This addendum references my image sensor cover story in EDN's September 16, 2004 issue.
Below please find links to Zip'd collections of the report files outputs by Starzen Technologies' free Dead Pixel Test utility. The 21 ASCII text files inside each ZIP have names reflective of the shutter speed which generated them. For the moment, at least, I've decided not to post any of the TIFF files that Dead Pixel Test analyzed, due to their extremely large sizes (approximately 13 MBytes each, uncompressed). If I can free up sufficient server space, I'll post at least a few of the files later; if you have a burning need for them and are willing to provide blank CD-Rs and return postage; drop me an email and I'll consider your request.
Original camera, noise reduction off
Original camera, noise reduction on
"Replacement" camera, noise reduction off
"Replacement" camera, noise reduction on
Thanks to Foveon for arranging for the loan of a Sigma SD10 digital SLR, containing a Foveon 3.4 million pixel site (10.2 million photodiode) X3 CMOS sensor. I compared pictures taken with it to those taken with my Pentax *ist D DSLR (containing a 6.1 Mpixel CCD, with 3.05 million green-spectrum, 1.525 million red-spectrum and 1.525 million blue-spectrum photodiodes) at a variety of focal lengths and illumination conditions. As my prior research had suggested, I found that the two cameras created images of quite similar quality; not terribly surprising because they contained almost exactly the same number of green spectrum-sensitive photodiodes (remember that the human visual system is most sensitive to green-spectrum light, and therefore detail in that portion of the visible spectrum is most important to capture).
The SD10's CMOS sensor contains roughly twice as many red- and blue-spectrum sensitive photodiodes as those in the CCD in the *ist D. In spite of Foveon's claims to the contrary, I found no practical indication that this X3 sensor 'specification superiority' led to superior images. Could I create handcrafted test images that would result in artifacts when Bayer pattern matrix filter interpolation was applied to them? Sure. But for real-life scenes, at least the ones I chose, interpolation held its own. Compare the SD10's price (along with the prices of lenses you might want to buy) against that of a Canon, Kodak, Nikon or Pentax DSLR containing a conventional filter matrix image sensor, and decide whether test image results are sufficient justification for any incremental SD10 price you might incur.
My primary criticisms of the SD10 centered not on the images the sensor created, but with the features offered by the camera body containing the sensor. The SD10 is a much bigger and bulkier camera than the *ist D, unlike the *ist D it doesn't offer a built-in flash unit, and it presents (at least in my opinion) an overall much less 'polished' package. Its center-weighted autofocus system is comparatively archaic, as is its exposure metering system, it accepts only nonstandard battery packs (the *ist D will optionally run on AA batteries), and it outputs only RAW-format files which must be run through Adobe Photoshop CS's RAW import filter, or through Sigma's standalone conversion utility, to transform them into JPEGs, TIFFs, or other formats which common image editing and viewing software packages support.
Any comments or questions? I welcome your emails!