4:2:2 - 4:4:4 and conversions
 

I need some help in understanding better the concept of 4:2:2 and 4:4:4 color ratios and converting these ratios. Is there someone out there who is experienced and can lay it out for me without getting too wordy? Or perhaps direct me to a thread that can explain it for me, at least in a better way than the material I've reading. It is much appreciated, thanks!

4:4:4 colour space has one pixel of discrete colour information per colour channel for every pixel of luminance information. ie 1920 x 1080 has 1080 rows of 1920 pixels of luminance, and 1920 pixels of each of the colour difference channels R-Y and B-Y.

4:2:2 has one pixel of discrete colour information for every TWO pixels of luminance information. ie: 1920 x 1080 has 1080 rows of 1920 pixels of luminance and 960 pixels of each of the colour difference channels.

4:2:0 has half the horizontal colour resolution of 4:2:2. For 1920 x 1080 there are 1080 rows of 1920 pixels in the luminance but only 540 rows of 960 pixels for colour signals.

If you think of the way color is stored on a camera recording system, the conventional approach is to convert the RGB signal to YUV, where Y is luma, and U and V are difference signals R-Y and B-Y.

So, in 4:4:4 space, there are 4 bits of info in each Y,U,and V channel.
In 4:2:2 space, there are 2 bits of U and V info for every 4 bits of Y or luma info.
4:2:0 is really shorthand for 4:1/2:1/2...sort of anyway. In the 4:2:0 scheme, the U and V channels are sampled twice as often in the horizontal direction than the vertical. The resulting resolution in the chroma regime is, therefore, quarter rez in the vertical for every 1/2 rez in the horizontal.

Actually, colour space doesn't take into consideration bit depth. You can have 8 bit or 10 bit depth in 4:4:4 or 4:2:2.

The first 4 comes from the analog days when luminance was modulated at a frequency 4 times (13.5 MHz) that of the colour subcarrier (3.4MHz).

I think you meant to say sampled, rather than modulated?

It's actually a bit more complicated than that, as in the early days of digital video it was indeed the composite signal that was sampled. It was at 4x colour subcarrier frequency for NTSC (hence the "4"), and at 3x for PAL, so sampling frequencies of 14.32MHz for NTSC and 13.29MHz for PAL.

Come the move to digital component and a decision to standardise on a common sampling frequency for both systems, as there is no longer any need for it to be locked to subcarrier, and hence 13.5MHz for both NTSC and PAL. The "4" notation then becomes nominal, it's no longer exactly 4x sc frequency.

A common query is "why 4:2:2"? Why halve the chroma resolution vertically but not horizontally? In the early days of digital component, television meant interlace, an asymmetrical system, and it made a lot of sense to retain full chroma res vertically due to the interlace structure. Move to progressive, and the advantages of 4:2:2 over 4:2:0 are lessened.

What's important to realise is that it only defines the sampling and recording format, and says nothing about the origination of the signal. That's important because many camera sensors mimic the performance of the human eye by generating better luminance detail than chrominance - they inherently can't be 4:4:4. Examples are Bayer sensors, and ones using pixel shift techniques.

Also frequently overlooked is that colour space numbers are RATIOS, not absolutes. Full raster 1920x1080 4:2:0 is obviously going to be a lot better than NTSC 4:2:2. That's pretty obvious, but the same principles apply with sub-sampled HD formats like DVCProHD (1280x1080) and HDV (1440x1080).

Thank you David for expounding and correcting.

At the camera end it certainly would be sampling, for some reason I was already onto transmission so modulating made sense at the time. No more technical posts after a long hard day!

Nice job Bill very easy to understand that way.

Thanx, Paul.
Disclaimer: sorry for using the word "bits" in my post. Of course, I didn't mean to refer to digital precision, I was ambiguous. Perhaps "units" is a better word than "bits".

And sorry for "calling" Bill on the use of the word "bits".

4:2:2 question
 

I understand the 4:2:2 will halve the horizontal resolution. So if the original 4:4:4 image pixels are 2 rows of 4 pixels:

YCbCr YCbCr YCbCr YCbCr
YCbCr YCbCr YCbCr YCbCr

--> 4:2:2 will result in following:

YCbCr Y YCbCr Y
YCbCr Y YCbCr Y

But I am not sure what the wikipedia page of 4:2:2 means, particularly the bottom checkerbox.

Image:Chroma subsampling ratios.svg - Wikimedia Commons

Doesn't why 4:2:2 halve the chroma resolution horizontally?

Chroma subsampling - Wikipedia, the free encyclopedia
The two chroma components are sampled at half the sample rate of luma: the horizontal chroma resolution is halved.

You are absolutely correct, my apologies, it was a typo. I don't seem to have the ability to correct the error in the original post?

It is actually more accurate to say that a 4:2:2 system records as many chroma samples as luminance vertically, but only half as many in the horizontal direction. The nomenclature only really refers to recording systems, not overall performance.

To talk of a 4:2:0 system "halving chroma resolution vertically and horizontally" is only true if you are starting from a system with equal luminance/chrominance properties in the first place, and equal or better than the basic recording resolution.

the bottom checker box is the combination of the top 2 boxes. the top box is luminance, and the middle box is chroma, and the bottom box is the final color output when the luminance is applied to the chroma.