a thought on 4:2:2 chroma subsampling

Y' data is collected at every pixel. With chroma subsampling the number of Cr or Cb data is smaller than the number of Y' data. This can occur in two different ways:

• Bundling method: Cr or Cb data is collected for groupings of pixels (functioning like bigger pixels).
• Omission method: For some pixels no Cr or Cb data is collected.

Note we're describing how the Y',Cb, and Cr data is acquired from the optical image, not how it is organized in the file or the stream.

Whether chroma subsampling is by the Bundling method or by the Omission method makes a real difference in digital imaging quality. The Bundling method accords better with human vision, and corresponds more nearly to reduced chroma signal frequency in analog video. Figure 2 of BT.601 (available online) illustrates digital 4:2:2 video chroma subsampling as a time diagram. That figure shows each chroma sample spread over two consecutive luminance samples: the Bundling method.

Sometime between BT.601 (1982) and the Apple paper "Uncompressed Y'CbCr Video in QuickTime Files" (1999) digital 4:2:2 video changed from using the Bundling method over double-size pixels to using the Omission method skipping every other pixel. Beware that the published literature is not always clear about this distinction. For example Figure 2 in Poynton's "Chroma subsampling notation" (2008, available online) shows each Cb and Cr having "exent" 2 pixels wide. I do not think he is illustrating the Bundling method because he had a hand in writing the Apple paper. But according to the Omission method the "extent" of where the chroma values come from is 1 pixel wide, and the "extent" of their influence on display is 3 pixels wide.

I've verified that FCP uses the Omission method for its 4:2:2 subsampling. I made a 720x576 frame with 1-pixel columns of alternating colors in codec "None". And the same frame mirrored right-left. When these frames are transcoded to Uncompressed 4:2:2 they come out quite different, each getting just one of the two colors' chroma. If the chroma subsampling used double-size pixels the two transcodes would have come out identical.

Do any 4:2:2 cameras, or other converters, still use the Bundling method (that is, heed BT.601, which is still a current standard)? This would make quite a mess, since the two methods are incompatible.

Dennis Couzin
Berlin, Germany