NIR, Rosco fire #19 and my elph 300HS from Christmas

[Alexandre Maltais]

Hello everyone,
I replaced the original filter in my elph 300HS by the Rosco Fire Flash Gel to look smart for my girlfriend!
Great, the opening is easy enough, the old filter has been removed, the new fit perfectly in its place, everything is closed, everything is sealed, CHDK and all...!

Ok test! Well, it is pink-red, only pink-red! Not the splendid blue-orange as we can see on the beautiful photos on infragram website, no, just pink-red!
Here the NIR photo along with the useless NDVI version!

Ok, that was not the real Rosco Infragram from the kit (Roscolux Fire #19), it's only a flash gel, one from the package for the "artistic" picture fun, this one: E-Colour #019: Fire.
But that Is still a Rosco Fire! Is it possible that there would be some small difference between the filters? Does it have to be exactly the one sold in the kit?

On the Infragram.org sandbox app, do I just need to put my jpg or raw image or I need to do something before with any image software?

Thanks a lot
Cheers from Montréal
Alex

[Chris Fastie]

Alex,

The two filters are probably the same. The #19 filter will work best if the you do a custom white balance on the camera before you take photos. Flood the sensor with red light while doing the custom white balance (FuncSet/WhiteBalance/Custom) e.g., by pointing at red paper in direct sunlight. Photos of healthy foliage will then be turquoise, and the NDVI values will be closer to where they should be and have more range. Still, it might not help with your girlfriend.

Chris

[Chris Fastie]

Alex, 

How were you still able to achieve this result without a balanced camera and image?

By stretching the histograms and then adding a fudge factor.

Do I do the custom white balance every time for every scene; or, the opposite, I do it once in the bright sunshine, then keep it that way to make sure my results are consistent?

Yes, a good white balance is probably good for all subsequent photos. 

Also, maybe related: with this kind of camera, is there a way to make my results consistent so I can do comparative analysis of the same area over time? Or is this the privileged fiefdom of high-end multispectral cameras? I’m looking to automate my process; I want to eliminate manipulation as much as possible.

Ned Horning has just made progress automating a calibration process:  http://publiclab.org/tag/ndvi-calibrate

Finally, there are so many reds! Which red is best to do a white balance with this setup? Should I try to find the same one as the filter? Is that the way it works?

The red color is not too critical. The objective is to fool the camera into thinking that it must exaggerate the blue end of the spectrum so the blue channel which is used for NIR  has higher values. This is a kludge. It is very hard to get the values where they need to be to produce authentic NDVI results. To eliminate this subjective process, you need Ned's new calibration procedure.

As you see, I’m at the early stages. I’m sparing you a bunch of other questions. But, in your opinion, what is the best roundup of information (books, websites) covering the basics of how cameras work with channels and spectra? I’m mystified by how the information is stored and retrieved in the image file : discrete channels with specific spectral ranges?

Yes, three channels: R,G, B. Computers know how to display them so it looks like a photo to us. Some basics: http://www.cambridgeincolour.com/tutorials/camera-sensors.htm. With a red filter the blue channel captures mostly NIR. Computers don't know what to do with that until we tell them.

 

Or is the information stored all together and the software is able to separate the information after the shot has been taken? And how are these channels (or bulk information) affected by the subtraction or addition of filters?… In a very basic way, I’m completely baffled by the fact that, after lots of experimentation, publiclabs figured that it is a red filter (fire!), and not a blue one that gets the job done! Intuitively, I would have figured that it would have been a job of juggling with different subtle shades of the same color… no, a totally different color, far down the spectrum!

Making an NDVI image from a single photo from a modified consumer camera is a wonderful hack. Today a lot of research into improving the hack is being driven by the drone boom.  Better filters, better calibration, and better post processing will soon follow.

Chris