Chroma subsampling stores colour information at a lower spatial resolution than brightness information. Human vision is much more sensitive to luminance detail than to colour detail, so this discards data most viewers cannot see.
The notation describes the sampling pattern:
- 4:4:4 — full colour resolution. No subsampling.
- 4:2:2 — colour at half horizontal resolution. Common in broadcast and higher-end camera formats.
- 4:2:0 — colour at half horizontal and half vertical resolution. This is what essentially all delivered video uses, including every H.264 file your phone has ever produced.
Going from 4:4:4 to 4:2:0 removes half the total sample data before the codec has done anything, which is why it is standard.
Where it becomes visible: saturated colour against saturated colour, at edges. Red text on a blue background is the pathological case, and it is not rare — it appears in slide decks, UI screenshots, subtitles with coloured outlines and logo animations. The edges fringe and smear in a way that more bitrate will not repair, because the colour resolution was thrown away before encoding.
For screen recordings that contain fine coloured text, 4:4:4 is worth it if the pipeline supports it, and the compatibility cost is real — many hardware decoders handle 4:2:0 only. For camera footage it is almost never worth thinking about.