(originally posted here∞)

The loss may not be intuitive from your linked diagrams because of all of the special cases, interlacing, etc. Let's think about a simplified version of the problem.

Consider the made-up colorspace that uses YUV pixels with the chroma subsampled vertically by 1/2. I'll call it 4:4:0. How would you convert YUY2 (4:2:2) to this colorspace?

X = luma and chroma sampled
o = luma only sampled

Source 4:2:2
X o X o
X o X o
X o X o
X o X o

Desired output 4:4:0
X X X X
o o o o
X X X X
o o o o

If you were just given that so-called 4:4:0 stream without any context, you might assume that all 8 of those X's represent information sampled from the original source. However, we can see than in actuality only the X's that I marked as bold are taken from the source; the others have to be interpolated. The bolded o's represent sample points where chroma information has been lost forever. (A complex multi-tap resampler will take some of them into account when interpolating new values for the unbolded X's, but that's not the same thing.)

---

Back to CollectedWritings