I've always liked Adam Wilt's site for his explanations and pictures of this issue.
For example:
4:2:2, 4:1:1, 4:2:0;
DV Pix - Sampling Methods (more links at the top right of this page);
DV Pix - Multigeneration test images.
Also, he's written several articles on this and similar issues for DV magazine. You'll need to register but it's free.

Mike

>>HDV is 4:2:0. Does this mean the 0 represents zero Blue component (the V)? <<

No... the 4:2:0 doesn't imply that there's no blue component, of course... there has to be.

4:2:0 means that there isn't a color sample on every scanline (like there is in 4:1:1, 4:2:2, and 4:4:4). Instead, 4:2:0 delivers a color sample every other scanline.

Put in numerical grid terms: if you're looking at a 720x480 DV image, you get effective chroma resolution of:
4:4:4 = 720x480
4:2:2 = 360x480
4:1:1 = 180x480
4:2:0 = 360x240

Luma resolution is black & white (well, shades of gray actually). Chroma resolution is what color those pixels are -- the luma is how light or dark they are, chroma is what color they all are.

In 4:2:0 (and 4:1:1), there's one color sample for a block of four pixels. That means that all four of those pixels are the same color, but they're different degrees of bright/dark (depending on what the luminance pixel shows). If you're shooting a strictly black & white image, chroma resolution is irrelevant. If you're shooting a color image, chroma resolution is quite relevant.

In 4:2:0, each chroma sample represents a 2x2 block of pixels. If within that block the upper-left pixel is green, the upper-right is blue, the lower-left is red, and the lower-right is gray, the 4:2:0 chroma compression would have to try to reconcile all four pixels and come up with one color that best describes all four. In that case, all four pixels would be turned to gray (the best compromise between equal parts red, green, blue and gray). If they were different tones, then the luminance value would still be different, so you could distinguish four different pixels, but all pixels would be gray. If they were all the same tone (i.e., 128,0,0; 0,128,0; 0,0,128; and 128,128,128) then in a 4:2:0 system all you'd see would be one big 2x2 block of gray; the luminance and chrominance would all be recorded at the same level.

In a 4:2:2 system, the green & blue on one line would be color-averaged together, and the red & gray on the next line would be averaged together, so you'd get two different color samples.

Of course, in 4:4:4 you'd get exactly what you'd expect from the original: individual red, green, blue and gray pixels.

The reason the 4:X:X notation is confusing is because it is illogical. Unintuitive I believe Adam Wilt calls it. If you try to read meaning into those three numbers you will go mad if you are not lucky enough to realize it just does not stack up.

Instead think of it this way. Forget the number 4, but know that luminance (Y) is always sampled 100%. Now consider the varying percentages of sampling for the color in the horizontal and vertical.

The Notation........% Horizontal sampled.........% Vertical sampled

4:4:4..............................100%............................................100%
4:2:2................................50%............................................100%
4:1:1................................25%..............................................50%
4:2:0................................50%..............................................50%

Mike S

Could this explain why HDV and mpeg-2 for DVD uses 4:2:0, from those figures given the wider AR of 16:9 used in both those formats the 4:2:0 sampling holds up better than 4:1:1?
Another question though, I've seen sampling quoted with an extra digit, does that refer to the alpha channel?
Thankfully one day soon it'll all be RGB and this YUV fudge will be behind us.
Bob.

The fourth digit, when present, refers to the alpha channel.

To add confusion, the 4:2:0 of PAL is not the same 4:2:0 as that of MPEG-2...

Holding up better? You mean that HDV MPEG-2 4:2:0 holds up better when converted to an SD DVD which is also 4:2:0?

Going 4:1:1 to 4:2:0 you lose even more color information, this is certainly true.

Long live 4:2:2!

The YUV fudge was probably inspired a long time ago by b&w compatibility, then found to reduce bandwidth needs.

Mike S, that chart is quite good, but there needs to be one correction: it doesn't accurately represent 4:1:1. 4:1:1 is sampled 100% on the vertical.

And yes, as Coursedesign says, there are two different 4:2:0 systems. So 4:2:0 doesn't necessarily mean 4:2:0...

dang coursedesign, if you didn't say that, i'd be content right now. 'not all 4.2.0's are alike...' lol

isn't there some drug reference to 4:20? i'm just asking... i wouldn't know about this stuff.... hmmm, got the munchies....

Barry, you've helped out even more. Thanks guys for your 4.2.2 'input'... now from what i've learned, will my 'output' be 4.1.1 or 4.2.0?

. . luv it! Yah gotta luv it! - I AM listening to this, very closely .. G

That depends. If you render to NTSC DV then it's 4:1:1, if PAL DV then 4:2:0, if you use the Sony YUV codec then 4:2:2 which is a better way to go.
Another consideration though is no matter which one you choose it's only 8 bit.
Bob.

Another consideration though is no matter which one you choose it's only 8 bit.

I would take 8-bit 4:2:2 over 10-bit 4:1:1 anytime though.

The color sampling makes a much more fundamental difference than 8-bit vs. 10-bit (although I do like 10-bit for the improved tonal range and added flexibility in post).

If Sony wants to keep Vegas as a leading edge and pro package version 7 MUST be 16 bit (in my book).

If Sony wants to keep Vegas as a leading edge and pro package version 7 MUST be 16 bit (in my book).

K, you lost me here. Who has a 16bit video editor in the sub 50k price bracket?

Who has a 12bit in a sub 50K price bracket?

I think MH is thinking of After Effects Pro, this handles 16-bit color (although it is not an editor, but a compositing and motion graphics package).

Perhaps because of Vegas' great strengths in compositing it is tempting to ask for the same capability.

AE Pro uses a lot of extra disk space (and memory) dealing with 16-bit color. This is still useful if you are dealing with chained effects that can create banding if you don't go to a higher color space (note that some other packages go to 32-bit and even float precision).

For editing, it would be nice to support 10-bit video, stored as 10-bit for efficiency reasons. This will come, it's only a question of which year.

10 bit, I can agree with wholeheartedly. 10 bit holds a bit of benefit when working with 8-bit originated video as well, but not enough to justify the extra cost, IMO. But you're right, it will come.