Digital camera chromatic variations seem to be a fact of life. Some cameras have them worse than others, and the most well-known aberration is purple fringe.
Purple fringe isn't unique to the Powershot cameras in any way -- it's been noted as particularly strong on Olympus and Nikon cameras, and just about every digicam suffers to a greater or lesser degree.
Purple fringe seems to have a variety of causes, and can show up in unexpected places. The image above (with two zoomed-in detail areas on either side) shows the typical, expected case: purple tones around the edges of a highlight or a lightsource, bleeding into the neighboring pixels. Each of the detail images are taken from opposite sides of the image, and you can see that the fringing on the Powershot tends to be on the side of a highlight that faces away from the lens center.
Such chromatic aberrations ("CA") need not be restricted only to highlights. The image below shows a more problematic circumstance: purple highlighting on a face. Here is a small fragment of a larger image, in full color and then with each of the red, green, and blue channels displayed as grayscales.
The image was recorded in RAW mode -- the camera meter indicated an exposure that almost exactly matched the incident reading from my Sekonic meter (the lightsource was a single 600W tungsten lamp to the right of the camera, slightly reflected by a hanging metalized auto sunshade just out of view to camera left). The ultimate target was to create the B&W image seen in the sidebar at the bottom of this page, but I was certainly struck by the amount of purple in the shot -- at first I feared it was a hopeless case.
Fortunately, the green channel carried just enough detail to save the day when converting to the final B&W version -- if my target had been color, the image would have indeed been lost.
What is causing these problems? One issue appears to be the light source itself. Compare the aberrant picture above with the image below, shot with the same girl in the same place but differing in one key way.
In the less-aberrant image, a different lightsource was used. This time the shot was lit by an umbrella'ed Sunpak strobe, placed very similarly to where the tungsten lamp had been. As you can see, purple fringing is hard to find in the strobe-lit image.
Even when we deliberately increase the strobe exposure in an attempt to cause aberration, the results are minimal -- the purple fringe is almost entirely absent. It's there, all right -- but just barely.
(An even more extreme example can be found at the top of this page -- the strobe is off to the side and aimed directly into the lens, but... no CA!)
Overexposed by strobe but
without strong fringing
Strobe illumination is very different from the illumination that we get from a source like a tungsten light bulb or even the sun. Those "hot" light sources are black body sources -- they glow because something has heated them up enough to emit heat energy as photons. They tend to radiate quite a bit of light at all wavelengths -- the average color is specified as the color temperature, but it's only the dominant color -- light is emitted across the whole spectrum.
Strobes, on the other hand, operate through electrical excitation of a gas, and only emit photons in a few very narrow bands. The average color of these bands is called the color tempterature of the strobe, but the actual light coming from a "daylight-balance" strobe is quite different from real daylight, particularly in the colors on the outlying parts of the rainbow: infrared (IR) and ultraviolet (UV).
Strobes don't emit a lot of IR or UV. For the Powershot's CCD, this seems to be a good thing. The CCD is sensitive to these "invisible" colors -- to prove it you yourself, point the IR remote at the camera and press a few buttons while watching the monitor. To your eye, the remote appears black -- to the Powershot, it looks like a little flashlight (or see here for some nice examples of Powershot shooting using only IR light, passed by an IR filter). Similarly, you can see quite well with the Powershot under a UV "blacklight."
While optical lenses perform well in the visible spectrum, some distortions can appear outside that color range. The UV/IR light can focus slightly off-center from the place where the visible light focuses -- when this happens, the distorted IR caused a "lateral aberration" just past the highlight.
Lateral aberration is related to the wavelength of the light, and so when aberration occurs, longer wavelengths (reds) will tend to be away from the center of the optical axis, while shorter wavelengths (blues) will tend to be toward the center. This points even more convincingly to IR as the hidden source of purple fringe, because it appears only of the "outside" of highlights.
If the aberration is infrared, why magenta? How does the blue get in there, rather than the aberration being strictly red?
The IR can show up in both red & blue channels, because those colors are interpolated across pixels -- different pixels on the CCD are masked with different colors, and those masks are apparently transparent to IR. DPreview has published a small guide to CCD color coding here. Canon uses the CYGM system.
The dyes used for the color mask are tuned to visible light, but not (as far as I can determine) to infrared. The IR passes through not only the red-sensitive portions of the mask, but also leaks through the other colors as well -- least of all for the green channel. What's white less a little bit of green? Magenta.
Finally, the camera appears to combine the various channels differently at different exposures. The characteristic curve described in the neighboring page on exposure shows a little of that varying relationship. This is to be expected -- the Canon engineers have attempted to compensate for non-linearities in the sensor by using different compensation formulae at differing intensities, which is a good choice for most images.
The purple fringing appears to be the collision of three problems -- inconsistent lens performance outside the visible spectrum, combined with the excess sensitivity of the Powershot's CCD to IR light, combined with the camera over-compensating the red and the blue channels on some highlights. These aren't the sorts of problems that can be fixed by a firmware revision -- they are the natural result of optical design, dye manufacture, and CCD physics. Currently, the only ways I know to defeat purple fringing are:
The color rendition of the B&W image at right was created using -- you guessed it -- more tricks. The most important trick was the addition of Photoshop "film grain" -- by inverting the image before adding film grain, the "highlight size" creates B&W negative-like clumping. Once re-inverted, the resultant appearance is more like regular 35mm B&W printing (or based on the aspect ratio, 6x4.5 -- though with rather less detail). A sepia-based duotone and some painted-in vignetting complete the image.
©2001 Kevin Bjorke