Posted on Thursday 23 January 2003 to Visions and Illusions
Digital photography that's better than film
As you can see the sensor's ability to capture detailed resolution, both in the grayscale and colour resolution goes way beyond what we could expect from even the best Bayer pattern sensors. Images from the new Foveon X3 sensor are more reminicent of super high quality slide scans, but go even beyond that with no trace of grain. Very impressive. This could be the first sensor to truly surpass film.
Foveon X3 exclusive preview
The way digital cameras work at the moment is pretty brain-dead. Conventional capture technology works by using a single layer of photo detectors arranged in a mosaic pattern to capture the red, green, and blue components of light.
Each photo detector filters for a single colour and it is only though the combination of signals from adjacent detectors that it is possible to calculate the correct colour of each pixel.
One downside of this scheme is that colour filters are very inefficient. A blue filter throws
away all red and green information, a green filter throws away all red
and blue information and a red filter throws away all green and blue
information. In effect, an arrangement like this only uses about one third of the
available light.
Also, in order to reconstruct the correct colour of a pixel, the camera needs to perform a large series of interpolations -
about 100 calculations per pixel. If you double the resolution of the camera, you need to quadruple number of interpolations.
This places a rather large computational burden on the camera which translates into increased battery drain.
Another problem with interpolation is that, at best, it is only an approximation.
At worst, it's a major source of artifacts like colour jaggies, rainbows and moiré
patterns. This is especially noticeable with fine textures and text.
X3 does far better than this approach, capturing all three colours per pixel
by layering the three colour detectors on top of each other in a manner similar
to conventional 35mm colour film.
By doing this it becomes possible to take advantage of the natural colour-separating properties of the silicon itself
i.e. that light of different wavelengths is absorbed at different depths in the silicon: blue
light near the surface (0.2-0.5 microns), green light farther down
(0.5-1.5 microns) and red light even deeper (1.5-3.0 microns).
X3 arranges photo detectors at each depth to read the colour components
for each pixel directly and without any need for colour interpolation.
A Side-by-Side ComparisonThe Sigma SD 9 is currently the only camera sporting the X3 sensor. At around USD $2000 it's out of the range of most us happy snappers although Foveon claims that cheaper cameras around the USD $400 mark will be on the market fairly soon.
Foveon X3 technology visibly improves image quality, as any comparison of equivalent pixel counts will demonstrate. In this case, a 2-megapixel file taken with a mosaic sensor is compared to a 2-megapixel file taken with Foveon X3 technology (more fully described as "2 MP x 3 layers" to reflect that each pixel measures three colors instead of one).
Mosaic Capture
Foveon X3 Capture ![]()
![]()
Sharpness
Mosaic
Foveon X3
![]()
As you can see, the camera equipped with Foveon X3 technology takes sharper pictures. That's because it captures twice as much green as mosaic image sensors, and the green wavelengths of light are critical in defining image detail.
Color Detail
Mosaic
Foveon X3
![]()
These pictures demonstrate how Foveon X3 technology improves color detail. The difference is that Foveon X3 image sensors measure full color at each and every pixel location, while mosaic sensors capture 50% of the green and just 25% of the red and blue.
Artifacts
Mosaic
Foveon X3
![]()
As shown here, Foveon X3 technology offers resistance to unpredictable artifacts. A mosaic image sensor is more vulnerable to artifacts, largely because it must rely on complex processing to interpolate the colors it missed. No amount of processing power can completely take the guesswork out of color interpolation.