Where do the false colors in digital infrared come from? And other questions.

I’ve been wondering why digital infrared images have the colors that they do. That’s a very broad thing to wonder about. So, here are four specific observations that have been on my mind:

  1. Digital infrared images (at 720nm cutoff) are red straight out of the camera
  2. Digital infrared images (at 720nm cutoff) take on interesting differential hues once color-balanced
  3. Digital infrared images (with Wratten 87c / 093) have less R, more B, and less G compared to images with a 720nm cutoff filter
  4. Digital infrared images (with Wratten 87c / 093) are effectively monochromatic

Why? Why? Why? And, why?

I’ve read a lot of discussions about infrared photography and have never come across a thorough answer to all of these questions. But, I stumbled upon an interesting piece of information which may explain three of these four observations. I’m sharing it in case anyone else is also puzzling it over.

Take a look at Figure 2(c) in this paper. It has the responsiveness of one digital camera to different wavelengths of light. (I’ll assume that all cameras are pretty much similar.)

For observation 1. The thing to notice is the responsiveness of R to infrared vs. visible red light near either side of the boundary between infrared and visible light–effectively the same. And the responsiveness of R there far surpasses G and B anywhere in the infrared spectrum, allowing R to dominate the image’s characteristics. This means that

  • if your filter cuts off at 720nm, the whole image without adjustments will be dominated by red
  • if your filter leaks a little visible red light, the whole image without adjustments will also be dominated by red

Whether or not your filter leaks visible light, the result is approximately the same (except when something has a different response to red and infrared).

For observation 2. Notice that R’s responsiveness is decreasing with increases in the wavelength, whereas G and B climb for awhile, and decrease only later. This means approximately that

  • if you color balance so that some particular object is neutral, anything that reflects relatively more low-frequency infrared than that object will have a reddish hue, and anything that reflects relatively more high-frequency infrared than that object will have a non-reddish hue

A filter doesn’t need to leak visible light to produce different hues; a relative preponderance of low-frequency or high-frequency infrared, with a camera with the sensitivities in Figure 2(c), will do it. It only requires R’s sensitivity to move in one direction while G and B’s move differently.

For observation 4. Once you get out to about 850nm, the responsiveness is the same for R, G, and B. So no matter how you color balance with a filter around this cutoff, you change only the overall hue of the image, without differentiation.

Now, I’m left with observation 3. Unfortunately I still can’t explain it! Figure 2(c) seems to predict that, with a deep cutoff filter, which the Wratten 87c / 093 should approximate, the image should be neutral gray. Instead it’s dominated by R, but not as thoroughly as with lesser filters (maybe some light passing below 850nm), with more B than G. Why? Maybe the diminished G is just due to the way that cameras deal with the extra G pixels in the Bayer array. But that’s just a wild, speculative guess.



Leave a Reply

Your email address will not be published. Required fields are marked *