When pixelshifting matters... Keying/Compositing

gunleik

gunleik

Well-known member
This fall I've been doing quite a lot of greenscreen (and some bluescreen) work with several HVX's and I've found some interesting stuff.

This is an un-scientfic, but observed problem description, that makes sense - to me -:)

The pixelshift gives you "better" images from the 960x540 sensor. Mostly seen in better gradients, thus giving the impression of a "better" image. - which is good.

The 4:2:2 colorspace adds to the "superior" image compared to 4:2:0 or 4:1:1 images.

BUT
When you do compositing and keying, you quite often relate only to one channel, and then the rez difference between SD and the HVX becomes debateable at best.

To me it looks like the pixelshifting actually makes it worse, because the fringes seems to be approximated from a mix of the channels.

So the pixel-messing technology which makes for a beautifull full image, actually messes up in compositing and keying situations.


Anyone else noticing this?

Gunleik
 
No... quite the opposite; Jim Arthurs said the HVX was a breakthrough for keying, and the guys from WETA Workshops put an HVX in an extensive chroma key test (against an XDCAM-HD cam, for that matter) and ended up buying the HVX.

We've just finished shooting a multiple-hour training DVD, all greenscreen... so I'm not sure what you're running into?
 
I noticed on the Canon MiniDV cameras with pixel shift I would get some funky edge artifacts, but I always attributed it to the 4:2:0. If you watch "open water" you can tell it was shot with Canons because any time you get silhouettes the edges get blocky, like they are half resolution.

I like the HVX for chromakey but if I could fix one thing it would be the CA issue.
Barry, where can I read more about that Chroma Key test?
 
Open Water was shot on a Sony, not Canon.

The WETA thing was told to me directly by the employees who did the test, there's not any sort of web-based report on it or anything.
 
Just to be clear:

I get ok/good keys from the HVX, but better from say digi -:) and have made a fair amount of commercial $$ with it this way thae last half a year.

What I (think) I see, when I open the images in Shake and look at the channels, is that the shifting makes for better colorgradients and say circles and diagonals, but when I look at only one channel (mostly green) they're more steppy (which makes sense...) BUT it seems like the camera is doing some "judgement" where it mixes the channels in gradients @ the edge to make it look smoother than it is, which in turn messes up the key a bit. This is quite solveable, but adds time to the keying process.

Of course, this can have several sources (in addition to bad lightning et al)

Main suspect:
960x540

Second suspect:
DVCPRO HD

Third suspect (related to second) :)
Lack of a HD-SDI clean signal

Sometimes it helps recording from component out instead of P2, but the component signal on the HVX is also not too good...

I see now - as I type, I should do a DVCPRO 50 test to compare results - for pure comparative problem fixation. But then again: All our productions are 1080p, so that doesn't make a lot of production sense for us.

I have read several times that "the internal processing of the HVX is 1080p" - and then recorded into a DVCPRO-HD stream. I wonder if this process also adds to the mess, if the convertion from square to none-square pixels also is "done on the cheap". Would correspond well with what I think I see. (muddy waters in single channels around edges, when the "full" image is pretty clean.)

I'm not comparing the HVX to 4:2:0 or 4:1:1 cams for obvious reasons (be it XDCAM HD or not). The 4:2:2 colorspace was one of the main selling points. All I say is that the rez of the chip takes away some of the advantage of the color-rez, when you need to work with single channels. This is actually a fairly obvious statement, I just managed not to be aware of it before I started working much with it. Comparing it to DV/HDV/mpeg2 is not really relevant....

We can record 3 streams of 1080 4:4:4 uncompressed live in studio and key close to realtime in Shake through dual HD-SDI to a server setup -:). We now try to make this workflow more practical. (Bummer that Red doesn't currently give you 4:4:4 through HD-SDI...)

Back to the subject: It will be interesting to test the HVX against the 1/2" new Sony EX (through HD-SDI)

And to be clear: We have 5 HVXs with a bunch of P2 cards, firestores and, field monitors and fun stuff. We are not totally unfond of the little bugger. That is NOT the issue at hand.

Gunleik
 
Last edited:
Actually the WETA story makes a lot of sense, despite my findings - if they mainly shoot close ups of small scale elements for comps.

The cams deep DOF and good colormetry will excell in such a situation.

Problem arises on "bigger" sets with more actors @ full size within the image.

Then the lines and hair I'm talking about will be comparatively smaller and the issues bigger...

Gunleik
 
I'd be surprised.

It is possible to get rid of the black halo, but not in a satisfactory way.

Remember: This is little/no issue for real closeups.

A headshot wavering in the wind with hair all over is ok.

Try to do that with 3 full human figures in motion, that is when the resolution-loss and huge halo crap hits the fan.

Gunleik
 
You must log in or register to reply here.
Back
Top