Almost all the video we watch every day, from streaming platforms to smartphone footage, stores colour at a lower resolution than the picture itself. This is not a hidden flaw but a deliberate choice, based on a feature of human vision: we are far more sensitive to changes in brightness than to changes in colour. Chroma subsampling exploits this asymmetry to save data without the eye noticing, in most cases.
Luminance and chrominance
Screens work in RGB, but video is almost always encoded in a model called Y'CbCr:
- Y' (luma) represents brightness, essentially a black-and-white picture;
- Cb and Cr (chroma) represent the blue and red differences relative to brightness.
Separating the data this way lets brightness and colour be treated differently. Luma is always kept at full resolution; the two chroma channels can be reduced.
How to read the notation
The J:a:b notation describes a reference block 4 pixels wide and 2 high:
- the first number (4) is the width of the block;
- the second indicates how many colour samples there are in the first row;
- the third indicates how many new colour samples there are in the second row.
| Scheme | Colour resolution | Data relative to 4:4:4 | Typical use |
|---|---|---|---|
| 4:4:4 | full in both directions | 100% | graphics, visual effects, screen recording |
| 4:2:2 | half horizontal, full vertical | about 67% | broadcast production, editing |
| 4:2:0 | half horizontal and half vertical | 50% | distribution, streaming, discs, smartphones |
In 4:2:0, then, a single colour sample is shared by a 2×2 square of pixels. In a Full HD video luma is at 1920×1080, while each chroma channel is at 960×540.
When you don't notice
On natural footage, such as landscapes, people or interiors, 4:2:0 is practically invisible. Boundaries between colours in the real world are rarely sharp to the single pixel, and luma keeps all the detail the eye uses to perceive sharpness.
When you do notice
Problems arise when colour edges are sharp and thin.
- Coloured text on a coloured background: red lettering on a blue background can show smeared outlines or a slight fringe.
- Screen recordings: interfaces, code and spreadsheets have thin characters and solid colours. In 4:2:0 coloured text loses definition, while white or black text stays sharp because it depends mostly on luma.
- Chroma key: keying a subject off a green background needs precise colour edges. With little colour, the outline of hair and fingers becomes ragged.
- Heavy colour grading: pushing saturation hard or separating close hues exposes the colour blocks.
Practical implications
For those shooting
Many cameras record 4:2:0 at 8 bit in consumer formats and offer 4:2:2 at 10 bit in more advanced modes. If you already know you will do chroma key or significant grading, record in 4:2:2 whenever possible.
For those editing
During editing it is better not to reduce colour further. An intermediate codec in 4:2:2 or 4:4:4 preserves information through the various stages; the move to 4:2:0 should happen only in the final export.
For those recording the screen
For tutorials and software demos, recording in 4:4:4 or, at least, choosing high-contrast colours with light text on a dark background or vice versa greatly reduces the problem. Bear in mind, though, that most platforms re-encode to 4:2:0: text that is too small and colourful will be sacrificed anyway.
With FFmpeg
The pixel format is controlled with the -pix_fmt option. For maximum compatibility in distribution:
ffmpeg -i input.mov -c:v libx264 -pix_fmt yuv420p output.mp4
Many hardware players and some browsers will not play H.264 in 4:2:2 or 4:4:4: if an exported file plays on a computer but not on the TV, the pixel format is one of the first parameters to check.
Subsampling and bit depth
These are two different things, often mentioned together. Subsampling concerns how many colour samples there are; bit depth concerns how many levels each sample can take. A 10-bit 4:2:0 video has fewer colour samples than an 8-bit 4:2:2 one, but finer gradations, and therefore less banding in skies and gradients.
In short
- Video separates brightness (luma) from colour (chroma) and reduces colour resolution.
- 4:2:0 keeps one colour sample per 2×2 pixels and is the standard for distribution.
- It goes unnoticed on natural footage, but shows on coloured text, screen captures and chroma key.
- Shoot and edit in 4:2:2 or 4:4:4 if you expect heavy processing.
- For final compatibility, export as yuv420p.
