I wrote a fairly lengthy response to a question about desaturating video clips on Apple’s FCP message board, but it seemed like something worth posting here too.
YUV should correctly be called YCrCb, with Cr and Cb begin the red and blue channels, the 1:1 in 4:1:1. Y is the green channel plus luminance, the 4 in 4:1:1. Being able to select the green channel produces better results in DV because that’s the channel that carries the luma information.
Yes, broadcast technicians seem to consider ‘YUV’ to be slang. That said, there are plenty more hits on Google for YUV than for YCrCb, and FCP and QuickTime use the term ‘YUV’ internally.
YUV refers to a method of storing RGB data in a smaller bandwidth. The information in YUV is converted back to RGB in order to display as a full color image. Part of the reason for separating Luminance from Color was to create a transmission method for color television which was backwards compatible with b/w tv and used less signal bandwidth than sending three b/w (rgb) channels of information. The solution is an incredibly elegant hack and a testament to the genius of the people who invented television.
The reason the Green channel looks clearest is because 60% of luminance is made of green light. The two color channels are not red and blue, they are ranges between Red-Green (Cr, U) and Yellow-Blue (Cb, V). This is sometimes confused because those ranges are determined by subtracting from Red and Blue. Something to note is that green exists in all three YUV channels: it’s 60% of Luma, explicit in Cr and Yellow is an additive combination of Green and Red. Blue appears as 10% of Luma and only half of the Cb color information. I suspect this is a primary reason for using greenscreens instead of bluescreens for video keying.
I wrote about YUV here, with links to several good sources: FXScript Reference: RGB and YUV Color.
I’m not remembering whether PAL is 4:2:0, but the 4:1:1 sampling in NTSC DV is quite easy to see. Sampling color information every four pixels is the reason DV Chroma is so blocky and drifts off to the right side of hard edges. I’m pretty sure the 4:x:x syntax has nothing to do with RGB and instead refers to the recorded sampling of information in YUV space. One of the great things about DV is that the Luma information is the same resolution as any other professional video format. The only degradation is in the recording of color information, which is usually very hard to distinguish unless you’re specifically looking for it.
When using the channel-blending trick the results mimic black and white film shot with a colored glass filter. The color channel(s) used are the chromatic opposite of the glass filter which would produce the same effect. For example the Green channel looks like b/w film shot with a pink/magenta filter, the Blue channel looks like b/w film shot through a yellow filter. It’s possible to combine channels to get more specific intermediary colors, something a lot of people do with the Channel Mixer in Photoshop. Combining 50% Red with 50% Green yields similar results to a blue glass filter. That idea is the basis for my Saturation & Colorize filter, I wanted to give the same level of control over b/w results as a person would have when shopping for glass filters at B&H. One cool side effect of math instead of reality is that the same methods can be used to saturate as well as de-saturate, with some useful results.