Dave, I like the photographs you took to examine the SD9 and 350D. They show up a number of issues in both cameras quite clearly, and are wonderful diagnostic images. Thanks for sharing them with us.
I looked in the metadata and see the Canon was converted with Bibble - is this correct? The demosaic is awful. Now, as I'm saying that, it's only fair to say that such a comment comes from my experience with the the RED camera project, and we're currently producing images like this:
http://www.reduser.net/forum/uploaded/9_2kpomona_01033.jpg at 72fps. I'm not going to say the images are perfect - there's always room for improvement, and I'd also point out that we're recording to CF card through an advanced RAW compression scheme that's lossy (uncompressed or losslessly compressed is not really possible to record to a CF card at the fps and frame sizes we're doing) with a remarkably low data rate. If you're interested, here's the forum thread about those images:
http://www.reduser.net/forum/showthread.php?t=5624 if you're interested.
I started this thread to talk about aliasing. It's all the more important as moving images show up the issues of aliasing much more clearly than stills, and you can't in all honesty, go in and paint and fix up a movie the way you can on a still.
And then we took a track that started to talk about Foveons. That's an interesting topic in it's own right, but it's also a tad "religious". They get brought into any discussion on aliasing because the don't have an optical low pass filter. Most Bayer CFA cameras do, because without, the bayer pattern reconstruction algorithms don't always work as well as you'd like and chroma fringing artifacts are a problem. Interesting that Phantom, maker of high speed Bayer CFA movie cameras have announced a "upgrade" to put an OLPF in front of their sensor. And it's removable, so if you don't want / need it, you can remove it.
But, when we talk about images, and what we see in them, we are always talking about a "reconstructed" image. Look at a RAW Bayer CFA image, as I often have to do, and it looks awful - black and white, a mosaic of little dots, and very dark as it's linear light. It needs to go through complex algorithms to allow you to see the RGB image. Those algorithms are often proprietary, and include all kinds of processing from noise reduction, demosaicing, sharpening etc. etc. Similarly a RAW Foveon image would also look very dark, with funny colours, but it would not have the mosaic pattern as it uses co-sited samples. It also needs a number of steps to turn it into a viewable RGB image that looks like the scene you've photographed. The Foveon white papers hint at the processing involved, and pixel-peeping can tell you more about what they do, but for the most part, it's proprietary. I don't know precisely what they do, but I can make educated guesses as to the nature of the processing needed to generate the RGB image, and I suspect it's of the same order of complexity as we use for demosaicing Bayer CFA images, just different.
So, when you compare an image, there's a whole host of factors you're adding into the mix, not least all the algorithms that conver the RAW to RGB. So when you start a discussion about aliasing, and try to keep it about aliasing, how it works, how OLPFs work, what positive and negative effects they have, you can keep it all quite reasonable. When you look at the real world on how cameras and photographers work with their images, the number of variables go up through the roof, and it's very hard to get to the bottom of it all.
I'd dearly like to see some Sigma images from a camera modded to include an appropriate OLPF. Lin, do you know if anyone has done this? There are enough examples of Bayer CFA cameras modded to remove the OLPF, but to see the other way around, I think, would be enlightening.
Graeme