To Jan Crittenden: HVX-200 Green Screen Halo Problem

[Per Holmes]

Thanks!

Per

[Dean Sensui]

Per…

Here’s what I do if I’m faced with what you experienced.

Create separate keys to suit specific types of detail.

Key for the smooth or sharp edges.

Then do a separate key for frizzy hair or other types of detail.

It’s more troublesome to do it this way, but from what I understand it’s not unusual for the high-end production composite artists to do this.

Good luck!

Dean Sensui — Imagination Media Hawaii

[Per Holmes]

Hi Dean,

Thanks for your post. Yes, it’s very common to use even 5 or 6 keyers, one optimized for hair, other keyers optimized body, shadows, contact shadows etc. Also, several keyers is the way to handle a green screen that is not evenly lit, and with a cyc, it’s almost impossible it to get it completely even.

Several keyers don’t actually fix this particular problem, because since the issue affects one entire side of the actor, including their hair, there’s almost no way to avoid some sort of rim with the keyer alone, without eating considerably into the edges. The edge itself needs to be evened out before it’s fed to the keyer, and treating the halo as if it’s a chroma echo, as described earlier I’ve been quite successful doing a sort of ‘reverse chroma echo’.

I still can’t figure out this issue, although I’ve determined that the HVX-200 is not the only camera to exhibit it. My Nikon D70s also has it, but since I have no way of knowing what kind of edge treatment they do, I don’t want to use it as an absolute reference, or even say for sure that it has the problem natively.

The D70s uses a Bayer pattern on the CCD, meaning that the RGB image is assembled from underlying CCDs that themselves have lower resolution than the final image, and for most types of photography, this is perceptually irrelevant. The HVX also assembles an image from CCDs of lesser resolution, and although what the HVX does is completely different. But the bottom line still is that the color channels from the CCDs (or single Bayer CCD) are assembled in unique ways to produce a resolution that is higher than the native CCDs. Normally, it’s not a problem, but it may give these cameras a hard time with extreme changes in color.

Regardless, since I’ve seen DVX-100 Andromeda (4:4:4 10-bit) footage that didn’t have the halo, I’m still extremely curious what the actual native data from the CCD in the HVX looks like. I’m not completely convinced that the halo is not an artifact from some DSP processing, something that could possibly be disabled, yielding a yet more powerful camera basically for free.

Best,

Per

[Dan Hundley]

I’m also seeing the halo problem. Always on the right side of the subject. Here’s a screen grab. Shot with HVX200 1080i 24PA, b.press, everything else default, Quicktime DVCPRO HD codec.

https://www.token.com/dh/test1.png

[Per Holmes]

Hi Dan,

Great! I mean Great 🙁

Please leave the PNG file on your server, I’ll leave mine as well so this thread may serve some value for a while.

I’m also always seeing the halo on the right side. I’m noticing one thing quite clearly on your shot because of the contrast between the shirt and the background, that your left side has a corresponding white rim which I’m sure is not a backlight. This therefore reeks of a chroma-shift of a couple of pixels. However, when I convert to YUV and shift the UV channels to the left two pixels, it doesn’t look right.

I think that at least one possibility is that it is the result of how a higher resolution image is assembled from lower resolution CCDs, which is usually photographically irrelevant, except when you do dramatic color and contrast shifts as you do around subject edges on a green screen.

Another thing that has struck me is the possibility of the infra-red light used to run the auto-focus, that it somehow front projects on the green screen and comes back as a shadow, like a flash on a camera. I think this theory is not really plausible, but I’m grasping at straws to come up with an explanation.

For sure, I think the ‘lighting explanation’ is bankrupt, that lighting is somehow the cause. There is no law of physics that allows that kind of ‘dark matter absorbtion’ literally outside the subject. That’s total science fiction. If different lighting for some people makes it go away or minimizes it, then they’re really just suppressing the effects of the problem — dark clothes transition with the black halo so much that you can’t see the halo. It’s not really gone.

Anyway, thanks for posting! I hope we figure out what it is. And if it’s fundamental to the design at this fairly low price point, then I’m totally fine. But if it’s a bug, it should be fixed for the benefit of everyone.

Per

[Joel Arvidsson]

Thanks for an intresting thread, and I hope its not dead quite yet. Can you explain you process how to remove the chroma eco. I’m not quiet following. Can you even post a picture of the nodes. I like the idea of fixing this before the key.
Cheers Joel

joelarvidsson@gmail.com

[Joel Arvidsson]

Thanks for an intresting thread, and I hope its not dead quite yet. Can you explain youre process how to remove the chroma eco. I’m not quiet following. Can you even post a picture of the nodes. I like the idea of fixing this before the key.
Cheers Joel

joelarvidsson@gmail.com

[Joel Arvidsson]

Just wanted to say that I got it to work.

joelarvidsson@gmail.com

[Andrew Shanks]

Just chiming in a few months late on this one, but I wanted to back up Per as we encountered this last year with the HVX as well. I hasten to add I am a senior compositor and vfx supervisor, I work on everything from TV to feature films, I am not a newbie who thinks a one click key is a finished product. I know a lot of folk who have the hvx will be fine with the one click solution in their editing packages (and won’t even see the halo), …for almost all corporate, music video, certain TV shows, etc, it will be totally fine. For those that want realism in their comps, well, the halo is an issue. You will see this no matter what keyer and what compositing system you use (I have seen it in shake as well as after effects, using primatte and keylight on both systems).

At the time we did our own tests and at -7 (for the sharpening) the haloing was the least present, but the super soft screen freaked the DoP and Director out (despite us saying we could introduce the sharpening in post), so we compromised and let them shoot at -3.

I can’t speak for certain, but I’d assume the lads down at WETA worked with the footage the way we did, and that was with proceedural keying of different areas to build the final mattes. When you’re used to working with 35mm and RED dpx files, it is a shock to the system to have to deal with these fringing artifacts, …but as someone else says, I guess thats the tradeoff for the price of the camera.

Will be interesting to see what the new ccd in the 170 is like and if it has the same issue.

One thing to say about the HVX, …at least you don’t have to deal with 3.1.1 colourspace or rolling shutter issues (…probably more a bain in my life presently than halo issues).

I just wanted to speak up because Per seemed to get a raw deal in this thread, he was just seeking an answer and his methods were questioned just because others haven’t seen it, others who don’t do this day in and day out (to the levels some of us do, …I will be the first to agree that if you crunch your matte like most users tend to do you won’t have an issue, …but edge is what keying is all about and if you can avoid matte choking, you should).

Can I suggest that someone who “hasn’t” had this issue when keying using the hvx posts a still for us to see, that would boost the arguement for that side of the discussion (I’d be intrigued to see them myself), …unless I have missed a post, so far the only posted shots have been from those that have had problems.

Cheers,

andrew

[Jeremy Garchow]

[Andrew Shanks] “When you’re used to working with 35mm and RED dpx files, it is a shock to the system to have to deal with these fringing artifacts, …but as someone else says, I guess thats the tradeoff for the price of the camera. “

I don’t think it’s fair to put the HVX200 against 35mm film or the Red. Three totally different cameras and acquisition situations and I can’t believe it was a shock to the system that the HVX underperformed when compared to an industry standby and ground breaking technology. Don’t you think that’s true, Andrew?

Thanks for the follow up and I agree that Per did get a bit of a raw deal, but hey at least we have a place to hash this all out.

Did anyone ever capture live out of the component to get a better key and perhaps no fringing? I ran a VERY informal key test capturing at the same time the hvx200 via p2, via component to ProResHQ and via Uncompressed 10bit via SDI through a component a/d. The ProRes and Uncompressed came out remarkably better, but I didn’t check for fringing. I can probably post stills for you when I return from the desert.

Jeremy