Your 4K Video Has Less Colour Detail Than You Think

When you set your action camera to 4K, you get a frame 3840 pixels wide and 2160 pixels high. Most people assume that the resulting video contains both brightness and color information for every one of those 8.3 million pixels — but this usually isn't true.

Most action cameras record brightness information at every pixel position, but sample color differences at only a quarter as many positions. This is called 4:2:0 chroma subsampling.

We can get away with this because our eyes pick up fine detail much better from changes in brightness than from changes in color. Video formats exploit this to reduce the amount of data they need to store. You can watch the resulting footage on a large 4K display and struggle to see that anything was lost.

Think of a really sharp black-and-white photograph. You can see individual hairs, fine lettering, the texture of brickwork. There's no color information at all, but there's plenty of detail. Much of what we recognize as sharpness comes from boundaries between light and dark, something our eyes are really good at.

A similar idea works in drawing. Make a precise pencil sketch, then add a loose watercolour wash. The color can be applied with a broader brush because the pencil lines already tell your eye where the edges are.

The same detailed lighthouse sketch in black and white and with loose watercolour added
The pencil lines carry the fine detail, even where the colour is applied loosely.

How 4:2:0 works

Many video formats describe an image using three signals. One is luma (Y′), which carries brightness-related information. The other two are chroma (Cb and Cr), which describe colour differences (how much the blue and red values differ from the luma value). Because luma already contains information from red, green and blue, only two additional signals are needed to reconstruct all three RGB values.

With 4:2:0 subsampling, a group of four luma samples has just one sample from each of the two chroma signals. For UHD 4K video, this results in:

  • Luma: 3840 × 2160 samples
  • Each chroma signal: 1920 × 1080 samples

Each chroma signal has 75% fewer sample positions than the resolution of the frame. When you play the video, the missing color values are estimated from neighbouring samples. The image is rebuilt at 4K, but its finest color boundaries are usually less exact.

4:2:0 sampling diagram showing four luma samples and one sample from each chroma channel per two by two block
Each color-difference channel is sampled at half the width and half the height of the luma channel.

What detail actually gets lost?

Consider red lettering on a grey background with the red and grey close in brightnes.  Because there isn't much brightness difference, the shape of the letters is mainly made up by the color information. When the chroma is sampled less often, those edges can lose some definition. Black text on white behaves differently because there is a strong brightness edge to preserve.

The enlarged comparisons below show the effect on lettering and fine colored edges. They were made by reducing the chroma sampling to 4:2:0 and reconstructing it, without adding video compression or an artificial blur.

Actual 4:2:0 subsampling comparison: black lettering stays sharp while equal-brightness colored lettering and thin colored edges soften
The crops are enlarged so the differences are visible. The effect in ordinary footage is usually much subtler.

How different cameras handle it

The GoPro HERO13 Black records its 10-bit video in 4:2:0. The Sony FX3, on the other hand, offers both 4:2:0 10-bit and 4:2:2 10-bit modes.

Those labels describe two different parts of the recording. The 4:2:0 or 4:2:2 ratio tells you how closely the color is sampled across the frame. The bit depth (10-bit in these examples) determines how finely each sample's value can be recorded (i.e. dynamic range). Both Sony modes use 10-bit values, but 4:2:2 records color differences at full vertical resolution rather than half.

The ARRI ALEXA 35, a high-end cinema camera, offers both 4:2:2 and 4:4:4 recording. With 4:4:4, the colour-difference signals are sampled at the same resolution as the brightness information. Even expensive cinema cameras use different sampling formats depending on the recording mode.

Here's another way to look at 4:2:0: the colour detail is essentially recorded one resolution step below the video resolution. In a 4K recording, the chroma channels have the dimensions of a 1080p image. In an 8K recording, they have the dimensions of a 4K image. And in a 1080p recording, they're just 960 × 540. 

Does it matter?

For most action-camera footage, probably very little. It's an efficient way to reduce file size without visibly sacrificing much detail. Higher chroma sampling is more useful for small colored graphics, green-screen work and other jobs where exact color edges matter.

If you're curious about your own camera, open an original video file in MediaInfo. Look for the field marked "Chroma subsampling".

Back to blog

Leave a comment

Please note, comments need to be approved before they are published.