Frame correspondence

[Sharc]

As far as I understand there is no correspondence between the frame types (I, B, P) of the original and the re-encode.
Would the quality of a re-encode benefit from a 1:1 correspondence of I - frames, i.e. an I frame of the original would become am I -frame of the re-enode?
Sorry if this should have been discussed before and I did not find it.

[jdobbs]

No, I really see no advantage to a 1:1 correlation. The encoder makes that determination based upon the GOP size. Most of the time it is also influenced by scene-detection, which decides at what point a new GOP may be started outside the GOP size.

[Sharc]

Thank you for the reply jdobbs.
I was mainly thinking about the I frames, because these constitute a reference for the derived P and B frames, with some error propagation or accumulation possibly occurring until the next I frame sets a new reference.
So I thought if the re-encoder makes a new I frame out of an original B frame, the errors that are inherently present in the original's B frame may become a reference I frame in the re-enocoded stream .... which might not be optimum.
Just some thoughts, maybe not correct.

[Fishman0919]

Some info...

http://bmrc.berkeley.edu/research/mpeg/faq/mpeg2-v38/faq_v38.html

http://bmrc.berkeley.edu/frame/research/mpeg/mpeg_overview.html

http://bmrc.berkeley.edu/frame/research/mpeg/

[Sharc]

Thanks, Fishman
Something to digest ...
My "frame translation model" for the decoding and re-encoding was probably a little too simplistic.