I now have three infrared filters and one ultraviolet-pass filter of varying specifications, for different purposes. Two of the IR filters are cheap, one of them was expensive and the UV-pass filter was found in a charity shop and I have no information about it. They all behave differently – so I wanted to test them to find the truth about which wavelengths they transmit and block. Here’s our line-up:
- Generic 720nm filter in 58mm screw fitting (for 35mm cameras)
- Generic 760nm filter in 77mm screw fitting (for medium format e.g. Mamiya RB67)
- Astronomik ProPlanet 807nm in EOS Clip fitting (for Canon DSLRs)
- Ilford UV filter in 2″ square size (for “instruments”, apparently)
Note for photographers: the UV filters you are probably familiar with blocks UV light from reaching the camera, and allows visible light through. This UV-pass allows UV light and blocks visible light.
The human eye can see up to about 700nm so it isn’t possible to see anything through these filters. To my infrared-sensitive DSLR camera (up to roughly 1100nm) which still has its Bayer filter, all three of these filters should produce an effectively monochrome image with a strong red cast. When the white balance is corrected, these images will appear plain black & white. Black & white images are exactly what I observe when taking landscape photographs with either the 720nm or 760nm filters. However, the 807nm filter which should allow even less visible light actually produces traces of blue and green in its images (see below for an example) and on a sunny day I can actually see a very faint image through it with the naked eye.
Clearly this is not expected behaviour. Whichever way you look at it, the 807nm filter should be less permissive than the 720nm and 760nm filters and this doesn’t appear to be the case. Fortunately, one of my friends is a research scientist* with access to a spectrometer that can accurately measure exactly which wavelengths are permitted through each filter. When I went to visit him, the only spectrometer available was only able to measure wavelengths between 300nm and 900nm, rather than all the way out to 1100nm as I would have preferred. Still, it was able to measure the critical region between 700-800mm where the band cut-off is supposed to occur.
* I prefer to call him a Professional Scienceman, pronounced similarly to “policeman” or “fireman” with the deadened “a” sound in “-man”
The spectrometer, like all good scientific instruments, is attached to a Windows 98 computer which has no network connection, no USB mass storage support and only a floppy drive for communications with the outside world. It’s 2013 and I haven’t carried a floppy disk on my person for at least a decade so I had to rummage around at work until I found a couple of disks.
These blue graphs show the percentage transmittance of each filter, plotted on a linear scale against the wavelength.
Lessons learned from these graphs:
- The 720nm and 760nm filters appear to be pretty much identical – which isn’t terribly surprising given that they are no-name filters that were not many pounds each. Clearly the 720nm filter was mis-labelled so I might as well treat them both as 760nm filters.
- The more expensive 807nm has a sharper cut-off (steeper slope) at its threshold, although it also permits some UV through
- The 807nm filter does indeed have a longer long-pass wavelength than the 720nm and 760nm filters, as its name would suggest.
- The Ilford UV-pass filter also permits a small amount of IR through
So why does the 807nm filter still permit some visible light through when the 720nm and 760nm filters do not? It ought to permit fewer wavelengths through. The answer is because when a filter blocks certain wavelengths, it doesn’t block 100% of that light. Cheap filters might block 99%, better ones might manage 99.9% or 99.99%. The linear scale of the transmittance graphs we just looked at doesn’t show this well. However, in the scientific field of optics I am told it is more common to plot the absorbance (also known as optical density) rather than transmittance of a filter, and to plot this using logarithmic units. We measured the filters in the spectrometer again, and this time recorded their absorbance. Here are the graphs.
Suddenly, plotting on a log scale, we see that the 807nm filter only has an optical density of less than 3.0 for much of the visible spectrum. Compare this to the other filters, which mostly have an optical density of at least 5.0 for most of the blocked wavelengths. It looks like the optical density goes off the scale at 9.999 on this particular spectrometer but I think anything above OD 5.0-6.0 probably can’t be trusted.
For photographers, OD 2.0 and 3.0 are reductions of about 7 and 10 stops respectively. It does sound like a lot, but landscape photographers quite commonly use a 10-stop ND filter. A camera can easily detect light through a filter of this density. So the green and blue light visible in my test photo isn’t being totally blocked, just reduced by about 9 stops.
If I wanted to reduce the blue and green light to an undetectable (by DSLRs) quantity I could use both the 807nm filter for its sharp cutoff and one of the other filters which has a greater optical density to blot out the shorter wavelengths.
Alternatively I could simply take the red channel from a colour picture when using the 807nm filter. All the green and blue colour information would be in separate channel anyway, thanks to the camera’s built-in Bayer filter. Ditto for the Ilford UV filter. If I wanted to take UV-only pictures, that filter permits some IR through too. But the UV information would be in the blue channel and the IR information would be in the red channel. Easy!
I don’t really have a summary for this article. Mostly that I now know exactly what each filter does, and that the inexpensive no-brand infrared filters block unwanted wavelengths well, and the expensive one blocks them less well. It seems that’s because it’s an interference filter (reflective) rather than an absorbtive filter (opaque) like the cheap ones. Shocker!