De-interlacing a PAL progressive stream???

[Corrado]

Hi,

I know that the subject may sound crazy :) , but let me explain...

I thought I had some interlacing artifacts on a PAL movie converted to DivX4 (marked as interlaced by DVD2AVI, as always), while it turned out that probably it was just a player problem (see this thread (http://rilanparty.com/vbb/showthread.php?s=&threadid=15993) for more detail, if you're interested.

In fact, after trying 11 different alternatives, I found no differences between the pre- and post-deinterlacing movies (see below for more details).

So my question: what does it happen to a PAL progressive stream if you apply a de-interlacing algorithm??? Is it possible that the movie remains unchanged (a part from image blurring)?

Here the comparison I made:

1. No deinterlace, sharp bicubic resize
2. No deinterlace, bilinear resize
3. GreedyHMA(1,0,0,0,0,0,0,0)
4. VerticalReduceBy2
5. SmartDeinterlace(2,15,true,true,true)
6. GreedyHMA(1,0,1,0,0,0,0,0)
7. GreedyHMA(1,0,1,0,1,0,0,0)
8. GreedyHMA(1,0,2,0,1,0,0,120)
9. SeparateFields+Trim(1,0)+Weave
10.SeparateFields+ComplementParity+Trim(1,0)+Weave
11.DoubleWeave+SelectOdd

1 & 2 were just made for reference in sharpness.
3, 4 & 5 are the ones in Gknot's .avs files.
6 to 11 were found on this board as solution for PAL deinterlacing or for field swap/shift problems.

The results:
3 to 8 = equal to 1-2 (a part from blurring, more or less evident depending on the method)
9 to 11 = the movie became remarkably interlaced (you can spot two different overlapping fields during scene changes).

Those tests (expecially the last 3) show, I think, that the original movie wasn't interlaced at all (although DVD2AVI says so), do you agree? But then, why the methods 3 to 8 didn't change a thing?

Thanks
Corrado

[hakko504]

If I understands Greedy correct, it shouldn't change anything if it detects progressive frames. (3,6,7,8)
Vertical reduce by 2 will blend the pixels and halve the vertical resolution (2 pix become 1)
SmartDeinterlace will only blur areas that it thinks has interlacing effects.

Conclusion: If you have a nice progressive stream, then none of the methods should change anything. Have you tried something like this:
video1=AVISource("c:\myvideo.avi")
video2=SmartDeinterlace(2,15,true,true,true)
return Subtract(video1,video2)

(the Levels filter might be needed in order to actually see anything)

[trbarry]

In examples 7 & 8 above some blurring would have occured because the 5'th parm turned on the vertical filter.

The main processing option in GreedyHMA is the AutoPulldown type, the 3rd parm. Setting it to 0 tells GreedyHMA you have Video source (as opposed to film). In this case Greedy is motion sensitive and will tend to blur or bob on action scenes, even if there would otherwise be low comb artifacts.

Example 6 above set GreedyHMA on automatic and probably would have given the best results if there was both film and interlaced source in the file.

But note that film source does not alway mean progressive in PAL. It is still possible that the fields are matched perfectly but out of phase by 1 frame. GreedyHMA will attempt to do 2:2 pulldown reversal (IVTC) and handle this.

But if the entire clip was from film, with nothing from a video camera, then you would be best off still forcing film with something like:

GreedyHMA(1,0,2,0,0,0,0,120) # force film for PAL

which is Greedy's equivalent of PAL force-film, but with a max-comb level of 120 to still deinterlace any really bad frame.

- Tom

[Corrado]

Thanks for the info.

I may be stupid, but can't find a reference for the following doubts, please help me telling if I'm right or wrong:

1. Video means "interlaced" source (in PAL = each field is shot at 1/50sec. and 2 are combined in a frame at 1/25sec.)

2. Film means "progressive" source (in PAL = both fields are shot at the same time at 1/25sec - or at 1/24 then "accelerated" to 1/25 - and then merged in a frame)

3. 2:2 pulldown is something I don't catch completely. Is it only what you say, adjust the phase shift? Why this phase should be uncorrect in a movie?

And finally, related to point 3 maybe, why do I read "interlaced" in DVD2AVI if the movie isn't so?

Thanks a lot for your patience...:)

Corrado

[ppera2]

"But note that film source does not alway mean progressive in PAL. It is still possible that the fields are matched perfectly but out of phase by 1 frame. "

I'm not expert for all this, but it looks that there is not clean what 'frame' means. By my knowledge, frame should mean 1 whole frame (picture) and field should mean half frame - every second line - top or bottom field.
Why sometimes stays 50 fps for PAL ? Is it usual by interlaced video ?

Why talking about shift by 1 frame ? It should be 1 field. So we have second field from first orig. frame and first from second orig. frame etc. in this case.

[trbarry]

Corrado -

Your comments are correct AFAIK, but see below.

ppera2 -

Yes, my terminology is sloppy, and maybe just plain wrong. Even when we discuss frames we often use the phrase to mean two different things. There is the 'real' frame of the video, as shot by the film camera that contains 1 large field, or two matching fields. It is a single picture representing a point in time.

However there is also a frame that is presented internally inside most video processing software. This also contains 2 fields but only on very lucky days will those two fields consistantly represent the same point in time. In the case of video capture this is often because interlaced video has probably gone through a telecine process and is sent 1 field at a time and then combined by video capture software.

In the case of a DVD it is probably because the folks that created it were evil, incompetant, or just did not care. On NTSC DVD's of course the telecine process adds extra fields to get the (field) rate up to 60. I live in NTSC-land but even on PAL DVD's there supposedly can be phase errors like this, and they have to be put back together.

It has been my experience that it is easy to match up the fields again but in automatic situations it is not always easy to decide that you are supposed to. Even video cameras or material often seems to have a sort of cadence to it that makes it appear slightly like you should apply 2:2 pulldown removal, even in US NTSC.

I can offer no good explanation for this last but it does sometimes confuse GreedyHMA automatic pulldown processing for PAL sources. That is one reason why it is good to tell GreedyHMA to use the force film modes when you know you have a properly mastered film based movie, even if there are cadence changes, edits, and/or dropped frames.

- Tom

[ppera2]

trbarry-

Yes, unfortunatelly we have lot of bad digital video material. It additionally complicates all video conversions. And most of people don't understand whole thing. And when terminology is confuse, it makes it real nasty.
Is there some reference about all this terminology ? Can we make it cleaner somehow ?

[trbarry]

ppera2 -

Sorry, as I already sorta confessed, some of my terminology is not either standard or proper. Some of it is a carryover of jargon developed by the some of us on the DScaler team and is maybe not the way the rest of the world talks about it. I remember there was a thread on AV Science where we were talking about pulldown removal and Blight interjected a question about "IVTC". At first I didn't even know what he was talking about.

So I would stick with the terminology in the guides and related links, at least here on Doom9.

- Tom