4:4:4 color sampling versus 4:2:2

[Luba C]

In 4:4:4 nothing is thrown away so that's the best, but how is it compared to 4:2:2 in real world terms? If we could quantify it, are we losing 15% color information in 4:2:2, and would we humans notice?

With 4:2:0 and 4:1:1, I believe we humans would notice . . . which one is better (has more color information/resolution)? And if we could quantify it, are we losing, say 30% compared to 4:4:4?

Thanks!

 

[Johnny Friday]

4:2:2 Or
4:2:2 HQ with Chroma Enabled? So, even more difference

 

[Gabe Conroy]

At 4:2:2 you are loosing 50% color resolution over 4:4:4
At 4:1:1 or 4:2:0 you are loosing 75% of the color resolution

4:2:0 and 4:1:1 technically have the same amount of information, just sampled differently. Depending on what the content is one may show less artifacts than the other.


It may help to think of it this way:

Think of a 720x480 image,

If you have a 4:4:4 720x480 image your black & white (luma) 720x480 and your color (chroma) has a 720x480 resolution

At 4:2:2 your luma still has 720x480 but your chroma resolution is now only 360x240

At 4:1:1 your luma is still 720x480 but your chroma is now only 180x120. So if you only looked at the luma signal it would look the same as the 4:4:4 image but your chroma signal is now a blurry mess.

I can see a clear difference between 4:1:1 and 4:2:2, it's tougher between 4:2:2 and 4:4:4. If you are doing something like greenscreen or secondary color correction the difference in the results is drastic.

Your chroma resolutions are off...halving both vertical and horizontal results in 1/4th the resolution, not 1/2. Likewise 180x120 is 1/16th of 720x480, not 1/4th.

 

[Mark Phelan]

Here is a comparison of a green screen image in ProRes 422 vs 4444. The 442 image is on the left, 4444 is on the right. Compare her sweater sleeve on the right side against the grass.

This was just thrown together for illustrative purposes, no color correction done and the background image is just placed for a quick-see at how the green screen might look.

 

[Stacey Spears]

At 4:2:2 your luma still has 720x480 but your chroma resolution is now only 360x240

That is incorrect. 4:2:2 would reduce chroma to 360x480 in your example. 4:2:0 would reduce chroma to 360x240.

Here is a really old article I wrote on 4:2:0 http://www.hometheaterhifi.com/the-...or-and-the-420-interlaced-chroma-problem.html

 

[Alexander Ibrahim]

Here is a comparison of a green screen image in ProRes 422 vs 4444. The 442 image is on the left, 4444 is on the right. Compare her sweater sleeve on the right side against the grass.

Its not just by the grass ... that's merely one place its easy to see. Take a close look at how the hair composites with the sky.

Also, look at the foreground image. To match the background you have to increase the contrast quite a bit.

Later, on top of the color correction to make the composite work, you have to do any color correction you need for creative purposes.

For that reason, I'd typically lower contrast in the background while raising the foreground/subject contrast minimally.

My point is that it becomes a cascading problem. Whatever you do to the original plates limits what you can do further down the post chain.

The more limited your color information, the more these limits will restrict you.

The flip side is 4:2:0 footage. I'm doing a conversion using 5D to RGB of 7D footage to ProResHQ. So, that process actually interpolates some color data in the original. Mostly that ends up being good- but sometimes keys are not as clean as they would be using camera masters. Its hard to predict when that will be the case.

So ... stay in your best color space as much as you can for as long as you can.

 

[Matthew Greene]

You guys are correct, I deleted the post, no time to correct it right now.