HDV1080i60 with Green Screen Footage

Hello!

Currently I am working on green screen footage in FCP 6.0.5 which was digitized with compression setting : HDV1080i60 1440x1080 29.97 fps

However, it looks very compressed and I see tons of artifacts in the skin and unforunately my keys look very stepped.

Can anyone out there please lend me some advice for a cleaner look to my footage?
The end result will be projected in a movie theatre and I need it to look as clean as possible.

Here are some screen grabs.

That looks perfectly normal for HDV footage, to my eye.

You might try adding a denoise before your key, but that might make it harder to pull clean keys, because it'll soften everything. The truth is, that's just how your footage is, and you're going to have to live with it.

Whoever made the choice to shoot for theatrical presentation in HDV and at 60i, however, should be fired immediately. The HDV thing, okay, I can kind of understand that, maybe. Different cameras perform differently, and while none of it is great, some of it can be acceptable. But shooting 60i for film out? That's a worse-than-rookie mistake. That's an "I have no idea what I'm doing" mistake.

Thank you...I was afraid of that. I had better quality with standard definition in the past!
There must be something better.
Just to clarify...It's not for film output. That indeed is crazy. It will be a video projected in a movie theater.

HDV is HIGHLY COMPRESSED. I mean, how else can they fit the same amount of data as DV, on the same tape?! Tells you something about it. It will be that steppy. You can try rendering in a ProRes sequence setting, That might improve things a tad, but once recorded onto HDV tape...that info is gone.

4:2:0 Chroma Sampling. 0. something is missing.


www.shanerosseditor.com

Listen to THE EDIT BAY Podcast on iTunes
[itunes.apple.com]

> It will be a video projected in a movie theater.

There's that magic word- 'projected'. Shoot progressive.

>I had better quality with standard definition in the past!

But you don't mean DV, do you? DV's pretty steppy especially with bright over saturated colours.



www.strypesinpost.com

Shane Ross Wrote:

> 4:2:0 Chroma Sampling. 0. something is missing.

I think I need a clarification here. As I understand it:

4:4:4 It is all there
4:2:2 Chroma resolution is reduced
4:1:1 Chroma resolution is reduced further
4:2:0 About the same chroma resolution as 4:1:1 but the samples in the chroma channels are not co-sited with the luma samples, but rather are averaged between two adjacent lines.

Not picking nits exactly, but as this is actually a departure from the ratio nomenclature.

Your basic point, that something is missing is true, but at about the same rate of missing as other compression schemes. It is not as if we don't sample the Y-B all of a sudden.

-Vance

Yea, that's pretty much it. I think Shane was mentioning about the heavy chroma compression, which of course, affects keying greatly, or just about anything to do with color manipulation.



www.strypesinpost.com

The names for the different chroma subsampling mechanisms are confusing. It's generally unhelpful to think of them as ratios; they're really just names, not mathematical descriptions.

Let's say you're working in full-raster HD. This would not be the case with HDV, 'cause HDV isn't a full-raster format, but I like keeping things simple, so we're going to assume you're starting out with 1920x1080.

Television is a YUV format; each image is separated into one luminance image and two color-component images. This is different from film or synthetic computer imagery, where each frame is composed of a frame of red, a frame of green and a frame of blue added together. In television the math for generating a full-color image out of the components is more complicated, but it has the virtue of giving you a separate black-and-white image that can be displayed by itself with no math. Back when color television had to be invented with no color televisions on the market, that was a big deal, 'cause it meant people could watch color broadcasts on their black-and-white sets.

So: one black-and-white image, and two color-component images. There's no reason, technically, why the black-and-white and color-component images can't all be the same resolution: same number of samples per scan line, same number of scan lines per frame. But to keep the bandwidth of color broadcast television within reason, it was decided that we'd be thrifty by having only half the number of samples per scan line in each of the two color components. So your 1920x1080 footage has a 1920x1080 luminance image, but the two color images are only 960x1080, with each sample being "twice as wide," as it were.

That's 4:2:2. Why is it called 4:2:2 instead of 2:1:1? I'm sure there's an excellent answer out there somewhere, but I don't know it. That's just how it is, as far as I'm concerned.

If all three of your components were full-resolution, we'd call it 4:4:4. There are no 4:4:4 broadcast formats, but 4:4:4 is the standard in film production. Where film and television production meet ? stuff that's shot on film but delivered exclusively on television, like TV commercials ? the transition from 4:4:4 to 4:2:2 has to happen somewhere. Sometimes it's at the telecine stage, and sometimes it's in the online room. If you scan your film to something like HDCAM SR, then you're (probably) scanning to 4:2:2. If you scan DPX sequences for online assembly, then you're scanning 4:4:4, but everything gets converted to 4:2:2 when you output to tape.

So 4:2:2 is television, and 4:4:4 is a higher-quality variation that's used universally for film and occasionally in television post.

But the advent of digital video made it possible for other chroma subsampling schemes to crop up. The first of these was DV's 4:1:1. Whereas normal broadcast television has half the samples per scan line in the color components as compared to the black-and-white image, DV has one quarter the samples. So a scan line with 1920 luminance samples would have only 480 chroma samples. Although old-school post pros love to get snotty about it (including and especially me), the overwhelming majority of TV news, and a significant amount of TV sports, was shot with 4:1:1 chroma subsampling in the days before HD.

So 4:1:1 was kind of crap technically, but gosh if it didn't actually work in practice, so for a time it was good. But there's yet another chroma subsampling scheme out there, employed by basically every format that's based on MPEG-2 or MPEG-4. It's called 4:2:0, and here's where the names totally break down. If 4:2:2 has half the color samples per line as luminance samples, and 4:1:1 has a quarter the color samples per line as luminance samples, then 4:2:0 must have half the color samples in one channel, and no samples at all in the other! Which is actually kind of true, on a scan-line-by-scan-line basis, but it's misleading in the extreme.

Without going into too much detail ? 'cause the details differ from implementation to implementation, and I can't keep them straight in my head ? 4:2:0 basically works like this: As in 4:2:2, the number of color samples per scan line is half that of the number of luminance samples. So you have 960 color samples per scan line instead of 1920. But unlike 4:2:2, you also have half the effective number of vertical color samples. In 4:2:2, each scan line contains samples of luminance and samples of two different color components. But if you imagine that in 4:2:0, each line contains luminance and only one of the two color components, and each scan line alternates, then you're pretty close to the truth. In all the other formats, each scan line of the image is the same structurally speaking, but in 4:2:0, half the scan lines have no samples of one color component, and the other half have no samples of the other color component.

In a nutshell, 4:2:0 includes one quarter of the color information that a hypothetical 4:4:4 image would include. But 4:1:1 also includes only one quarter of the color information. The difference between the two is how that color information is reduced.

Generally speaking, 4:2:0 is superior for progressively-scanned material. DVDs are 4:2:0, as is Blu-Ray, because both of those media formats are mostly for 24p stuff. But because of the sharing of color information across scan lines, 4:2:0 is inferior to 4:1:1 for interlaced material, which is part of what makes HDV such crap. HDV at 24p is actually pretty darned good. It's not perfect, but it holds its own pretty admirably against DVCPRO HD in relatively static shots like interiors. But at 60i, HDV is just noise that just coincidentally happens to resemble what you shot.

Now, the nightmare is converting from 4:1:1 to 4:2:0. Countless thousands of enthusiastic AV nerds have gone out and shot their home movies on DV, converted them to DVD and wailed in despair at how assy their lovingly shot and edited mini-masterpieces look on television. Well, yeah, Poindexter, 'cause you're throwing away virtually all the color information in the image. Better by far to just desaturate everything in post and pitch your short as a retro pastiche, since the your footage won't survive the color-sample genocide you're imposing on it during format conversion anyway.

So how does all this apply to green-screen? Professional green-screen work is always done at 4:4:4. Period, end of paragraph. If some 4:2:2 stuff has to be keyed out, be prepared to listen to the Flame artist grumble and complain every step of the way, loudly and frequently asserting that this shot sure as hell isn't going on his reel, and could his credit maybe say something like "creative consultant" or "craft services" instead of "compositor?"

Okay, not really. In truth, ever since the second "Star Wars" prequel went into production using an F900 and an ocean of green paint, techniques for dealing with 4:2:2 stuff have improved. But the pros can spot the difference between 4:2:2 and 4:4:4 within ten seconds in the master keyer in Flame, and they won't hesitate to express their opinions on how the shot should have been done.

But really, the 4:2:0 chroma subsampling is only one reason why your HDV material looks so much like ass. Another is that it looks like your greenscreen was astonishingly underexposed on-set. Depending on your camera's response, a greenscreen should be exposed around 50 IRE ? maybe 40, maybe 60, but always in that neighborhood. Finally, from some of the ringing around the details in that yellow jacket-thing, it looks like the camera was set up to do some sharpening of the image. NO BAD WRONG STOP. Artificial sharpening is the sworn enemy of compositing. Once I had to spend an extra day fiddling with dilate/erodes to get rid of the ring around the talent's sportcoat, because the DP thought in-camera sharpening was the bee's knees.

It is possible to shoot good greenscreen plates on HDV. But unlike shooting on, say, SFX 200T, with its combination of fine grain and 10+ stops of latitude, HDV is a powerfully unforgiving format. If you didn't shoot it exactly right, then you shot it wrong, and precious little can be done to it in compositing.

Jeff-

Great and detailed answer! When is your book coming out? All you need is about three more answers at this level and chapter 1 is done.


Jeff Harrell Wrote:
-------------------------------------------------------

> That's 4:2:2. Why is it called 4:2:2 instead of
> 2:1:1? I'm sure there's an excellent answer out
> there somewhere, but I don't know it. That's just
> how it is, as far as I'm concerned.


Hey I know the answer to this one. Some of the earliest use of digital sampling of analog video pre-dates the common use of component video. Those early systems, like time base correctors, need a sample clock to drive the D to A convertors. One available choice was the subcarrier signal. But to meet the needs of digital sampling you have to sample a signal a minimum of at least twice the highest frequency you need to capture. As video has detail at even higher frequency than color information engineers used either 3 or 4 times the subcarrier frequency. Thus the "4" in 4:2:2.

When D1 digital systems appeared they featured a 13.5 MHz clock that worked for both 25 and 30 fps systems. That was close to the 4 times subcarrier frequency so the nomenclature was retained.

See? I knew there was a great answer out there.

It reminds me, though, of the old anecdote about why the standard US railroad gauge is what it is.

With of the south side of a north bound horse?

-V

Basically, yeah.