Posted 20 September 2009 - 05:27 PM
Posted 22 September 2009 - 01:40 PM
That being said, the spatial arrangement of the Bayer pattern has a sort of 4:2:0-ishness to it in that it gives twice as much spatial resolution to green as to red or blue, which are undersampled equally both horizontally and vertically. The human visual system has much more resolution for brightness than for color, which is why coarser sampling of the color differences works. Green makes the greatest contribution to luminance, about 60 - 70%, so Bayer's extra weighting for green is using the same thinking.
Bottom line, trying to use 4:x:x notation to describe the output of a Bayer sensor is the wrong tool for the job. It's kinda like trying to check your tire pressure by using a tape measure.
4:2:2, BTW, is kind of a weird old thing. It's an artifact of analog CRT TV, where you couldn't undersample vertically because there was no way to know what was on the adjacent lines. As soon as we had digital TV and enough memory to store a frame, it became unnecessary. Now that we have memory, we should forget about it. ;-)
Posted 22 September 2009 - 03:49 PM
Shoot some resolution test targets and see what you're actually getting, and make sure you shoot some very saturated ones too (this is what nobody's ever done on a Red, which is why it sometimes looks better than we know it can be).
This will characterise the optics and transport mechanisms you're using, which will probably cost you quite a bit more than the bayer patterning on any sane sensor. I'd have said 4.5K bayer is probably a good 2.5-3K RGB, but you'll lose more than that in the glass.
Posted 23 September 2009 - 01:21 AM
Posted 23 September 2009 - 05:40 AM
I think Bayer is fine so long as you oversample sufficiently. You'd probably make much more difference to perceived image quality and the amount of useful information you're capturing by using your three exposures to bracket brightness.
That said, the best stuff I've ever seen done with a lashed-up scanning rig such as what we're discussing here did nine exposures per frame, three each for red, green and blue. I have no idea what he was gaining out of this or whether other factors made his rig good.
Posted 25 September 2009 - 07:32 PM
If I shot three separate, monochrome, RGB captures of each image through a filter wheel and jammed them together digitally, would that put me back into something 4:4:4-ish?
If you do that with a monochrome sensor, yes, that's 4:4:4. That's how scanning is done. If you do that with a Bayer chip, you'd get the same spatial sampling structure, but you could achieve more saturated primaries.