What Is Chroma Subsampling? (4:4:4, 4:2:2, 4:2:0 Explained)
Why JPEG stores colour at lower resolution than brightness, what the ratios mean, and when you can safely give up colour detail.
Human eyes are far more sensitive to brightness than to colour. JPEG exploits this: it stores the brightness channel at full resolution and the two colour channels at reduced resolution. The reduction ratio is chroma subsampling, written as three numbers.
Reading the notation
The numbers describe how many colour samples exist per group of brightness samples, in the form J:a:b.
- 4:4:4 — for every 4 brightness samples there are 4 colour samples. No reduction.
- 4:2:2 — 4 brightness samples, 2 colour samples per row. Colour resolution is halved horizontally. File size typically drops 15–25%.
- 4:2:0 — 4 brightness samples, 2 colour samples on alternating rows. Colour is halved in both directions, so only a quarter of the original colour data survives. Saves a further 10–20% over 4:2:2.
The file size difference between 4:4:4 and 4:2:0 is real, but it is smaller than most people assume — usually under 35% of the total.
When 4:2:0 is fine
For natural photographs of foliage, skies, clouds, and skin at normal viewing size, 4:2:0 is hard to distinguish. Broadcast television used 4:2:0 for decades.
When it is not
- Red and magenta detail. The colour channels carry the red difference, so saturated reds smear first. Red flowers, red text, warning signs.
- Thin coloured lines and text. Coloured UI screenshots, charts, maps with coloured boundaries.
- Flat colour with sharp edges. Logos, illustrations, pixel art, anything synthetic.
- Any image that will be edited again. Chroma loss compounds. Two rounds of 4:2:0 is visibly worse than one.
If your source is synthetic or generated — the output of an image model, a screenshot, a vector export — it is very likely 4:4:4, and it is the kind of image that suffers most from subsampling.
Why "just raise the quality" does not undo it
Subsampling happens before quantization. Once the colour channels have been averaged and stored at half resolution, no quality setting can recreate the discarded samples. Raising the quality only spends more bits on the already-degraded data, which makes the file bigger without restoring detail.
That is the whole argument for keeping the sampling mode of the source: it costs some bytes, and those bytes are the only ones that buy back real colour information.
Checking your own files
Drop a JPEG into the tool above. The line under the filename shows the sampling mode we detected in the file header, along with the source's approximate quality and whether it is progressive. Nothing is uploaded, so you can check private images safely.