Question about NTSC

[Cyberia]

If I understand correctly, there is no way to start with 29.970 NTSC and truly get 23.976 Film. You can IVTC it, but no matter how good the algorithm, reconstructing progressive frames is not possible. This is because with NTSC each field is it's own timeslice. You can't truly reconstruct a frame without two fields representing the same timeslice.

This is why IVTCing always produces very slighty jerky motion or blended fields? Really then, Field Deinterlace will produce the clearest and smoothest frames in this situation.

Besides a 20% increase in filesize, is there a drawback to always using FD on NTSC material?

[Leak]

Originally posted by Cyberia
If I understand correctly, there is no way to start with 29.970 NTSC and truly get 23.976 Film. You can IVTC it, but no matter how good the algorithm, reconstructing progressive frames is not possible.

That depends on the source. If it was video material shot at 59.xxx fields per second to begin with, then yes, you're out of luck with IVTC. But if the NTSC material was produced via telecining (which you try to undo by inverse telecining) you've got material that went from 24 frames per second to 59.xxx fields per second, where both fields of the reconstructed frames your IVTCer produces belong to the same point in time (modulo some inconsistencies created by edits after the telecine process).

Take anime (unless it was animated for 30FPS like Haibane Renmei, or has 60 fields per second stuff mixed in like some background CGI animations in Chobits) or movies for instance. You start off with 24 frames per second progressive material, then every frame is split into 2 fields (that would be easily recombinable without a quality loss) and in a cycle of 10 fields 2 fields are duplicated (which is what inverse telecining tries to undo) to get to 60 fields per second. If you manage to undo the field duplication, you should be getting more or less perfect progressive frames, unless there was some bad editing or other processing going on as well.

Of course, if you take filmed-for-TV stuff or sports broadcasts that are filmed at full 60 fields per second you can't apply IVTC to it - you have to deinterlace it (which means going from 60 fields per second to 60 frames per second), and perform some frame rate conversion if you want to get it down to 30 or 24 or whatever frames per second.

np: Underworld - Big Mouth (Underworld 1992-2002)

[Guest]

Generally sports, adult material, and general TV crap like the weather is shot on video and so cannot be IVTC'ed, but must be deinterlaced. Films, high-end series (like Sopranos), and anime are usually progressive to start with and so can be IVTC'ed. But as Leak said, sometimes things are hybrid too.

It's easy to tell whther the material is progressive by separating fields of a motion sequence and stepping through. If you get a new picture every field, then it is interlaced. If you get 2 or 3 fields with the same picture, then it is progressive.

Of course it is possible to truly recover progressive frames from NTSC 29.97 content, as long as it is progressive to begin with.

[Cyberia]

Thanks guys. That clears things up nicely. I've done IVTC before and it worked well, but recently I have been trying to process some (made for TV) video that has clearly has no repeated fields. ie: each field is a unique picture. I discovered FD worked better on it than Telecide, now I understand why.

Would it be possible to indicate this difference in DGIndex?
eg: NTSC (Telecine), NTSC (video)

[jggimi]

It requires examination of the frames. See www.doom9.org/ivtc-tut.htm for more info.

[Guest]

What jggimi said. :)

It could be hard telecined and DGIndex would not know that. DGIndex only knows about the MPEG flags for soft telecining. DGIndex would be a liar if he told you a hard telecined clip was video when actually it was hard telecined and IVTC'able. (This brings up my long-standing goal of making a general video analysis tool.)

[jggimi]

AutoGordianKnot, software in beta that len0x is developing, includes a hard analysis. It makes a determination of progressive, interlaced, or hybrid.

[Cyberia]

I wonder if lenox would make a stand-alone program to do this, or a dll that could be called against a d2v?