Why chroma sub-sampling is important

Most people know that 4:4:4 is better then 4:2:0, but what is the difference?

Have a look at the following image which is a part of Kodaks Marcie LAD. The image color sampling is  below is 4:4:4

When comparing it to a 4:2:0 version of the same image you can see some artifacts, especially in the hair and between the colors at the top left.

This is because we just have a quarter of the chroma resolution, even if we still have the same number of pixels in both width and height.

The reason why some codecs use lower chroma sub-sampling is because it saves space. We will have a look at this further down

Now let’s have a look at the details. In the comparison below you can see the difference. Between 4:4:4 and 4:2:2 there are some pixel groups that loose detail. When you look at 4:2:0 it’s pretty blurry and the details are lost. When grading it’s very important to keep as much detail as possible, it can be the difference between a great looking image and an OK looking one.

Here is another comparison. You can see there is a slight hue shift in some of the pixels in the 4:2:2 version compared to the 4:4:4.
So what is the difference between 4:4:4, 4:2:2, 4:2:0 and 4:1:1

The eye is more sensitive to contrast then color information, that’s why you can sample color in a lower resolution (sub-sampling). Sub-sampling is described in 1/4th, which means 4:2:2 means we sample 100% of the Y-values, 50% of the U values and 50% of the V-values. Y is the luminance value which describes the how bright the pixel is. U and V are chroma information.

Let’s have a look at these diagrams, they are very conceptual.

4:4:4

The top row displays the Y sampling, each pixel gets its own value.

The middle row displays the U and V sampling, each pixel gets it’s own U and V value.

The bottom row diplays the combined YUV signal, each pixel has its own unique value.

Now compare it to the following diagrams.

4:2:2

The top row displays the Y sampling, each pixel gets its own value.

The middle row displays the U and V sampling, every second pixel gets its own U and V value.

The bottom row diplays the combined YUV signal.
Pixel 1 gets its own Y value and also its own U and V values.
Pixel 2 gets its own Y value but the U and V values from Pixel 1.

4:1:1

The top row displays the Y sampling, each pixel gets its own value.

The middle row displays the U and V sampling, every fourth pixel gets it’s own U and V value.

The bottom row diplays the combined YUV signal.
Pixel 1 gets its own Y value and also its own U and V values.
Pixel 2 gets its own Y value but the U and V values from Pixel 1.
Pixel 3 gets its own Y value but the U and V values from Pixel 1.
Pixel 4 gets its own Y value but the U and V values from Pixel 1.

This creates hue shifts that start to look bad.

4:2:0

The top row displays the Y sampling, each pixel gets it’s own value.

The middle row displays the U and V sampling, the first pixel in the top row gets its own U and V value.

The bottom row diplays the combined YUV signal.
Pixel 1 row 1 gets its own Y value and also it’s own U and V values.
Pixel 2 row 1 gets its own Y value but the U and V values from Pixel 1 row 1.
Pixel 1 row 2 gets its own Y value but the U and V values from Pixel 1 row 1.
Pixel 2 row 2 gets its own Y value but the U and V values from Pixel 1 row 2.

 

So why do we use this “crap”? It’s because we save space and bandwith.
We do a little math in an 8 bit colourdepth image. We use 4 pixels for the example.

	pixel 1	pixel 2	pixel 3	pixel 4	total
4:4:4	24 bytes	24 bytes	24 bytes	24 bytes	96 bytes
4:2:2	24 bytes	8 bytes	24 bytes	8 bytes	64 bytes
4:1:1	24 bytes	8 bytes	8 bytes	8 bytes	48 bytes
4:2:0	24 bytes	8 bytes	8 bytes *	8 bytes *	48 bytes

* these pixels are on row 2

The conclusion of all this is, use the best quality image you can work with when grading, and don’t only look at the resolution, but also the which chroma sub-sampling is used. Talk to the DOP you are working with before he/she chooses which camera to film with.

© 2011 Nikolai Waldman