From a linear raw RGB file, all you need is a colour correction matrix so that's relatively simple for Adobe to support linear DNGs and Tiffs. There's a lot of work needed to get a Foveon raw to linear RGB though - you need special processing to extract RGB from the three layer data (which diagrams depict as RGB, but it's not really) and some strong chroma NR to cope with the noise induced by the above processing. If the Foveon did produce good RGB data as is often assumed, indeed there'd be little work for Adobe or anyone else to support it's raws, but producing an RGB image from a Foveon raw needs more processing than it takes to get RGB from a Bayer mosaic colour filter array camera. It's worth reading the Foveon white papers:
http://www.foveon.com/article.php?a=74 to understand more about their chip.
http://www.foveon.com/files/CIC13_Hubel_Final.pdf is perhaps the most interesting. Gems include "Using a direct image sensor such as Foveon’s X3 technology, reduces the requirement of a blur filter because aliasing occurs in all color channels equally giving no brightly colored aliasing fringing." - reduces, not eliminates, and the obvious conclusion from this is that without an OLPF aliasing will occur on a Foveon. Obviously Sigma omit this aspect of the camera an allow for luma aliasing which can look like a sharper image, but is really just artifacts. Given their removable dust shield IR filter, I'm surprised that now that they have a good native resolution they don't supply an OLPF/IR combo filter for this position to remove aliasing for photographers where such artifacts would be an issue.
On page two of that paper, under "New Image Processing Methods" we see the comparison of a typical image processing pipeline "A" for a Bayer CFA, with "B" and "C" from a Foveon. "the color correction transformation matrix required to convert the native sensor signals to a standard color space is aggressive compared with conventional camera sensors and this amplifies noise." - the diagrams show the "noise suppression" blocks necessary in the Foveon pipeline, and although they show such a block in "A", from experience I know that such a block is not particularly necessary. They also employ decimation tricks to get more performance out of the slow NR blocks, because as we know, good NR is not fast.
I don't agree with Michael's statement in the review "Some people may not be aware of it, but Color / Chroma Noise and its associated removal tool in various processing programs is there to remove noise introduced by Bayer Filter Array decoding. One would therefore assume that a Foveon type X3 sensor, which doesn't use a Bayer Matrix, wouldn't have Color Noise, and indeed it doesn't." Chroma noise comes from the colour correction matrix amplifying sensor noise. Without such a matrix a Bayer CFA image looks a bit under-saturated, and has some hue errors. As noted, the Foveon has a very aggressive colour correction matrix, and you can see how necessary that is by looking at the paper above with the spectral responses of the three layers and by how much they overlap each other. It's the overlap that needs to be removed by the CC matrix, and the more overlap, the stronger the matrix, and the more noise amplification, hence the necessary NR steps in the Foveon processing mentioned above. In a Bayer CFA, chroma noise does not come from the demosaic process, but instead from the CC matrix, which due to the use of good dye filters on the CFA is a much less noise inducing matrix than that used on Foveon.
I think the colour anomalies noticed by Michael, the red hue in the shadows, the colours not being "right" are fully due to the issues with extracting colour information from the Foveon sensor. As all things are a balance, these colour anomalies and associated NR processing are the price paid for lack of chroma moire. If you don't get chroma moire in your shots (and even if you do, it's usually quite easy to remove), you'll achieve much better colour accuracy from a Bayer CFA.
Graeme
