Crisp › Learn › Chroma subsampling

What is chroma subsampling?

Here is a fact about nearly every video file you have ever watched: it stores far less colour information than brightness information. Not a little less. Typically a quarter as much.

This is chroma subsampling, it has been standard since analogue television, and it works because of how your eyes are built.

Your eyes are not symmetrical about this

Human vision has far more rod cells, which resolve brightness, than cone cells, which resolve colour. We are extremely good at seeing fine detail in light and shade, and comparatively poor at seeing fine detail in hue. Video encoding exploits that directly: keep full resolution for brightness, throw away most of the colour resolution, and the loss is close to invisible on the sort of images cameras usually capture.

The notation describes the sampling of a small block of pixels. 4:4:4 keeps colour at full resolution. 4:2:2 halves it horizontally. 4:2:0 halves it both horizontally and vertically, so one colour sample covers a two-by-two block of pixels. That last one is what your phone shoots, what streaming services deliver, and what almost every h.264 and HEVC file uses.

When you actually notice

On ordinary photographic footage, essentially never. The places it shows up are specific and worth knowing, because they explain problems that otherwise look like a bug.

Saturated text and graphics. Red or blue text on a contrasting background is the classic case. The edges look soft or fringed, because the letterforms are defined by a colour boundary that is being stored at half or quarter resolution. Screen recordings suffer the most.

Chroma keying. Pulling a clean key from green screen footage is much harder at 4:2:0, because the edge between subject and background is exactly the fine colour detail that was discarded. This is why people shoot 4:2:2 or better for compositing.

Heavy grading. Pushing colour hard exaggerates whatever colour resolution you have, so banding and blockiness in gradients become visible.

Repeated re-encoding. Each generation resamples chroma again, and errors compound around sharp colour edges.

What you can and cannot do about it

You cannot recover colour detail that was never sampled, in the same way you cannot recover focus. Converting a 4:2:0 file to 4:4:4 upsamples what is there; it invents nothing.

What you can do is avoid making it worse. Shoot 4:2:2 if your camera offers it and you plan to key or grade heavily. Keep an intermediate master rather than re-encoding a delivery file repeatedly. And when adding text or graphics, do it at the final stage so the letterforms are encoded once rather than surviving several generations.

Crisp exports 4:2:0 for the same reason everyone does: universal compatibility. A file that plays everywhere is worth more than colour resolution most viewers cannot see, and the exceptions above are narrow enough to plan around.

Colour detail, in detail

Should I export 4:4:4?

Only for a specific reason, such as delivering to someone who will key or grade the footage, or archiving text-heavy screen content. For anything headed to a viewer, 4:2:0 is what plays reliably everywhere, and the difference is invisible on photographic material.

Why does red text look blurry in my screen recording?

Chroma subsampling. Text edges are defined by a colour boundary, and at 4:2:0 that boundary is stored at a quarter resolution. Recording at a higher resolution helps more than raising the bitrate, because it gives the colour channel more samples to work with.

Does upscaling fix it?

It can improve the appearance, because a good model reconstructs plausible edges, but it is not recovering the original colour samples. Treat it as making the best of what survived rather than a genuine restoration.

Is this why green screen keys look ragged?

Frequently, yes. The subject-to-background edge is precisely the fine colour detail that 4:2:0 discards. Shooting 4:2:2 or better, and lighting the screen evenly, both matter more than the software doing the key.

Download Crisp for Mac Free to try, one-time $19 to remove the watermark. Runs entirely on your Mac.

See also: why screen-recorded text goes soft.