Public Lab Research note


Superblue

by cfastie | April 20, 2013 02:49 20 Apr 02:49 | #6923 | #6923

cfastie was awarded the Basic Barnstar by warren for their work in this research note.


The new infrared camera under development at Public Lab will probably make two kinds of images that reveal how much plant growth is happening in the scene. By combining information about both the infrared and normal visible light reflected from plants, the camera can produce false color infrared images like the familiar pinkish satellite images. These are called NRG because instead of the image being composed of Red, Green, and Blue (RGB) channels, those channels display Near-infrared, Red, and Green (blue is not used). An infrared channel can also be combined with a visible channel to compute the Normalized Difference Vegetation Index (NDVI). This quantifies the difference between light that plants use for photosynthesis (red or blue) and light they don't use (infrared). The bigger this difference, the greater proportion of the light hitting the leaves is being absorbed to power photosynthesis.

BG3curve.JPG MaxMaxElph.JPG

The Schott BG3 filter passes lots of blue light (400-500 nm) and lots of near infrared light (>700 nm), but passes very little red light (600-700 nm). Canon Powershots with this filter instead of the standard IR block filter record little red light in the red channel, which instead records near infrared light.

Although the current Public Lab infrared tool includes two cameras -- one for visible light and one for infrared -- it is possible to make NDVI images with only one camera. A filter can be used to block all red light, so that the only light captured in the camera's red channel is infrared. The blue channel captures normally, so NDVI is computed using the difference between the blue and red channels. NDVI computed from satellite or aircraft data uses red light because blue light is scattered by the atmosphere and introduces unwanted variation. At close range, this scattering is not a problem so blue light can be used to make meaningful NDVI images.

Rosco2007.JPG Rosco2007Ebert.JPG

The Rosco #2007 VS Blue filter has a spectral transmission similar to the Schott BG3 filter. Above is their published spectral curve and Ebert's curve.

Glass filters are available that block all red light but pass the blue end of the spectrum and also infrared light. These are expensive, so I have been experimenting with inexpensive polyester filters made for theatrical production. One company makes filters in many dozens of different colors and publishes a spectral graph of each filter. I bought a Rosco #2007 VS Blue filter to test, which only cost about US$7.00 for a 20 x 24 inch sheet. My cousin gave me a new Canon Powershot A810 ($70-$100 on eBay) that she had picked up as swag at a conference, so I removed the IR block filter from it to make it sensitive to infrared light. I attached it to a Public Lab spectrometer (Ebert) and photographed spectra of halogen light passing through a piece of the Rosco filter. There is some similarity between this spectrum and the one Rosco publishes for this filter. In Ebert's curve it appears that less of the far blue light and infrared light past 750 nm is being transmitted, but the sensitivity of my A810 falls off at both ends of the spectrum, so those parts can't be compared with the professional spectral graphs. The important characteristic evident in all the spectral graphs is that very little red light is transmitted, but blue and infrared are. The question is whether the cheap Rosco filter will allow the A810 to capture photos with mostly blue light in the blue channel, and mostly infrared light in the red channel.

Rosco_G11.jpg

Ned's G11 has a LifePixel Superblue filter (probably a Schott BG3) in front of the sensor instead of the standard IR block filter. My A810 has some Rosco blue filter taped in front of the lens.

Ned Horning has a Canon Powershot G11 with a Schott BG3 glass filter which has been professionally installed in it where the original IR block filter used to be. The US$250 price tag for this Superblue conversion does not include the camera, which is $250-$350 on eBay. It captures photos which can be converted into very good NDVI images. He lent me this camera this week and I have been using it every day to shoot whatever green plants I can find outside (it's still a little brown around here). I also taped a piece of the Rosco blue filter in front of the lens of the A810 and have been taking the same shots with it.

The strong blue color that reaches the camera's sensor overpowers the standard white balance settings. I tried each white balance preset in the A810 and the photos were all purple. Then I used the A810's custom white balance feature while filling the frame with light bouncing off a gray stone. This makes the images from the modified A810 resemble the photos from the professional Superblue G11. So just about any camera can be converted to a Superblue camera, but the custom white balance feature might be a requirement.

WBalanceRosco.jpg

Custom white balance is probably a requirement of cameras to be modified into NDVI cameras. Pointing the camera at anything while calibrating the white balance works. More experimentation is needed to learn how white balance affects the final NDVI image.

Ned has a special version of his ImageJ plugin which uses the red and blue channels of a single image to compute NDVI. This makes it really easy to make NDVI images from single Superblue photos or from entire directories of them. Some early results are below. The Rosco filter on the A810 did not produce the nice discrimination of green plants that the professional Superblue camera did. I took more photos today with different white balance which might improve the usefulness of the RoscoCam. I will put those in another note soon (here it is).

Filterdaylilly.jpg NDVIdaylilly.jpg


31 Comments

It'd be neat to test the roscoe with our spectrometer against their spectrum graph. I was excited to see Jeff Hecht working on testing some filters on SpectralWorkbench: https://spectralworkbench.org/sets/show/336

Reply to this comment...


I've been messing with different NDVI gradients, and have been trying to get one which shows "cool" colors below zero, and doesn't circle back around to violet (which i find disorienting for the "heat" metaphor). What do you think of this one?

ndvi-flip.jpg

It replaces violet at the "cool" end with black, and seems to read pretty well.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


That's an excellent color table for the heat metaphor. This topic of how to display NDVI deserves much more attention.

"It'd be neat to test the roscoe with our spectrometer against their spectrum graph."

Did you actually READ this note????;)

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Apparently, from my phone, not nearly well enough. There are disadvantages to a mobile friendly website...

Using baseline subtraction in Spectral Workbench you should be able to produce a graph corrected for input light source and sensor sensitivity, although the resolution of that graph wouldn't be consistent. Have you attempted that before? Maybe I missed that on the researh note too

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I've been modifying lots of cameras with the new superblue filter (using the Rosco filter you sent). For those (like webcams) which don't have custom white balance, do you feel that we can do white balance in post processing if we have a reference white balance card in the frame?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Jeff, Your first question: I subtracted the Rosco spectrum from the halogen light spectrum, but the result made no sense at all. So I just used the Rosco spectrum which at least makes sense. I'm not sure why that is. White balance question: Yes. More discussion on the infrared forum.

Reply to this comment...


Can you link to the spectra so I can give it a try? Thanks.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


This set has the Rosco and the halogen. https://spectralworkbench.org/sets/show/366

Reply to this comment...


@Jeff i really like this NDVI color gradient that replaces violet with black.

Reply to this comment...


Hello cfastie!

I follow your job with interest, but there are a thing that I not understand well. You say "A filter can be used to block all red light, so that the only light captured in the camera's red channel is infrared. The blue channel captures normally, so NDVI is computed using the difference between the blue and red channels." But the NDVI is not defined as the difference between Red and NIR channels?

Thanks in advance for any clarification.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Hi jaluna, NDVI tells us something about plant health because it quantifies the difference between light that is absorbed by plant pigments and used for photosynthesis and light that is not used and is reflected from the leaves. Both blue and red light are used by plants for photosynthesis, and NIR light is not. Satellite and high altitude sensors always use red light because Rayleigh scattering by the atmosphere degrades the signal in the blue part of the spectrum. For very low elevation aerial photography (balloons, kites, or UAVs) and for normal ground based photography, the blue wavelengths provide an acceptable substitute for red. However, on hazy days, infrablue photos of landscapes can show a gradient in NDVI with distance because Rayleigh scattering effects blue light more than NIR.

Reply to this comment...


Hi, so if I am understanding this correctly I only need to remove the IR filter from my camera and replace it with Rosco #2007 VS Blue filter. At that point I can upload the image onto my comp to compute NDVI. That means only one camera to do all of this and not two. Did I understand the article correctly?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


That's right, jalarid!

Reply to this comment...


Hi jalarid, Yes that is correct. There is lots of active development about which cameras and filters work best, and how to get the most meaningful NDVI from the infrablue photos. Check out related research notes (http://publiclab.org/tag/infragram) and forum discussions (https://groups.google.com/forum/?fromgroups=#!forum/plots-infrared).

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


BTW, this post is now the 7th most popular research note (by view count) on the site!

Reply to this comment...


Hi, I don't understand how can be blue color the Rosco #2007 VS Blue filter when it allows blue light to pass. We see the color of the objects because they reflect it, don't we? Thanks!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


kykytos, This is a really good question. Lots of blue light can pass through a #2007 filter, so a white light viewed through the filter will appear blue. Red light is absorbed by the filter material, and that light turns to heat. The filter has a shiny surface, so all colors are reflected off of it. If you hold a piece of the blue filter under a light, but with a very dark area behind the filter, the filter does not look very blue. Even red is reflected off the surface. Some of the blue light that enters the filter material is reflected and scattered on the inside of the material, and some of that can come back to you when you look at the filter. Red light that enters the filter material does not come back to you because the filter material absorbs it. So the filter can look slightly blue even when no light is being transmitted to you from the other side.

Reply to this comment...


Hi cfastie! I really think that your work is pretty awesome and for that im trying to do some experiments with the NoIR camara and your "Rosco filter". As I understand you have 3 band in a RGB photo, Red, green and blue. when you are taking off the IR filter what you are doing is increasing the amount of energy who is been register for your Red band, and what you do with your filter is leasing just the IR in the RED band, so you will have a IR-G-B photo... I´m right? BUT! in http://publiclab.org/wiki/ndvi, liz say that what will have in the differents channels is NOT the IR-G-B, and will be IR-R-G .... can you please see that? I really think that is wrong because if I´m right any sensor in the camara can read some wavelenght of light and if you take off the IR filter you are not changing the other channels! Thanks for your answer!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


@camilochiang, Yes, you are correct that with a blue Rosco filter, the photo captured by the converted camera is (more or less) NGB. I think the diagram you are referring to is this one: .
5396083368_8d8e441daa.jpg
.
That diagram attempts to show how false color IR images are made from two camera IR systems (with both IR converted and unmodified cameras). The result is an image referred to as NRG because channels have been swapped so N, R, and G data are displayed with the colors R, G, and B.

It is not possible to make an NRG false color IR image from the photo from a single camera (Infragram) system with a blue filter because the red channel is used to capture NIR so no visible red data is captured. You can make a similar false color IR image with a blue filtered infragram by using blue data instead of red. These look generally similar because NIR data is displayed as red, and the G and B data are displayed as colors other than themselves. These images could be referred to as NBG. Here is an example from this note: http://publiclab.org/notes/cfastie/12-08-2013/root-crops .
Pots6pan27k.jpg
.
The abbreviations I use for the false color IR images are different because I was trying to indicate that some of the visible channels are so contaminated with NIR light that we should acknowledge that. But that just makes it so confusing that little useful information is transferred to most people.

Chris

Reply to this comment...


Hello! Is there any "easy to find" material that I could use to take some pictures and calculate the NDVI? I mean, before buying expensive filters I would like to test some materials and cameras. Thanks!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Theatrical gels are a good substitute for more precise filters. Rosco #19 "Fire" behaves much like Wratten 25A and costs $7.00 for a huge (20x24") sheet at Adorama or ebay. A small (2x2cm) piece of Wratten 25A gelatin filter costs $5 (+$2 shipping) at the KAPtery.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Thanks cfastie! I am from Argentina, so I will try to contact providers that could send the filters to my country.

Reply to this comment...


Hi, aguchosolano - the Public Lab store ships internationally and will send Rosco #74 or #2007. I believe they're going to start shipping #19 as well soon -- @natalie, is that right?

http://store.publiclab.org/collections/diy-infrared-photography

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Yes that's correct. We have about 100 of the Infrablu filters left. After those are gone, all the new Infragrams have one each of #74, and #19.

Reply to this comment...


Thanks!

Reply to this comment...


anyone tried the Rosco sample book... only $3.95 from adorama. The swatches will not be big enough for front lens filters on SLRs but easily big enough for compacts or behind the lens. http://www.adorama.com/ROSB.html

Reply to this comment...


Bob, this is a great idea. A secret admirer sent me one of these and I was very excited. I didn't know that any mortal could order them for only $3.95 (+$5.95 shipping). The pieces of filter are 3 x 1.25 inches, plenty big to convert many PowerShots, and apparently you get almost all of the Rosco colors. Each one has a little sheet of paper with the name and spectral curve on it. The samples are cut from their full sized filter sheets which have the filter name printed on one edge. So there is a small probability that each sample will have part of a printed name on it, making it less suitable for our purposes. This adds to the excitement when you receive your swatch book and flip through it looking for that one color you really want to use.

There is similar gambling involved in ordering antique Wratten filters on ebay. I could not resist buying a lot of five filters for $10 (incl shipping) which included #15 and #25A. But both filters arrived well used and scratched just enough to make them unsuitable for internal camera installation. They would be fine for taping in front of a lens, because then the filter ends up getting scratched anyway. I ordered another Wratten #15 that was supposed to be unopened (and cost almost 4X more than the whole lot of used ones). The seller then noticed that the seal had been broken, so he refunded the shipping cost. Still waiting to see how "new" it is. Chris

Reply to this comment...


Is it important the station in which the photo is taken? I mean, how do you differ infrared reflected radiation from infrared emitted radiation (thermal radiation)? Because I suppose it will be very different from winter to summer and it will affect to NDVI computation.

Another doubt: What's the difference between a spectral curve and an Ebert's curve? I don't find it.

Thanks!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


The angle from which the photo is taken and the angle of the sun do influence the NDVI result. NDVI values are a function of how much red and NIR light are absorbed by leaves. At low angles, much light just reflects off the surface and chlorophyll does not have a chance to differentiate between red and NIR photons. Also, deep shadows are illuminated by light with a different color balance, so the proportion or red to NIR will differ from sunlit surfaces. As the day or the season progresses, the proportion and total amount of red and NIR entering the leaf changes, and the proportions reflected to a stationary camera will change fro multiple reasons.

Consumer cameras are not sensitive to thermal infrared, so that is not a problem. Some NIR is emitted from healthy plants by chlorophyll fluorescence, but it is trivial compared to the NIR reflected or scattered by leaves in sunlight.

Ebert is the name of a particular version of the Public Lab spectrometer. Above, a spectral transmission curve for the Rosco filter made by Ebert is compared to the transmission curve supplied by Rosco.

Reply to this comment...


Hello, I wanted to ask if anyone has an idea on how to order the Rosco filter to Europe / Germany? Many Thanks!

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


I think the Public Lab Store will ship filters to Europe: https://store.publiclab.org/collections/diy-infrared-photography/products/infragram-diy-filter-pack

You can contact the KAPtery about special shipping arrangements for filter material: http://kaptery.com/product/wratten-25a-filter

Reply to this comment...


Login to comment.