red = 350, 360, 370.

 

This means that the 330 flter would give you a cyan component and 350 would give you a yellow component. I just tried that approach using your images but couldn't see any difference in the final result - don't know if that's because the approach or its execution is flawed or because there are no additional colour nuances in this particular subject.

 

Colouring each of the 7 images is going to be quite a lot of manual effort, but perhaps you have the ability to set up a workflow to do this. My suggestion is a lot easier because it uses the standard GIMP facility (presumably also in Photoshop et al.) to create a colour image from 3 greyscale separation images - but it may not give the nuances of colour that a 7-colour approach would.

 

I think you are right about the tripod movement. Some of the post-Hugin images have a transparent wedge at bottom right, probably because Hugin had to rotate the images slightly and hasn't cropped the images as a person might have done.

[Andy Perrin]

Bernard, I’m way ahead of you. I spoke to Jonathan about doing this a few days ago, although I haven’t started working on it yet. The right way to do this is to use some linear algebra and use the theory of projection onto a vector space to find a weighted average of the different images (very much like you were saying!). It can be automated in MATLAB.

[JMC]

Hi everyone,

 

Sorry, I've been quiet on this (and other) topics for ages. Snowed under with work at the moment, and haven't had time to do any more experiments on this or any other areas which aren't directly work related.

 

I have been reading all your posts with interest though, whenever I can get a break from work. Good to see what everyone is up to.

 

All the best, Jonathan

[JMC]

Another quick update. Given the issues with movement of the flower, and the long exposures I need with the light source I have I'm going to put this research on hold for the moment. I'd like to go down the route of using flash for this, but to do that I will need to to work out a full set of exposure times for each filter which will take quite a while. I may revisit it later though.

[Andrea B.]

Jonathan, this is an intriguing topic you have started. I hope to see more from you soon about this, but certainly do understand that work gets in the way.

 

I would love to see the underlying raw colors from your set of images.

 

Bernard, I also enjoyed your color stack.

 

I do not know how to channel stack 9 (or your posted 7) images into 3 channels and get nuances of color!!

 

It is possible that looking at the raw colours of the narrowband images would give some guidance as to how best to place them into channels?? I don't know for sure unless I were to play around with them.

 

Misaligned layers can be cleaned up as you go along by using the eraser tool. Ultimate success does depend on having nothing too far misaligned. But edges can be cleaned up a bit by either erasing or cloning. Lots of work, of course.

[Andrea B.]

Part of the problem in trying to get a "full color" UV image from this stack arises

because images which are near in peak wavelength have very little tonal difference.

For example, if you take the 310, 320 and 330 images and diff them in PS, you can see that.

The diff between 310 & 320 is almost all black. Similarly between 320 & 330.

So, if just those 3 images were channel stacked, you get a result with little color.

 

QV.

This was (arbitrarily) 310(red) + 320(green) + 330(blue).

 

Maybe if I try 3 which are far apart? More tonal differences, thus more color.

But still kind of meh.....

310(red) + +340(green) + 370(blue)

 

 

 

I so *desperately* want to stack the raw color images that I might have to go order some narrowband filters. I've wanted them for years now. Maybe I can get a loan somewhere? :lol:

 

The UV interval is only 100 nm wide. Unless a particular subject shows very different absorbing/reflecting characteristics in that interval, I don't think you can get much stacked color under narrowbands.

[dabateman]

Andrea,

From my recent UVC, UVB, UVA series I have images with, 253bp25, 303bp10, 313bp25, 335bp10, 370bp15, 390bp25 and 405bp10. I like to mix 313, with 370 and 390.

To me there seems to be 4 regions:

1. 253 better than and similar to 193bp20.

2. 313bp25, better than but different than 303bp10.

3. 370

4. 390bp25 better than but sometimes different than 405bp10.

So I like a 313, 370 and 390 mix usually.

 

These are taken from the monochrome output of the back of the Sirchie KSS 100 imager using a GM5 camera. I collect 3 images per shot in Raw and jpeg. Let me know if you want the raw files.

David

[bvf]

 

I do not know how to channel stack 9 (or your posted 7) images into 3 channels and get nuances of color!!

 

 

Nor do I, really! What I did was a fudge. I said "I combined 310, 320, for the blue channel; 330, 340, 350 for green; 360, 370 for red." So what I did was initially combine the 310 and 320 mono images to get a single mono image which would become the blue channel, combine the 330, 340, and 350 monos to get a single mono image for the green channel, and combined the 360 and 370 monos to get a single mono for the red channel. So all of the images were making some contribution to the final result.

 

However, this doesn't do what you would normally get with a tri-colour filter set, because the filter transmissions should overlap such that some frequencies trigger two channels to provide intermediate colours. What I suggested elsewhere was including some of the original shots in more than one channel: "e.g. Blue = 310, 320, 330; Green = 330, 340, 350; Red = 350, 360, 370"

 

You also say "I might have to go order some narrowband filters." Actually to do this probably you probably need broader-band filters that overlap - but I haven't been able to find any for UV. (I know you got some for IR, but they were horribly expensive.)

[Andy Perrin]

I have an idea for how to do this and I thought I was going to work with Jonathan on it, but he seems to be punting on it for now. Essentially the idea is to interpolate images made with narrowband filters to make a pseudo-wideband filter. Then the pseudo-wideband filters are overlapped to produce R, G, and B channels, as they would be with a Bayer.

[Andrea B.]

I think I understand that that's the only possible way you could merge 7-9 narrowband photos.

 

I have to think about how the interpolation would work. Again, some pairs of photos above have so few tonal differences that the interpolation where the filters didn't overlap would just be a copy.

(......I'm not at all sure that I correctly phrased that thought.....)

[Andy Perrin]

I'm going to try some computational experiments soon, I think. I will make the photographic setup entirely in the computer (SIMULATED) and see if I can get it working there. Then later we can try it in real life, hopefully with fewer major bugs, after debugging the methodology in the simulator.

[dabateman]

I'm going to try some computational experiments soon, I think. I will make the photographic setup entirely in the computer (SIMULATED) and see if I can get it working there. Then later we can try it in real life, hopefully with fewer major bugs, after debugging the methodology in the simulator.

 

Andy, you can use any of my UVC to UVA range images that I have posted, if it helps.

 

Let me know if you need anything. My attempt with the sunflower series, I liked.