Home > Photography > False Colour From Filters (and Simulated Filters)
False Colour From Filters (and Simulated Filters)
This last article on false colour covers a topic that I was initially somewhat scornful of - the use of various filters on a multispectral camera to produce strange and unusual colour combinations. Over time, and thanks to some discussions with Steve Smeed, I realized that the effect is often very visually interesting, and it's probably at least as related to the actual scene as some of the more exotic algorithmic false colour I'd been working with.
The cause of this type of false colour is the result of the behaviour of the filters outside of the human-visible spectrum, combined with the response of the red, green, and blue areas of the camera's sensor (which, in most cases, is due to the internal Bayer filter).
For example, a Hoya R72 filter allows near infrared light to pass, but also allows the long-wavelength end of the red part of the spectrum through. As a result, on virtually any camera that uses a Bayer-type design, the red pixels in the image will capture a mixture of red and near infrared light, whereas the green and blue pixels will capture near infrared only (because the filter material used for them is transparent to near infrared, but the R72 filter blocks green and blue light from reaching them).
Here are some simple examples for comparison. Each of these was shot in a way that allowed in one human-visible band in addition to near infrared light, either by using the R72 filter, or a primary-coloured filter:
More colourful results can be obtained by using secondary-colour filters, or filters with more complex curves. Yellow is - unfortunately - the only commonly-available secondary colour filter, I assume because of its historical use in "black-and-white" photography for reducing haze.
The magenta filter used for these examples was a very old Kodak 34A that I found secondhand. It is still easy to find magenta colour correction filters of various grades, but these do not block 100% of green light. You will know when you are looking through a 100% green-blocking filter, because it is a truly strange experience, and makes it extremely obvious how much of our standard perception of the world is based specifically on green light.
I have been unable to locate a true cyan filter, so the "cyan stack" used for these images was a glass CC110C with a gel CC50C in front of it.
The B+W 403 is used here without an infrared-blocking filter on top of it, so that image is really a mostly-infrared shot, despite the 403 being an "ultraviolet" filter.
Tiffen's "Cool Day For Night" is a specific type of blue filter used in cinematography to approximate the appearance of shooting at night (which is much easier in many cases than actually shooting at night).
When shooting using this method, depending on the lighting conditions and the sensitivity of the camera to near infrared light, it can be helpful to stack on a filter that cuts down on infrared, but doesn't block it entirely. This is where filters which are poor for their intended purpose (in many cases, at least) like Schott KG-series glass can be put to good use.
Ultraviolet-A: A Special Case
As was pointed out to me by Klaus Schmitt, in the specific case of shooting ultraviolet-A images with a modified Nikon D70, there is a kind of false colour that actually does reveal something concrete about the subject of the photo.
Shane Elen of BeyondVisible.com discovered back in 2005 that even though at lower-frequency/longer-wavelength portions of the ultraviolet-A band (e.g. 380nm) both the blue and red photosites of the D70 sensor pick up UV in roughly equal amounts, at higher-frequency/shorter-wavelength portions of the band (e.g. 360nm) the red photosites are much more sensitive (see this forum thread from 2007 for additional information). This aspect of the D70 sensor was later confirmed independently by Dr. Schmitt and some of his associates.
As a result, if the object being photographed reflects different parts of the ultraviolet-A band in different amounts, the image that is taken will have some areas that appear purple, and some that appear red. If the image is tinted only purple, or only red, then either the object only reflects one part of the ultraviolet-A band, or the light source being used for the photo only emits light in one part of that band.
Here are three flower photos that illustrate the mixed-reflectivity case. In addition, I've included two mineral photos that were shot under UV-C lighting. UV-C lights of this type also emit (among other things) 360nm UV-A light, giving those two images a "red kryptonite" appearance because it's mostly just the red parts of the sensor being stimulated.
In all of these cases except the ultraviolet-A example mentioned above, it is possible to simulate the appearance of one of these images using standard near infrared, red, green, and blue exposures. The process is one of combining the information from those exposures in a way that mimics the combination of light filtering caused by the filter(s) used in front of the lens and the sensor response of the camera itself. This is simply a multispectral extension of the "channel mixer" concept used in Photoshop® and the GIMP.
The Mirror's Surface Breaks includes a "filter emulation" processing configuration based on some "best guess" approximations I put together based on sample images. One of my longer-term projects is to take some much more accurate measurements and improve the results further, but even as-is they should be very close to the real thing. This means that it's possible to produce images which appear as though they were shot with cameras and/or filters I didn't even own, let alone have with me, when the original photos were taken (again, just like the non-multispectral equivalent in regular image-processing software). For example, the "R72 Classic" configuration is based on sample images sent to me by Steve Smeed, which were taken with an Olympus C-2020Z.
Here are a few examples using these initial approximations:
This same technique could be used to simulate the specific appearance of - for example - Kodak EIR "colour infrared" film (or Aerochrome, et cetera). EIR used a near infrared/red/green scheme very similar to the one I use for my digital photos, but was subtly different due to the specifics of the film. It has its own extremely-loyal fanbase, and they've spent a lot of time trying to replicate it as closely as possible in digital format.
The EIR folks (in particular, "Infrachrome"/J.W. Wong) have also developed processes to perform the reverse type of transformation (e.g. taking an image that was shot with a multispectral camera that had a yellow 12 filter in front of it, and extracting the near infrared, red, and green bands from the raw image). I've done a little work along those lines, but have not included the functionality in TMSB yet. There is some interesting potential there, and the same model could be used to capture near infrared/red/blue images in one shot using a magenta filter, near infrared/green/blue images using a cyan filter, and so forth.