Colour Sampling 4 2 0 not good for Chromakeying

G

GuyS

Active member
It is my understanding that Sony XDCam-HD uses 4:2:0 colour sampling. The Panasonic HVX 200 uses DVC PRO HD, which is 4:2:2 colour sampling. There is a huge advantage when chroma keying with 4:2:2 technology.

I am interested to see if this was a concern for the Sony buyers. I am contemplating purchasing a pro level HD cam but the ability to chroma key will be a major decision factor. If I am understanding the technology correctly the Panasonic is better than the Sony for chromakeying.
 
I'm no pro in this, but I've always heard the same thing. 4:2:2 is much better for green screening. Red is coming out with a Scarlet, which will probably be 4:4:4 (no color sampling). So, if I were you I'd wait until April. :happy:
 
I've been concerned about this and following threads about it for awhile and the general consensus is.
1. 4:2:0 is less a big deal in progressive than interlace.

2. Those numbers are relationships of chroma info to luminance and since the luminance info on the EX is mountains greater than the HVX the chroma ends up very close.

3. in real world tests we have been hearing that it keys well. Jim Arthurs who does a great deal of Chroma Key work and has used many video cameras was knocked out by the quality of its chroma keys, and even said it was better than the HDCAM 900! I'll take the "better " with a grain of salt, but it does make me feel pretty confident it can cut a good key.
I think Jim posted here somewhere about that.

4. No one has posted that it cuts a bad key yet.
Lenny Levy
 
OTTOEMEZZO said:
I'm no pro in this, but I've always heard the same thing. 4:2:2 is much better for green screening. Red is coming out with a Scarlet, which will probably be 4:4:4 (no color sampling). So, if I were you I'd wait until April. :happy:
Click to expand...

True, but if RED announces something, IMO you will not see the camera hit the shelves for almost two years.
 
4:2:0 4:2:2 4:4:4 mean nothing unless you know what resolution the codec is running at, and for live video, codec means nothing unless you know where the image is coming from (ccd, adc, dsp, etc). in theory the 640x1080 (691k pixels) color resolution of DVCPROHD has 33% more pixels than 960x540 (518k) color resolution of XDCAMEX. in the case of the hvx200, however, the color CCDs are 960x540 so it is not possible to resolve more than 540 lines of resolution in any color, making its effective color resolution closer to 640x540 (346k), 33% less than the EX1 . in my experience, the hvx200 also introduces some odd chroma sampling artifacts which are difficult to deal with and cannot help the chromakeying process.

as for RED, any single CMOS system using a bayer filter will probably give you 4:2:0 at best. you can always store it in a 4:4:4/RGB format, but if there is only one blue/red pixel in every 2x2 quad of pixels, it is unlikely youll get any unique data in the three other pixels youd be using to store that color information.

what seems odd to me is that people call HDCAM 3:1:1 but DVCPRO HD 4:2:2. really, by the same system, DVCPROHD should be called 2.6:1.3:1.3 or something i suppose.
 
nyvz said:
as for RED, any single CMOS system using a bayer filter will probably give you 4:2:0 at best. you can always store it in a 4:4:4/RGB format, but if there is only one blue/red pixel in every 2x2 quad of pixels, it is unlikely youll get any unique data in the three other pixels youd be using to store that color information.
Click to expand...

I think you should download some Red footage and try CC'ing it... 4:2:0, sorry dude but you're way off the mark there.
 
do you understand what i mean, though, in that if there is only one blue pixel for every four pixels? maybe saying 4:2:0 is confusing because any good debayering algorithm and good system will antialias and interpolate color information, so youll never see any of the blockiness like in an actual 4:2:0 system. maybe its more like shooting with a 4:2:0 system, converting to 4:4:4 and running an antialiasing filter on the color channels. instead of blocking youd probably see softening especially around blue or red elements. maybe youd see what i mean if you shot a resolution chart through a red or blue filter or something. in the end, with RED, its still a hell of a lot of resolution and i imagine youre more likely to be limited by the resolution of your lenses than your pixels. i might be wrong. ive never used red, but if you can explain how someone could achieve 4520x2540 of actual color resolution from an array of 2260x1270 blue/red pixels i will obviously rethink what ive said.
 
I can see what you are saying in theory, but just compare say a shot from the HVX in its own flavor of 100mbps 4:2:2, then match that with the EX1 at 35mbps at 4:2:0. The EX1 blows it away for resolution and how clean the image is. But it doesn't match the color of the HVX. I'd love to say it does, but it just doesn't. And that's why, for me, the HVX is more filmic and the EX1 looks like video. Phil Bloom's footage looks lovely when treated with Magic Bullet, but for me it still doesn't look filmic. It looks like very good video. I can only conclude that the reason why one looks more filmic than the other is the difference in color space. Still makes pretty pictures though.

Not meaning to knock the EX1, which for me is a better all round nicer camera to use than the HVX. And very nice, clean, high-rez video isn't a bad thing and I'm sure in the right hands would be able to be made to look very filmic.

Now the EX1 with 4:2:2, that would be something. And I'm sure much better for keying.
 
But the EX1 DOES do 4:2:2.

It comes out of the HD-SDI connector.

Booth,

I doubt if you saw the XDCAM from the EX1 compared to the SDI out from the same camera, you would say one looks like film and one looks like video. To the naked eye, I bet you couldn't tell the difference.

"Looks like video" is subjective and is something often said on this forum without backing it up, and is mostly meaningless when it comes to color. If the motion cadence is the same then you're 95% there. Film is graded in many ways and so is video. Video can be made to look like film and film can be made to look like video. Only at the extremes are the differences really apparent.
 
I'm no expert in EX1 but here's what I've read about the EX1. If you capture/record using the HD-SDI from the EX1 you'll not only getting the high resolution but also in 4:2:2.
 
booth said:
Now the EX1 with 4:2:2, that would be something. And I'm sure much better for keying.
Click to expand...

Don't forget that the live HD-SDI is actually real 10 bit 4:2:2 output, which is something to be considered. Not just 960x1080 true sensor-derived chroma samples, but better nuance in shades of color too.

Here's hoping that CineForm 10bit CF portable recorder makes its debut as soon as possible.
 
matthew77 said:
But the EX1 DOES do 4:2:2.

It comes out of the HD-SDI connector.
Click to expand...

I know it does 4:2:2 SDI out but I don't want to be tethered to a computer. That's why I said, 4:2:2 from the EX1 would be something. Once 4:2:2 is gleaned from other methods, I think it will really unlock this camera's potential.

I was talking about to card. As I said:

but just compare say a shot from the HVX in its own flavor of 100mbps 4:2:2, then match that with the EX1 at 35mbps at 4:2:0.
Click to expand...
And yes, it is subjective. My subjective conclusion was that the HVX image looked more filmic than the EX1.

matthew77 said:
I doubt if you saw the XDCAM from the EX1 compared to the SDI out from the same camera, you would say one looks like film and one looks like video. To the naked eye, I bet you couldn't tell the difference.
Click to expand...
If it was a well shot piece of work after the grade, I bet I could.

"Looks like video" is subjective and is something often said on this forum without backing it up, and is mostly meaningless when it comes to color. If the motion cadence is the same then you're 95% there. Film is graded in many ways and so is video. Video can be made to look like film and film can be made to look like video. Only at the extremes are the differences really apparent.
Click to expand...
You take 4:2:0 and 4:2:2 color sampled video to a professional color grader and they would take 4:2:2 any day of the week. The ease of working with 4:2:2 over 4:2:0 should not be ignored. And making something look filmic in the grade is far easier with the extra color space. I'm not talking in FCP color or suchlike, I'm talking about in high end hardware grading. For me, color space is a big thing for making filmic imagery. The further I can get away from the old DV colors the better (I'm from PAL land so we've always had 4:2:0 with DV video rather than the really awful 4:1:1 for NTSC).
 
There will be options coming up like the before mentioned Cineform portable recoder and the new XDR recorder. I will buy one of these depending on it's capabilies vs cost. The XDR is going to be around $5K.

That's a lot, but not bad when you consider how flexible it will make the EX1. Cineform's 4:2:2 HDMI is said to start less than (or near) $2K.
I imagine their later SDI model would be $2.5K to $3K
 
Trying to divine the performance characteristics of a piece of equipment, based upon advertised numbers, speculation and assumption, doesn't help anyone in terms of meaningful information.

The best way to find out how something performs is to either borrow or rent it and actually use it. Can it pull a key? Try it. Can it be graded? Try that, too.

Otherwise it's like trying to decide how well a guitar sounds by measuring the thickness of the wood, string tension, fret spacing and so on. Or judging wine just by reading about the history of the vineyard.

There are so many subtle qualities in color and tonality that can't be readily defined by mere measurements. The only way to truly know about a camera's performance is to actually play with one.
 
I sometimes record laser light shows that I produce here. I've been recording laser light for 21 years--starting out with 1" Newvicon tube cameras, evolving to 1/3" CCD cameras in 2001, and tried again with HDV in 2007 with the HVR-V1U. It looked better than standard def, but not as good as I'd expected.
Then last night I recorded a sequence with the EX1. I figured it'd be a tiny bit better than the V1U, but when I watched the playback on my HP LP3065, I was shocked--the image has fine detail--I could even see the "corpuscular" interference patterns in the reconverged laser light on the screen, which no camera has ever been able to pick up in the past due to chroma smear!

For those not familar, red laser Helium-Neon is 6328 Angstroms, an intense cherry red. Video cameras crumble in the presense of this light. All images go to smeary, blurry-looking blobs of light. Not the XDCam EX, however. It's the closest rendition to what my own eys see, of any camera I've used in 21 years of making laser videos.

Not all 4:2:0 looks the same.
 
DCSensui said:
Otherwise it's like trying to decide how well a guitar sounds by measuring the thickness of the wood, string tension, fret spacing and so on. Or judging wine just by reading about the history of the vineyard.
Click to expand...

I think you may be a bit overdramatic here. Digital video, by definition, is quantifiable.

It's certainly easier to quantify than a guitar or a vineyard. The numbers actually mean something.
 
matthew77 said:
I think you may be a bit overdramatic here. Digital video, by definition, is quantifiable.

It's certainly easier to quantify than a guitar or a vineyard. The numbers actually mean something.
Click to expand...


Actually, that's how physical modeling works with some modern MIDI musical synthesizers--they model the shape, length and diameter of, say, a pipe, and calculate the sound it will produce when air is moved through it. All done with computer simulation, but generates a realistic sound. Sound is completely quantifiable. :)
 
Just to clarify, I was referring to cameras and not recorders. There's been so much speculation about what the EX1 can and can't do that I just had to jump in.

A/D converters vary widely in quality. CMOS chips or CCD chips aren't all alike. Some are better than others.

Signal processing accounts for part of that difference as does the characteristics of the imaging chip. It's a system and every step along the way contributes to overall performance. So the system has to be judged based on what it actually does, and not just the underlying technology.

While you can say something can reproduce a given range of light, how that light is translated can be significantly different from one system to the next. Are subtleties reproduced smoothly or are there undesirable transitions from one color to the next? Is it noisy in the shadows? Is there a color cast in the highlights? Does it roll off toward the toe and shoulder of the response curve or does it just suddenly clip or crush values outside the operating range? Color is a three-dimensional space. What color space does that device employ?

The real world can quickly part ways with the theoretical world. You can have two different film emulsions, for example, and both can reproduce 10 stops of brightness. Both have the same ISO rating. But what's the quality of that image? What does the skin tone look like relative to other objects in the scene? Is there a color cast? Do certain colors saturate faster than others? The basic numbers might be identical but visual qualities might vary widely. National Geographic used to do extensive tests on batches of emulsions for a single film type from Kodak and buy out a specific batch. Same film. Same basic emulsion. But the color quality was different from one batch to the next despite strict quality controls.

Again, numbers alone aren't enough. It took trained human eyes to pass judgment.

Claiming one camera or the other can or can't possibly do a specific task, based entirely on specifications, is a little better than a guess based on assumptions.

And even if the camera exhibits certain problems the practical questions are: who will notice and does it matter? And is it worth spending a thousands of dollars more to avoid those problems? In some situations these might be insurmountable hurdles. In other situations it might not matter. Depends on the application.

The only way to really know is to try it. That's what some people here have been doing. CMOS rolling shutter issues? Try it out. See how bad it might be. How does the camera feel? Pick one up. Try it out for a day or two. Is long GOP bad for highly detailed scenes with rapid movements? Maybe. But if it's not tested no one will ever know for sure. Will it key nicely? Who knows? Maybe it might provide a great key but only with the right software.

What I'm trying to say is that when it comes to picking out a camera, determine performance factors by doing real-world tests. Numbers might be able to provide a starting point, but it shouldn't be the ultimate deciding factor.

While this is rather preachy it's intended to encourage others to approach new technology with an open mind. Set aside some of the usual presumptions and you might be pleasantly surprised. I was.
 
You must log in or register to reply here.
Back
Top