Reading the notation
| Notation | Colour resolution | Typical use |
|---|---|---|
| 4:4:4 | Full, same as luma | Grading, compositing, screen content |
| 4:2:2 | Half horizontal | Broadcast and professional acquisition |
| 4:2:0 | Half horizontal and half vertical | Delivery: streaming, Blu-ray, web |
4:2:0 stores one colour sample for every four brightness samples, which is a 50% reduction in total data before any codec runs.
Why delivery accepts the loss
At normal viewing distances the eye cannot resolve colour detail at full luma resolution, so 4:2:0 is close to free on photographic content. It stops being free on synthetic material with hard colour edges: coloured text on a contrasting background is the standard failure case, which is why screen recordings suffer more from subsampling than camera footage does.
And in HDR
HDR10 delivery is normally 10-bit 4:2:0, expressed as the pixel format yuv420p10le. The bit depth and the subsampling are independent choices: 10-bit is about how finely each sample is quantized, subsampling is about how many colour samples exist. HDR needs the first; delivery bandwidth dictates the second.
ffprobe -v error -select_streams v:0 \
-show_entries stream=pix_fmt -of default=nw=1 input.mp4
Questions people ask
Can I convert 4:2:0 back to 4:4:4?
You can change the pixel format, and it gains you nothing. The discarded colour samples are gone, and upsampling interpolates rather than restores. The only reason to do it is as an intermediate step in a processing chain that needs full chroma.
Does subsampling affect HDR more than SDR?
Not inherently, but the two are often confused. Bit depth and subsampling are independent: 10-bit 4:2:0 is the normal HDR delivery format. If a picture looks soft around coloured edges, that is subsampling, and no amount of dynamic range addresses it.