4.2.2. versus 4.2.0
 

My situation is unique and maybe someone can enlighten me a bit. I come from a background as a 35mm film shooter, without having edited or managed the end use/workflow. I'm moving to HD and have learned a ton of new technical information already. But what is the difference between 4.2.2 and 4.2.0? I understand the soon to be released Sony XDCAM EX Cinealta will deliver 4.2.2 under a certain workflow.

I chiefly produce very high end stock footage, destined for my agent, Getty Images, who distributes and markets my material worldwide. It needs to be the highest quality possible to be intercut into movies, commercials, and documentaries.

Chris.. I see you're in San Marcos. I'm in Austin. Any chance for a beer/coffee to talk tech sometime?

4:2:2 has half the colour resolution of the luma for any given resolution, 4:2:0 (and 4:1:1) have only quarter the colour (Chroma) resolution.

For most footage, 4:2:0 looks pretty good to the human eye, however if you need to do heavy manipulation, like blue screen etc, 4:2:0 can produce visible jagged edges around coloured objects. It is also noticeable with high contrast coloured areas such as coloured graphics and titles.

For high end stock footage, clients may demand more than 4:2:0 but it isn't always possible to drag a 35mm camera and rolls of film everywhere.

The EX outputs 4:2:2 over HD SDI which would require a separate recorder, such as a computer or high end HD VTR, however I'm not sure if this 4:2:2 output is "real" 4:2:2 or just upscaled 4:2:0 - can someone confirm?

All footage recorded on board will be 4:2:0 (even if bumped up to 4:2:2 and played back through the HD SDI) because that is limited by the camera's recording codec.

Initial reports suggest the EX picture quality may be unrivaled for it's size but if you need the best quality possible, you may need to spend a bit more.

Chroma resolution relative to luminance. With 4:2:0, chroma is half that of luminance both horizontally and vertically. With 4:2:2, it's half horizontally, but the same vertically.

It's much more significant with interlaced video than progressive, as the alternation of samples then takes place on a field basis, so in frame terms you get two adjacent chroma samples, then two gaps.

There's a lot more to it than that, especially when motion gets involved, but the advantages of 4:2:2 over 4:2:0 are far more pronounced with interlace than progressive.

http://www.dvinfo.net/conf/attachmen...0&d=1159424266

http://en.wikipedia.org/wiki/Chroma_subsampling

It might be easier to see what is going on other then reading about it.

It's worth noting here that the EX1 uses 4:2:0 color across a full 1920 x 1080 image (in HQ mode), which works out to 960 x 540 chroma samples per frame. That's a decent amount of color information compared to any other HD camera under $10,000.

The absolute number of color samples isn't a good metric.

You need a ratio of color samples to luma samples, which is why we use ratios to describe color in video... i.e. 4:2:2 or 4:2:0.

This won't often matter to the viewer, but it matters a lot if the footage is to be used for VFX work or extensive post.

For a client like Getty I'd be concerned about 4:2:2. After all if you've been turning in 35mm, then Getty has been getting 4:4:4 images from you in video terms. Check with their engineering staff though.

I wrote about this before here:

http://dvinfo.net/conf/showpost.php?...3&postcount=35

The HD SDI output direct from the camera head is 10 bit 4:2:2 uncompressed. (SMPTE 292M)

It is the XDCAM codec which discards color information, not the camera head.

So, if you play back media from the SxS cards, you get upscaled 4:2:0 video.

As far as field capture of 4:2:2 footage goes, one worthy workflow is HD-SDI to the AJA ioHD, and then Apple ProRes 4:2:2 direct to disk. This is "do-able" with a Macbook Pro and an external drive setup over eSATA. That's three more gadgets on set.

So, you are trading a somewhat larger film cam and mag for a small camera that has to be tethered to a video village. It is hard to choose.

If you can justify another $5K USD and want the best out of your EX1, go with Convergent Design's upcoming XDR Flash

If the EX1's camera head is as good as we believe, just imagine shooting full raster to 422 @160Mbps I-Frame.

Man, I really want one of these!
Well, actually I want the camera first. LOL

http://www.convergent-design.com/dow...%20XDR.pdf

I am not entirely convinced on that unit. It seems promising, and I'd like to be convinced. I am extremely excited by the prospects.

My first concern is that I haven't heard of Convergent Design, which may be my own fault. I have to be convinced about them as an engineering facility. Do they make solid products? For that I can turn to the community- what has been the experience of people here with Convergent Designs products?

By comparison AJA is a known quantity. I know their engineering, I've used their products, I've seen dozens of my peers using their stuff. I even see Apple talking about and working with them.

I need to see more information on workflows for their capture format on Final Cut and Avid. Can it use the built in Logging and Transfer option in FCP? Do we need additional software? I'd love to see a whitepaper talking about the capture format in detail.

Hopefully there will be more answers as the promised ship date draws near.

That said the Convergent Designs product is a bit high, but in the right ballpark price wise, and it provides some innovation and potentially great utility.

I'm eager to get the opportunity to test a Flash XDR and to hear from those of you that do get to test one.

Thanks Thomas - good link.

They do mention 4:1:0 : "This ratio is possible (indeed, some codecs do support it), but not widely used. It means half the vertical and quarter the horizontal color resolutions......"

Whilst it may not be chosen by choice, it can effectively become the colour space if material is originated in a 4:1:1 system and then coded for release in 4:2:0 - which is exactly what happens if material is shot NTSC DV25 (4:1:1) and then used to make an NTSC DVD (4:2:0). I understand it is one very valid reason why 4:2:2 codecs like DVCPro50 have more takeup in the US than in 50Hz countries.

In PAL the problem only comes up with DVCPro - both DVCAM and DV PAL are 4:2:0 - and may be one reason why DVCAM is far more popular than DVCPro over here.

AFAIK 4:1:1 is not used for any HD codec, so the problem shouldn't arise for HD.

Yeah 4:1:1 was a pretty bad color format. I hate the fact that NTSC users have to make DVD's with such bad decimated color. This is one of those things I try to explain to people why homemade DVD's don't look as good as Hollywood DVD's. NTSC 4:1:1 to DVD is really murdering the color. This is one of the main reasons why HDV down converted in software to DVD looks so darn good. Sure it has over sampled resolution to help but it also has over sampled color which helps a lot and the color starts as a native 4:2:0. In a world where every single other video standard works in 4:2:0 or 4:2:2 I will be glad when DV is gone and 4:1:1 right along with it. I don't think we will ever see 4:1:1 ever again. Now if we could only get rid of interlaced video forever along with 4:2:0 field based color spaces and just stick with the more natural looking progressive flavor of 4:2:0.

Yes - 4:2:2 and 4:1:1 really are engineered to be most relevant for interlaced systems. (And in NTSC, I believe 4:1:1 was chosen over 4:2:0 for DV25 more because digital devices were then seen as islands in an analogue world - all i/p, o/p being analogue.) In the progressive world, 4:2:0 and 4:4:4 are much more suitable.

The main point I wanted to bring out was that 4:2:2 is largely (and justifiably) seen as so desirable in the NTSC world is not because it *is* 4:2:2 - but because it avoids an eventual effective 4:1:0. Keep 4:2:0 throughout - as with PAL DVCAM/DV - and there is less advantage to a 4:2:2 system for origination, even more so with progressive systems.

I may be missing the mark here.

Can't the EX1 'record' 4:2:2 on the SxS cards? OR does one NEED to 'record' into a computer using the HD/SDI ports via external hardware (say Blackmagic).

I thought it was possible to record in a 4:2:2 colourspace in the field using the SxS cards and then dumping the media from the cards into a computer?

Uh... no.

4:2:2 matters a bit to poor folks that need to actually work with footage before we output it.

Clearly there isn't anything I can say to convince you, so take a look for yourself.

The image I attached is a key from the jpeg Thom linked to here:
http://www.dvinfo.net/conf/attachmen...0&d=1159424266

I took that image and did a quick key with Primatte RT. You'll notice some desaturation... ignore that. I upped matte density too much. Just notice the edges.

The result file is a PNG, and it has transparency enabled so drag it over whatever background you like when you get it on your computer.

Clearly the 4:2:2 image shows a marked improvement in key quality when compared to the 4:2:0 progressive image.

Consider that these are computer generated files- the easiest to composite.

A real image of say a person with this stuff called hair might present additional challenges.

Sometimes more data is good data.

SxS records only 4:2:0 Long GOP XDCAM.

The EX1 requires some sort of outboard system for 4:2:2 recording.

AJA's ioHD with a laptop and Convergent Design's Flash XDR seem to be the most popular outboard systems proposed right now.

While I am excited about the EX1, not including 4:2:2 recording could become a serious oversight with the indie crowd this camera is aimed at.

How do you record from AJA ioHD to a laptop - through the FireWire? If so, is the 400 mbps version enough?

No - I am convinced, and maybe the "largely" in what you quoted from me is being overlooked, I'm certainly not saying that 4:2:0 is every bit as good as 4:2:2, period. Just saying the subject has more to it than often made out, especially when referring to straightforward footage with no keying etc.

One thing your image does reinforce (and which is often overlooked) is the difference between progressive and interlace 4:2:0 - 4:2:2 is unquestionably the best of the three, but the difference is less marked between it and the prog 4:2:0 sample than it and the interlace sample.
Undoubtably. So why draw the line at 4:2:2? Why not insist on 4:4:4?

I suspect that if the enlargement of the image had been done on a nearly vertical edge just above the three o'clock position we'd have a very different result, and that whilst the 4:4:4 result would be very similar, there would be much less difference between 4:2:2 and progressive 4:2:0.

Thanks for clearing that one up. Looks like I have to recalculate some things relating to this camera. A big oversight to not have 4:2:2 recording onto the cards.

I am looking at a camera to use for indy filmmaking and I would definetely need to use with greenscreen. The EX looks pretty nice on paper for everything except not getting 4:2:2. If I can't get the 4:2:2 coloursampling recordiong onto the EX1's cards.. what is the cheapest method of getting the 4:2:2 into my computer via the HD-SDI? Thinking of price point here.

In this case should I consider the HVX over the EX?

By the way I will be stepping up from DV. So i feel a little bit in the woods when it comes to HD. I have used many (A1, HVX200, fx1, z1) and really see the difference in quality and resolution. But I haven't yet had the opportunity to use an HD camera for greenscreen work.

Makes sense, but if you had to choose between 4:2:0 with full 1920x1080 image resolution or 4:2:2 with 1280x1080 resolution, which do you think would produce the better key?

4:2:2 1280 definitely.

Now, if I determine the the image is "unkeyable" then I prefer to have the highest resolution image I can for rotoscope and tracking mattes.

Well Ben as a fellow indie I am buying into the EX1. I like the HVX, but I really want the 1/2" sensors, focus marks and full raster 1080p image.

The cheapest way to get 4:2:2 into a PCI Express desktop is BlackMagic's Decklink HD Extreme. $995 MSRP That unit is single link HD SDI, but it has analog and digital i/o. (The HD Pro models are SDI only.)

http://www.blackmagic-design.com/products/hd/

The best way to get 4:2:2 into a laptop right now is AJA's ioHD. $3495 MSRP.

http://www.aja.com/html/products_Io_IoHD.html

You output via HD SDI to the ioHD, and then it pipes into your Mac via firewire. It works over Firewire 400Mbps, but the ioHD comes with a 800Mbps connector on its casing.

That footage is in Apple's ProRes 4:2:2 codec in 10 bit color.

Alternatively there is the Convergent Design Flash XDR. That is MSRP $4995. It uses compact flash drives and records a 4:2:2 intraframe codec at 160Mbps. It also offers 50Mbps and 100Mbps 4:2:2 with Long GOP. That might come out a bit cheaper if you have to buy a Mac and external drives for the purpose.

http://www.convergent-design.com/dow...%20XDR.pdf

There is also V3HD, which captures to DVCPRO HD. It retails for $2750, making it the cheapest solution possible for a laptop.

DVCPRO HD limits you to 1080i or 720p capture. That may be better for keying than the SxS 4:2:0 footage, but the camera does so much more, I'd rather spend the extra cash and get the full performance.

http://www.v3hd.com

In any case I'd recommend an external SATA RAID 0 disk array for capture. It should work though on even a single USB connected drive. It should work on some of the 7200RPM laptop internal drives, but I've never seen that tested.

The main reason these solutions are rejected is that its a lot of cabling and hassle. For film style work I see that as a non-issue because my cameras are always tethered to video village for video monitoring (yes even on dolly and steadicam shots!) if nothing else.

With these solutions I get 4:2:2 video, and I can use the Mac as a waveform|vectorscope. In a pinch I could use the Mac for monitoring video too.

As a note I think I am going to go the AJA ioHD route.

Maybe I misread that. Sorry.

Only one reason: $$budget$$.

4:4:4 means HDCAM SR or uncompressed data on the drives.

Uncompressed 10 bit 4:4:4 is only available with RED and higher priced cameras. The only ones "readily" available are the Thomson Viper and Sony's F950 & F23.

Also we are talking dual link HD SDI, at about 3Gbps. Thats about a Terabyte every five minutes.

Thats a lot of disk space and it has to be FAST. Four disk RAID 0 at minimum. A 14 drive RAID 0 SATA array can give 12TB or about 60 minutes of recording. (That's real binary TB not the popular trillions (10^12) of bytes) That's two fully loaded Xserve RAID's.

HDCAM SR recorders can alleviate the disk space issues- at the meager cost of $95,000 USD. (Base configuration)

Eventually- but not yet. When I do start working with 4:4:4 the EX1 will be B cam or smaller roles.

Well they become more similar, but not the same, 4:2:2 holds the advantage.

By the way, 4:2:0 with a circular subject like this looks almost exactly like a 90 degree rotation of 4:1:1

Backing up for a moment here, what were the cameras and recording modes used for the circular images example? Not all video is created equal for any given chroma format, so one example of each format isn't informative without more background detail. Was the 4:2:0 example from the EX1, or some other camera?

I'm not convinced that having 960x540 square chroma samples spaced evenly across an HD image (e.g. from the EX1) is dramatically different from having 640x1080 non-square samples (e.g. from the HVX200).

Interlaced versus progressive is another matter, which raises the question whether the EX1 has a decent progressive recording mode?

Great information Alexander!!

The BlackMagic looks like the cheapest but the AJA solution offers allot more -but- obviously more $. Capturing into ProRes is another bonus.

The 1/2 inch sensors, lens and 1080p are what's attracting me to this camera. As many others as well! Alot to think about and a budget to be made.

thanx,

Ben

Mightn't the answer depend on a lot of other factors, most importantly whether they are both interlace or both progressive?
I think any confusion might be that I am speaking theoretically, you are speaking here and now, so yes, for present practical purposes I agree.

But for the future, and the ideal progressive only world we all long for, 4:1:1 and 4:2:2 are colour spaces conceived for the interlace world. Their non symmetry matched the non symmetry of interlace. Progressive images are symmetrical, and much better suited in theory by 4:4:4 or 4:2:0.

Kevin I created that image in 3D Studio Max. I rendered with a softer form of anti aliasing to somewhat simulate low pass filtering in cameras. I rendered out a 4:4:4 RGB sample and then used After Effects to convert to other decimated formats. I then took still images of each format and put together the image you see.

So it is computer generated but really that means it has a lot more raw detail then what most cameras can do. Real world fotoage also tends to be softer and the details are blended together a little bit more. This means even if you have 4:2:0 the chroma samples are a little bit more blended. If you usually shoot VFX stuff softer without any electronic edge sharpening this helps even more. In this case the edges are so naturally soft that the difference between 4:2:2 and progressive 4:2:0 becomes pretty small. Video is never 1 pixel accurate like a computer image is.



Alexander, Sure the straight keying looks better with 4:2:2 but your example doesn't show any chroma upsampling or softening. The key to good keys is upsampling the chroma to try to get back to a 4:4:4 format. Sure it isn't perfect but it does help a lot. I use Shake and I always soften the two chroma channels and relink the three channels back together before I do a key. In this case the super small edge you gain with 4:2:2 gets even smaller to the point where it gets really hard to tell on moving footage without inspecting a still image up close on a computer.

There are even ways of fixing interlaced 4:2:0 by reshfting just the chroma samples and then upsampling them. This still isn't as good as progressive 4:2:0 but it does help a lot. This is not something that can be done in Shake however and special plugins or processing tools have to be used. Nattress makes great plugins for FCP that help upsample the color.

Sure progressive 4:2:0 isn't perfect but it is not bad either. You can tell from the keyed image that other then slight pixelized edges it looks very clean. If you are downsampling to SD for your final project it will look almost perfect. Even if you do not it is going to be very hard sitting back on a ocuch to notice these single pixel sized edges. If you add chroma softening you get rid of the problem. Yes 4:2:2 is better but that is a no brainer. The question is by how much and is it really worth it to everybody. To me and you it is worth it but not to everybody else. In fact I see some people pulling off some pretty amazing stuff with progressive 4:2:0. With the proper processing it is going to look pretty darn clean.

It should also be pointed out that HDCAM tape uses 3:1:1 color and only samples 1440x1080 pixels. This give you chroma channels that are 480x1080 pixels in size and yet a lot of people pull great keys from this tape format and the tape format is regarded as the industry standard. I have worked with keying HDCAM footage and while it wasn't perfect I got some very good results from it. I also studied at the Academy of Art college under Jay Cooper who was one of the leading compositors for Star Wars Episode 2. He supplied us with some great ILM material to work with and gave us a great rundown on how much they hated working with HDCAM. ILM upsampled chroma on the footage however and ended up with some great results. A lot of it they also captured tied to a capture station so that helped a lot.

So if 3:1:1 was good enough for ILM and Episode 2 I'm sure the rest of us can make great use of other decimated chroma formats.

Alexander,

Quoted from Alexander Ibrahim
"Alternatively there is a Cineform Flash XDR. That is MSRP $5995."

I'm familiar with the Convergent Design product, but didn't know it was Cineform based.

Where did you hear that it was Cineform based? That would be very exciting for me anyway.

I once belived the same thing. I assumed that PAL's 4:2:0 would give an advantage over NTSC as far as DVD conversion would go, because of mpeg2 DVD's 4:2:0, but it was explained to me that is not the case. I am not an expert on this but I was told that PAL's 4:2:0 does not equate to DVD's 4:2:0. Therefor PAL is of no advantage.

I didn't that would be a brain fart. Or wishful thinking if you like. In either case it isn't true.

Please read that as, "Alternatively there is the Convergent Design Flash XDR. That is MSRP $5995."

I'm gonna edit the original too.

Rats! I was pretty excited.

I think their product would be a good complement to the XDcam EX though, and much more portable than hauling a PC around to shoots.

Just a quick note of clarification, the Flash XDR is $4995. Also the Flash XDR offers a much wider range of bitrates (19.7 to 160 Mbps), so you can dial in the rate depending on the job. You can also select between 4:2:0 and 4:2:2 sampling as well as Long-GOP or I-Frame Only. Flash XDR is small enough to snap onto the back of your camcorder for run and gun applications.

Mike Schell
Convergent Design

Hi James-
Flash XDR is MPEG2 based, either Long-GOP or I-Frame only in 4:2:0 or 4:2:2 recording at up to 160 Mbps. Also the price is $4995.

Mike Schell
Convergent Design

Thanks Thomas, I was about to say the same thing. Except I was going to say any camera.

Kevin, you never get a better key than with CG images.

Also, this image has relatively mild JPEG compression.

XDCAM 4:2:0 shoves a 1.4 Gbps stream into 35Mbps of data. That's about 41:1 compression.

Compression is complicated by noise levels. (One place the EX1 should outshine a lot of its competitors due to its 1/2" sensors.) Noise makes compression more difficult and can break a codec rather thoroughly.

Also when considering camera systems, you have to consider the bit depth of your codec. 8 bit or 10 bit or 14 bit. I don't think there are cameras that use float or half, then again 14 bit log is pretty impressive. Most of the formats we use, HDV, DVCPRO, XDCAM are all 8 bit formats.

That makes a 10 bit 4:2:2 HD SDI capture even more appealing.

I wasn't trying to get a great key.

I think you are missing the point. The point isn't to demonstrate advanced keying techniques, but rather to show the difference in the chroma sampling.

In this example I'd normally use Keylight in Shake. That pulls a cleaner key with default settings. I am not a fan of "magic bullet" keying- but in this case Keylight works a wonder.

I picked Primatte, because its included in Motion, which more people have.

In fact, thinking of Keylight and Shake, I think the alpha channel Keylight creates is instructive. I'll attach it here. (its actually just a screen grab of the Shake interface, about 1000x1000)

Its always worth it to see if one of the built in keyers can do an acceptable job, because they will always be faster than a tree/script you devise.

Back to my point.

This Alpha channel represents the default setting of Keylight. Right off the bat I can use the 4:4:4 and 4:2:2 keys... I can even just use the keylight node to lay in the background.

The 4:2:0 keys need a lot more work. I don't work the same way you do, but think of the time needed to key using the technique you describe.

Saving that artist time, and quite a lot more in many real effects situations, is the real advantage of 4:2:2 and 4:4:4.

Thanks for the info Mike, I corrected my original post. Glad to see you are reading here.

Is there a whitepaper on the codecs used in the Flash XDR?

Well, if I am going to dream about the future... how about mathematically lossless compression of 8K at 16 bits per channel at every pixel in log space? I'd ask for a 16:9 sensor, but I'd like the camera to offer a 2.39:1 cine aspect ratio, where the sensor could increase its frame rate in turn for using lower vertical resolution. Oh, and solid state or holographic recording please with faster than real time transfer to my NLE, which should of course be a mid range Mac laptop.

Actually I expect all of that will eventually happen.

Well except for the part about Apple having a mid range laptop. That's just madness.

I also wasn't talking about keying ability but the fact that progressive 4:2:0 is not garbage. In it's raw form 4:2:2 is better but with filtering the results are going to be much closer.

I don't know why you say my method takes a lot of time. It doesn't really take that much at all. A filter in FCP takes a second to drop in. In shake it just takes a few quick nodes which you can then save to reuse for the future. There are other tools as well that will process the footage right in your software so it doesn't have to take any time at all. To some people using such tools can mean getting great quality keys out of 4:2:0 material. Hey I agree with you that 4:2:2 is better and that I try to allows convince my clients to go 4:2:2 all the way but I just don't agree with you that 4:2:0 is garbage. Sure it takes some work but some darn good keys can come from progressive 4:2:0 material. I like where you come from but I cannot agree with you that 4:2:0 is garbage. The key you did of my image does of course show 4:2:2 as better but not by all that much. I have done some very high quality keys with good 4:2:0 footage and many other people do as well.

I don't think that is true. I also think that by "PAL's 4:2:0" you are referring to "PAL DV 4:2:0", and whilst it is true that there it is defined differently to the MPEG definition, I seem to remember the difference is fundamentally a half line shift in chroma.

Hence, in PAL, DV-MPEG (and therefore DVD) recoding should only result in a slight chroma shift, whilst in NTSC it results in throwing away half the chroma information. Dv25 to DVD should therefore be much better in PAL than NTSC, with the exception of DVCPro, which is the only example (AFAIK) of 4:1:1 in a PAL system.

Ok here is my chroma example image but this time I upsampled the 4:2:2 and progressive 4:2:0 sections. I didn't do the interlaced 4:2:0 section because I would have had to use some other tools to do it and I was too lazy to do it. Interlaced 4:2:0 really is a pain in the rear.


Edit: You will notice all jagged edges are now gone. I didn't fix the blown up cross section of course just the main image.

The only artifact you now have is a slightly thicker dark border around the subject which is due to the smoothed chroma samples. You now have some of the color bleeding a little bit.

Hmmm. I think most of that really is dreaming....... :-)

Whereas 4:4:4 seems perfectly feasible within proven technology - just 50% more bits than 4:2:2, keeping all else equal.

Desirable though much of your wishlist may be, I'd expect a general move to wavelet compression (as used by Red and JPEG2000) to be the next big step.