I don't actually know what he is referring to for sure, but it sure sounds like he is using a sensor which tries to get around the bayer pattern problem of lowered definition by interpolation of colour information.
Bayer pattern sensors use a pattern like this or something similar:
GR
BG
Upon taking the picture, the computer attempts to combine this information into full colour information on each pixel. This means that for the Green pixel, the chip will guess based on the nearby pixels and add their information to create a full colour peice of information. This is actually sliding colour information around in physical space as can be seen above. The Red Channel information has to be moved slightly over to be combined and keep the pixel numbers the same. The name for adding information is interpolation. It is a guess by the computer and typically causes a loss of fine detail. Additionally, because of the guesswork, it tends to increase noise. Canon sensors gather twice as much information in the green channel because there are twice as many. This means that Green Channel noise is less than blue and red channel noise (check dpreview for confirmation here - they have charts on noise in specific colour channels and have a little writeup on this on the review for the nikon d2x I think).
The Sigma Foveon sensor tried to get around this by using 3 photoreceptors per pixel, using around 10 million photoreceptors for an effectively 3.x megapixel sensor. This allowed each pixel to record three separate channels of information, allowing the colour information to have an absolute position. NO INTERPOLATION was required. The problem was that a 10 megapixel sensor was only able to yield a 3.something megapixel picture and megapixels do count in marketing. The 3.x megapixel sensor in the Sigma was able to capture the same, perhaps better resolution information as a 6 megapixel D70 or a 10d. Theoretically, it would be able to capture more real colour information with larger photosites allowing much less noise, requiring much less on-chip noise reduction than a similar bayer pattern sensor. Additionally, without interpolation, all resolution detail would be perfectly accurate and far sharper than any Bayer interpolated image. Unfortunately, on-chip noise reduction in the new Canons than in the Sigma chip and this technology is pretty much dead now. More than 2 years have passed since Sigma's last release.
So what are they talking about with multi-shot functions and not having to interpolate colour information? I can only assume that they are using 4 channel colour space and are using 4 shots with EACH PIXEL taking only one channel of colour information at a time. Information could then be layered, channel on top of channel, creating 4 layers of colour and luminance information in EACH pixel. This would totally bypass the Bayer offset pattern and create perfectly aligned pixels with no interpolation. Fine detail would be tripled and noise would be significantly less.
If the pattern of detail held to the same effectiveness as the Sigma, this would more than double the megapixel to quality ratio. A 22MP digital back would then be able to shoot around the 45-50MP level of clarity and quality or better when compared to a Bayer pattern sensor. The increase in clarity without interpolation would trash moire patterns which are usually a result of computer guesswork.
Obviously this would only be doable with VERY still subjects on VERY good tripods.
Hope it makes sense. I'm pretty sure this is accurate.
Message edited by author 2005-08-19 01:07:07. |
