A long time ago TV was black and white and snooker was confusing. Then those awfully clever Americans invented ‘color’ TV that you could watch on a black and white set. They did this by keeping the black and white part of the signal and adding two colour signals, which were mostly ignored by black and white TVs (see the sidebar on Y’, Cr, Cb). This television standard was called NTSC, which, for technical reasons, stands for “Never Twice the Same Colour”.
Equally clever German chappies came along in the ‘60s and invented PAL stealing a few bits of NTSC (which was a bit rubbish) from the Americans and a few bits of SECAM (which was a bit French) from the French.
Several years passed before some men with beards at the BBC, envious of the epithet ‘clever’, decided to show off by making television digital. The first thing they had to do was to make the analog signals (Y’, Cr and Cb) into digital signals. In the late 1970s this was very difficult and, though they built a big rack of equipment to do it, they couldn’t make it run fast enough to capture all of the bandwidth of the black and white signal, only about half of it. Of course, they knew that the colour signals were half the bandwidth of the black and white, so they could ‘sample’ them (as converting from analog to digital is often called) at half the rate. These sample rates became known as 2:1:1 – twice the speed for Y’ as for Cr and Cb. Later, they improved the electronics in their rack so that they could sample the full range of frequencies – which meant doubling the sample rates – to 4:2:2.
This means that each line is sampled the same way – the first ‘pixel’ has a Y’ sample, a Cr and a Cb sample. The next pixel has only a Y’ sample. The third pixel has Y’, Cr and Cb, the forth only Y’ and so on.
You don’t have to be clever to work out that 4:1:1 (used in the NTSC version of DV) has ¼ the horizontal colour resolution compared to the luma. 4:4:4 samples them all the same and Sony’s HDCam uses 3:1:1. Although the sample rates have increased for HD, the terminology for the relative rates for the three bits of the video signal has been retained.
So what is 4:2:0 all about? It’s used in PAL DV and HDV and I’m pretty sure there are some Cb samples in both of those standards. The pictures would look really weird without them.
It is, of course, a bastardisation of the terminology. I don’t know who first came up with the term, but when I find out I’m revoking their membership of the Pedant Club – that’ll show ‘em.
Basically, 4:2:0 is a term used for systems that also reduce the colour resolution vertically as well as horizontally. This was very complex to do in the old analog TV days, so no-one bothered. It’s much easier once everything is digital, and makes as much sense as reducing the horizontal colour resolution, based on the way our eyes work. I still hate the term though…
Along with interlacing, all these colour sampling schemes stem from the old analog TV days. They make a certain sense with uncompressed digital video, but nowadays almost every moving picture transmitted in the world has gone through some form of data compression algorithm at some point. Indeed, some people refer to the analog reduction in colour resolution as compression, but I think it too crude to warrant the term. Modern image compression makes far more intelligent judgements on what to remove from an image so we, the viewers, don’t really notice. It has been shown, by clever people from all around the world, that most of these schemes work better if fed with 4:4:4 video – they can output a compressed data stream which is no larger than one originated in 4:2:2 or 4:2:0, but which can be reconstructed to 4:4:4.
Why is that useful? Well, extracting a chroma key works better if you haven’t thrown away half (or more) of your chroma in the first place. Numerical resolution isn’t everything – 4:4:4 images seem to preserve more subtle tonality than even 4:2:2. At the end of the day it just seems archaic to hobble very clever modern, adaptive video compression schemes, which are quite capable of deciding how much colour information they can throw away without us noticing, because of a practical decision made by engineers in the 1950s.
Any of those incredibly clever Japanese people listening?
