It makes use of a characteristic of the human eye-- we're much more sensitive to luminance
information (brightness) than we are to chroma information (color). You just don't need to convey
all of the color information in order to deliver a pretty good picture as long as you provide good
brightness data.
Video cameras make use of this a lot. They might sample HD luminance at 1920x1080 points,
but the chroma might be sampled at every 4 locations, e.g. 960x540 points. Very high quality cameras
might sample at half the locations or even all of them, but these demand higher data rates and larger
storage media and a higher price.
Here's a picture that illustrates the concept of chroma sub-sampling.
The middle column ("chroma") contains 3 different chroma sampling examples. The saturation of the colors increases toward the right and the hue varies as you travel downward.
The third column ("complete picture") represents the combination of luminance and chroma to yield a complete picture.
The first row represents a full sampling of both luminance and chroma. It's the ideal sampling scheme.
The second row shows the 4:2:2 subsampling scheme. Color info is captured for each pair of pixels.
The third row represents a common sub-sampling scheme called 4:2:0. Color information is only captured for each 2x2 pixel block of four pixels.
The fourth row shows the sampling scheme called 4:1:1. It also samples the color values every four pixels but you can see that the configuration is horizontal rather than square.
The third column shows the result when you combine the sub-sampled color information with the full luminance info. Compare them to the complete picture at the top of the third column.
This presentation accentuates the differences. It's important to understand that a real 12x12 pixel grid takes up only about 1/10 inch (2.5mm) in each dimension. Stand far away from your computer, or blur your eyes a bit and see if the grids in the right hand column look different from each other.