[Jerry Kindall @ Sep 26th 2006 1:46PM]

The Foveon sensor has its drawbacks. However, in terms of image quality, it is roughly equivalent to a Bayer sensor of the stated number of pixels (14 in this case). You have to remember that 2/3rds of the information in a Bayer-sensor camera is simply made up. Furthermore, because the color pixels of a Bayer sensor are separated in space, the sensor requires an anti-aliasing filter in front of it -- anti-aliasing being, in this case, another way of saying "blurring," i.e., removal of detail. Foveon sensors don't require this. (Well, technically they do, but the Foveon artifacts are much less annoying. With Bayer sensors you get unacceptable color moire if you don't have an antialiasing filter; with Foveon sensors you get a bit of spatial aliasing of fine detail.)

You could in theory take a Foveon image and blow it up to 14 true megapixels using something like PhotoZoom Pro and have an image that matches a 14 megapixel Bayer image for detail.

Foveon's Achilles heel actually has been color, not resolution. The sensor has historically suffered from metamerism, i.e., colors that are distinct to the human eye become a single color in a Foveon image. The sensor has been particularly weak with the cooler end of the spectrum, if I remember correctly. It'll be interesting to see if they've addressed this weakness in the new sensor. If kdo is correct, and they're now using 4 layers rather than 3, they may have done something about the color response.

The ideal digital camera would have two aligned sensors: a monochrome one for detail (luminance) and a Bayer mosaic one from which color (chroma) is derived. I'm surprised no one has done this, frankly, it's such an obvious idea. Alignment between the sensors could be done largely in software.

[Foof @ Sep 26th 2006 2:22PM]

2/3 of the data isn't simply "made up" in most cases (only if the image data of adjoining pixels consist entirely of a primary color R, G, B. For any other situation (99.9% of the time) bayer processing uses and interpolating data from every photosite. In theory, Foveon should have no real advantage over Bayer if a scene consists mainly of grayscale, since every bayer photosite is doing "equal" work.

Having said that, I've seen quite a bit of Foveon output over the years, and the detail it resolves is incredibly impressive, especially at full-size. A 4.7MP Foveon should be equivalent to a 10-12MP bayer, just as the 3MP Foveon in the SD9/10 generally outresolves Canon's best 6MP bayer.

The 0.98x frame coverage is great too, though I wonder how large a 0.9x maginfication will look on a relatively small 1.7x crop sensor...