Originally Posted by Aleron Ives
The nomenclature seems to have developed around the concept of IVTC filters, since the general "best practise" technique in computer circles is to undo any field or frame repetition in order to restore the original progressive frames in any video.
Yes - I think because 24fps film content that has been telecined to video in the US has 3:2 pull-down that is introduced by the telecine, the addition of 3:2 pulldown (the pulldown refers to the physical act of pulling down the film during blanking on early telecines which didn't employ continuous scanning) has become intertwined with the telecine process. However 3:2 repetition can be added to 24p video, which has never been near a telecine, and in Europe we use 2:2 pull-down, where there are no duplicate fields to detect and remove/cope with.
There's so much misinformation and poor explanation of broadcast video - and in particular interlacing and de-interlacing - it gets a bit annoying. Particularly when a phrase appears that is based on a standard broadcast production process (Telecine-ing is the act of transferring film to video using a telecine machine) is used to mean something else (removing 3:2 pull-down/repetition which may or may not have been caused by a telecine)
As such, any video with an altered framerate is expected to undergo IVTC to restore the progressive frames and decimation to remove duplicate ones, whereas material that is truly interlaced is more difficult to decomb.
Yes - though in 50i/60i they are both forms of de-interlacing. 24p/25p content with 3:2/2:2 can be restored losslessly with the right de-interlacing algorithm, whereas native interlaced content will always be ambiguous (there is spectrum folding in the vertical spatial and temporal domains that cannot be undone - though some good guesswork can be made) Obviously native interlaced 50i/60i content should be de-interlaced to 50p/60p not 25p/30p if you want to keep the full motion (i.e. temporal resolution) of the original. How you do this depends on the quality you want to achieve (particularly the vertical resolution) and the processing power you have. The best de-interlacers will use stuff like Phase Correlation Motion Compensation to detect which areas are moving, and in what direction, to create the "missing" fields that need to be created to generate a decent 50p or 60p result.
(These processes aren't new, and in fact have been an issue in broadcast production since the 80s. High quality DVEs (Digital Video Effects devices) used in production control rooms, that zoom and shrink, rotate, page peel etc. pictures have used realtime de-interlacing to generate the highest vertical resolution pictures they can when you zoom into a live native interlaced source. If you process fields individually (which lower cost DVEs do) you end up with noticeably poorer quality results (particularly when you enlarge a picture))
Since you seem knowledgeable in this area, one thing I've yet to understand is what happens with US dramas shown at 720p60. Material on stations that broadcast 1080i60 can be restored through IVTC to achieve 1080p24 frames, but if the source is already progressive, I don't know if the frames are simply duplicated (rather than the fields), which would then mean that only decimation would be needed to restore the original framerate, or if there is a separate way to IVTC progressive content to restore its original framerate.
They're both very similar. In interlaced 3:2 you are looking for the repeated field (which will be alternating between field 1 and field 2) in a pair of source frames which you simply discard to give you 2:2 (which you can de-interlace to 1:1 with a simple field-merge - aka weave). You obviously need to continuously detect the field duplication - as 3:2 content edited in the video domain could break the 3:2 cadence (i.e. you could end up with edits that break the repeated 3:2:3:2:3:2 pattern) by comparing fields and where two are identical you discard one.
The same techniques can be done in the 3:2 60p domain, but in this case you also need to detect the repeated pairs of frames as well as the repeated triple frames as you need to go from 3:2 to 1:1 without any de-interlacing. You are still detecting and discarding repeated images - but you have 3 repeated frames rather than 1 repeated field to detect per pair of source frames. With a 3:2 regular cadence 24p within 60p you detect 2 repeated frames followed by 1 repeated frame.
Some compression systems optionally also allow you to flag 3:2 sources so you don't have to send redundant content (though repeated fields/frames compress quite well) so rather than sending the duplicate fields or frames, you flag them for duplication in the receiver. If you have access to the compressed stream you can detect the flags and thus recreate the 24p source without having to run any content-based repeated field/frame detection.