Originally Posted by Allan Jayne
(copied from another post)
I would be interested in finding out whether anyone has developed a process that actually corrected CUE as opposed to just detecting it or filtering it.
By "correcting" CUE I mean, out of every four scan lines, exchanging the chrominance between the second and third. The detection process is still needed, namely to find out which lines are first of a foursome. If first of a foursome changes from frame to frame every now and then, the process is more complex.
Exchanging chroma samples between the 2nd and 3rd lines won't necessarily correct a CUE problem. It would do this only in the case of simple chroma sample replication - i.e., when a given chroma sample is used unaltered for the 2 lines it's assumed to be located between. That's a pretty poor upconversion technique, though, and I would hope that it's not being commonly used. Of course I would hope that CUE doesn't happen either, so perhaps it's more common than I would like.
Assuming that a more proper interpolation or filtering is done based on the presumed position of the chroma sample relative to the luma samples being used, then all of the chroma values in the 4:2:2 output will be incorrectly calculated. None of them will be identical to the original 4:2:0 values and none of them will be correct. In addition, you'd need to know the exact method used to do the interpolation or filtering in order to be able to correct them. Even then, you may well find that reversing the filtering process to recover the original samples is not possible as information may have been lost in the upconversion.
Now, it may be that swapping certain chroma pairs in the upconverted 4:2:2 signal gives you a better-looking result than not doing the exchange at all, but they won't be the same as if the chroma upconversion was done correctly in the first place.
- Dale Adams