NIR Camera Calibration

Hi babu - can you upload an image of vegetation from your camera? It would be really interesting to see.

Hi Babu,

As I was writing this I see you got some good comments - here are some more.

Lots of good questions. I'm also looking for answers to some of these questions and can fill you in on what I've learned so far.

Ideally we would calculate NDVI from radiance (what the sensor measures) but unless we have a way to calibrate the camera, a fairly complex process, that's not possible. An alternative is to develop methods to standardize or normalize NDVI images acquired using different IR cameras so they can be compared . I think white balance is one step to accomplish that but only if the NDVI images are created using similar bands (wavelength range) for the red and NIR layers.

I haven't used a filter that passes red and NIR light but my guess is that it will be difficult to get a good red and NIR layer from your image for the reason you mention. Have you looked at the three individual image layers to see if they are very different. If you take a photo of green healthy vegetation a decent red band would be quite dark and the NIR band would be relatively bright for vegetated areas of the scene. If you don't have good enough differentiation to create NDVI images there are some other options you can try or you can use your converted camera with a NIR filter over the lens and a second unmodified camera for the red band.

One of the other options is to play around with subtracting different layers (e.g., red-blue, green-red...) to see if the differenced images produce suitable layers for NDVI. It's quite possible that will work. I would also try setting white balance using a gray card and also green vegetation before taking your photos to see if that helps.

I'd be happy to look at one of your images to see if they would work to make NDVI.

I hope this helps,

Ned

Babu,

The new camera is going to be very fun to use. It would be interesting to see a histogram for each channel from an image like that one. We can hypothesize that any data in the green channel is NIR light plus a little red, and any data in the blue channel is mostly NIR. The data in the red channel will include a mix of red and NIR. So:

Channel....... Data recorded
Blue............. NIR
Green........... NIR (plus some red)
Red ............. Red plus NIR

What you don't have is a channel that is mostly red with no NIR. And you need that to compute NDVI. Or you could use a channel that is mostly blue because blue is absorbed by healthy vegetation just as red is, but you don't have that either.

It might seem logical to compute a red channel by subtracting the blue channel from the red channel. So:

(red + NIR) - NIR = red

It makes no sense to me that this would work.

But you can test whether it does or not. Find a simple scene with healthy plants, some other green things that are not plants, and some non-green non-plants. Take a photo with your new super blue camera and also with a similar unmodified camera. The two photos should be precise repeats. Then calculate NDVI in two ways. First do the above math to compute a red channel and compute NDVI using that as red and the blue channel as NIR. Then compute NDVI using the same blue channel as NIR and the red channel from the unmodified camera as red. This requires that you register the two images so the red channel from the unmodified camera matches the channels from the super blue camera. If the two versions of NDVI are similar, you have good reason to believe that subtraction can produce a computed red channel.

Chris

Babu,

You might be right that the blue channel is more sensitive to NIR, based on the fact that the healthy foliage in your super blue photo has a bluish tinge. We know the red channel should be recording some NIR in addition to red, so why doesn’t the foliage appear reddish? The color in the final image is a combination of whatever is in all three channels, so there must be a lot of light getting through the blue Bayer filters. And I guess there must be more light getting through the blue Bayer filters than through the red Bayer filters. There is no blue or green light getting through your super blue filter, so all the light in the red and blue channels is red and NIR. Assuming there is very little red being reflected from healthy foliage, there must be more NIR being recorded by the blue channel than the red channel. But these differences are small (and my logic probably flawed), so it might be that the blue tinge is just an artifact of white balance. You could test this by repeating the photo with different white balance settings.

The camera you use to take a normal version of your super blue photo does not have to be identical to the super blue camera. The most important thing is to be able to register it very well with the super blue photo. That means the focal length and the resolution should be very close. You could resample both images to the same pixel resolution after they are taken. You just need a normal red data channel for each pixel of the super blue image.

The pattern of IR transmission of the super blue filter probably does not matter much. When NDVI is computed from modern satellite data, the NIR band used is very narrow, just a few dozen nm. We cannot replicate that, so we use the very wide NIR band we can get from a camera. The important thing is to make sure there is not much red data in the NIR band you are using. That is your main problem.

I don’t know why the super blue photo would look different at different exposures unless maybe automatic white balance was on.

I don’t know what ASD field spectrometers are.

Chris