Infrared Contamination: Good Color Gone Bad
Related: digital sensor, git, infrared, Leica M, lighting
Digital cameras rely heavily on a strong infrared-blocking filter for accurate color reproduction.
Infrared contamination means that the amount of infrared light “seen” by the sensor is significant (or even large) relative to the amount of visible light. A magenta/purple color shift will occur in such cases.
See Understanding digital cameras and infrared below for details.
Leica M8, Nikon D200, Canon EOS 5D-IR compared
For this comparison, I chose two items with very high infrared reflectivity, though they are both quite dark to the naked eye (visible light). These items are not all that unusual; many fabrics, upholstery, anodized metals, deciduous foliage, etc are highly reflective of infrared. Such subjects are at risk of a color cast, from subtle to extremely strong, depending on the camera and lighting.
The use of a strobe (flash) is problematic because enormous amounts of infrared light are emitted (just place your hand over one, and “pop” it—you’ll feel the heat or indeed burn your skin if the flash is powerful enough). Sunlight is less troublesome, though subtle color shifts are still likely with inadequate filtration. Even fluorescent lighting, with minimal infrared output, can be a problem.
Reduced-size images are shown below. The top row is the as-shot image, and the bottom row is the luminance (brightness) of the image above it. Click on an image to go to a larger version with comments, or go to the Samples with Commentary section. The Nikon D200 image renders accurate color as seen by the eye.
|Leica M8||Nikon D200||Canon EOS 5D-IR|
Digital cameras almost always have a strong infrared-blocking filter installed over the sensor. This filter is typically glass, usually a separate piece (Nikon D200, Canon 5D), whose removal allows conversion of cameras to pure infrared shooting, and sometimes it is permanently bonded to the sensor (eg Nikon D2x), which precludes such conversion.
Failure to attenuate (block) infrared light adequately allows the sensor to “see” visible light and infrared light at the same time. This manifests itself as a color cast, which is typically magenta or purple.
Side effects of inadequate infrared filtration include faces that look sunburned, black fabrics that look purple, etc. The results can be awful—and extremely difficult to correct, since different portions of the subject within the same image might all reflect infrared light quite differently, just as an image can have a bright red or green or blue or yellow item in it. Worse, this difference is invisible to the human eye, so the fatal flaw is not known until the image is examined, though in the worst cases the problem is obvious even on the camera’s LCD. No color profile can fix such problems, in spite of the optimistic claims seen in various web discussion forums.
Because this problem can be so troublesome, whenever I purchase a new digital SLR, I always test for infrared contamination right away. I have previously determined that the Nikon D2x, Nikon D200, Canon EOS 1Ds Mark II and Canon EOS 5D all have very strong infrared blocking sensor glass, and do not suffer from any color shifts (due to infrared at least).
In early November 2006, I retested the Nikon D200 for infrared sensitivity, and determined that it is approximately seventeen (17) stops more sensitive to visible light than to infrared—more than adequate to preclude any color shifts due to infrared contamination. By comparison, the Nikon D2h (which I used for over a year) had a very weak infrared-blocking sensor glass, leading to color-shifted skin tones until I began applying a B+W 489 infrared-blocking filter to the lens (see also Nov 10, 2006 blog entry).
The Leica M8 digital rangefinder hit the shelves in November 2006, and it too uses an infrared-blocking sensor glass, though Leica suggests that the blocking effect is achieved by special coatings on the glass, not so much the glass itself.
Let’s explore how the Leica M8 fares in comparison to a Nikon D200 and a Canon EOS 5D converted for infrared shooting (“5D-IR”).
Ironically, the very problem this article explores was quite troublesome in the clothing of the Leica Germany representative! As the representative stepped in front of the extremely bright projector beam, his vest seemed oddly bright, although it looked quite black under the room lighting. Such projectors emit a huge amount of infrared, and so the peculiar reflectivity of artificial fabric could be seen directly with the unaided eye (though this is not necessarily all infrared). This was observed both by myself and a friend. Incidentally, the human eye can see infrared up to nearly 750 nanometers, albeit weakly.
The flash contributed only partially to the exposure of the image shown below, so the color shift is less pronounced. The vest shown below is black to the eye, not dark purple. Note also the ruddy skin tones, a hallmark of infrared contamination.
Using direct flash, this subject has experienced an extreme shift to purple/magenta. Compare to the Nikon D200 image further below. The small black bag is on top of a dark blue sack on top of a black briefcase. Only the briefcase (plastic) has remained reasonably neutral.
This image was color-balanced in Photoshop (setting the bag to neutral), due to a very strong green cast from the fluorescent lights.
Even though fluorescent lights do not emit much infrared, this subject has experienced a significant color shift in the blue bag. Compare to the Nikon D200 image below and the Leica M8 flash image above.
This image is true to the original and can be used as the reference image. The small black bag is on top of a dark blue sack on top of a bright white background.
Taken with a specially-modified Canon EOS 5D [modified with glass over the sensor which blocks visible light (715 nanometer cutoff)]. The background is bright white paper, which by comparison shows that the two bags are very bright indeed in infrared.
See also November 8, 2006 blog entry.
The Leica M8 suffers from inadequate infrared filtration which can result in extreme color shifts to magenta/purple under certain lighting conditions, particularly with subjects that reflect significant amounts of infrared. Current digital SLRs from Canon and Nikon do not have this issue, using far stronger infrared-blocking filters—about 17 stops of attenuation with the Nikon D200, as determined by the author.
Because the results are unpredictable (to the naked eye), and very difficult to correct, the M8 is simply not ready for prime time (in terms of reliable color under varied lighting conditions). Like all Leica “M” cameras, the M8 is a camera useful for shooting under varied conditions, and in difficult lighting. No prudent professional should use this camera for a “money job” until and unless the variables of lighting and filtration are well understood—which rules out much of the utility of a small and quiet rangefinder in applications such as photojournalism and street photography.
This issue must be resolved by Leica, and indeed as of Nov 11, 2006 Leica has stated they are aware of the issue, promising communication within the “next two weeks”. Leica ought to also update their marketing hyperbole with a cautionary note on this problem.