Tech question on pixel shift

[Scott Auerbach]

I have no reason to NEED to know this, but inquiring minds, eh?

Is anyone aware of a good, thorough explanation of how pixel shift works? The standard description is that the G CCD is offset compared to the B & R CCDs. But I don't see how that alone would increase resolution any more than offsetting one color printing plate (relative to the others in the CMYK array) would. In fact, it would just look out of register.

I'm assuming there's something more sophisticated going on, and having read a brief description of the physical layout of an interline transfer CCD (something that's not accurately reflected in most diagrammatic illustrations of pixel layout on a CCD, where they're usually explaining the 4:2:2 Bayer layout), I can see how the active pixels could be shifted into the interline area. But that would (seemingly) require that the chip be physically oscillating, which I can't believe it does. The only other option involves some staggered checkerboarding of active pixels and interline areas, with rapid switching between active and inactive areas. But that would mean the G CCD is built to entirely different specs than the other two, and would seem to imply the smaller pixel area that pixel-shifting is designed to avoid.

So... as Ricky Ricardo used to say, "Loooocy, I'm confooooosed."

Anyone have a link to something with good illustrations that explains what's really going on?

[Barry Green]

Getting good third-party information on the process is vexing; the best I've found is a BBC white paper that doesn't describe the process in detail, but does say that it's a legit technique that can result in as much as a 1.414x resolution boost (square root of 2), although 33% is a more practical target.

There is no "shifting" or oscillating going on. The CCDs are permanently mounted in a 1/2-pixel offset. Panasonic calls it "spatial offset" rather than "pixel shift" because "shift" implies a verb, it implies that there's some manner of active shifting going on, and that's not how it works. It's more like a Bayer demosaic.

Panasonic's white paper description can be found here:
http://www.panasonic.com/business/pr...site/index.asp

Look in the upper right and click on "technical paper".

[Scott Auerbach]

Originally Posted by Barry Green

Panasonic's white paper description can be found here:
http://www.panasonic.com/business/pr...site/index.asp

Look in the upper right and click on "technical paper".

They must've pulled it off the site; I don't see it there. Dang.

[Rodrigo Otaviano]

Originally Posted by Scott Auerbach

They must've pulled it off the site; I don't see it there. Dang.

Click on "Resources" first.

[Chris Hurd]

It's definitely a "legitimate" technique that's been around for years. Pioneered by Panasonic and used by Sony as well as Canon. The best explanation I've heard is that pixel shift produces more sampling points for the analog to digital conversion process. The more points there are to describe a curve, the smoother it can be drawn. It's an effective way to boost resolution.

[Dee Joslin]

And this is good reading. It may already be in the forum but I haven't seen it.

ftp://ftp.panasonic.com/pub/Panasoni...200.CCD-WP.pdf

[Pete Tews]

That's a very informative pdf...

[Scott Auerbach]

Thanks. That kinda-sorta begins to answer some of the questions that'd been rattling around in my brain, though I suspect the real answer is in what the DSP 'black box' does when it combines the offset signals and interpolates extra lines out of that. It also clears up one puzzling misconception I had: I thought it was a full pixel offset, which didn't make any sense at all. Seeing that it's a .5 pixel offset in both axes makes a LOT more sense.