Trying to deinterlace source, 5 out of every 6 frames interlaced

[chilledinsanity]

Hey, I'm hoping this is a fairly easy question, I just haven't been able to figure it out looking at decomb parameters. I have a source where, like the title says, 5 out of every 6 frames are interlaced, with 1 progressive, it's like clockwork.

I've been trying to figure out what settings I should use in avisynth that will result in the best looking picture. I don't care what the end framerate is (I think it might be PAL), I just want to deinterlace the source and avoid the ghosting effect if possible. I feel like my answer lies in decomb with the decimate settings, but I haven't figured out which ones yet (I've been trying Decimate(6) in combo with various other combinations). To clarify, if I use Kernel or TDeint to deinterlace, then 1 out of every 6 frames will have the ghosting effect.

I feel like I just need the right command combinations for my script, anyone have suggestions as to what I should try? (I'll post results here).

[JMP]

OK, start with simplest doubleframerate deinterlacing (like Bob(), or even SeparateFields()) and tell us:
1) Do you see "ghosts" in any of the resulting frames?
2) How many different frames do you see per second of your video (pick a scene with movement and just count :) )
Information about the framerate ouf your clip could also help us to figure this out...

[chilledinsanity]

Thanks for the reply, here's the information you wanted, this is from using Bob():

1. Yes. I'm having difficulty making out the exact pattern/ratio of ghost frames to regular ones though using Bob.

2. They're all different. I can't find any duplicates (I checked carefully).

As for the source, it's a low budget Australian DVD movie. I really don't know the source, I'm guessing it's a PAL that's been converted to NTSC due to the number ratios.

[JMP]

OK... additional question to your answer 2: you mean they are all different in the original or in the bobbed stream? Actually if you have some progressive frames, then you should have, after bobbing, pairs of very close (almost identical) ones. Don't look at small differences resulting from bob() reconstruction of missing fields. Pick a sequence with motion and watch some object moving: in how many frames does it change the position?

Now to ghosts: the next question is: do you see in the bobbed stream individual blended frames, separated by "good" ones, or is it blend, after, blend, after blend? The latter case is beyond my knowledge... In the former you have so-called fieldblended video and you may try to clean it with Restore24(), or CDeblend() - see discussion threads devoted to these functions to learn how to use them.

[chilledinsanity]

They're all different in both. I mean here's the situation. In the original stream I'll have 5 frames with interlacing in them, then one frame with no combing lines at all, looks fine. In both the original stream and after a bob, I'm looking at sideways camera motion for my test and there's definitely a difference in motion in every single frame. Believe me, I would notice duplicate or near duplicate frames, I hate those.

As for the second part, it's the former situation, with most of the frames looking okay, with some ghosts in between. I'd say it's a 5 out of 6, but bobbing seems to add some fuzziness I didn't see with the others, so I can't swear to it. In any case, there is at LEAST one ghost image for every 5 regular images.

I'll try and research Restore24() and CDeBlend() in the meantime, thanks again for the help.

[Mug Funky]

i recently noticed something similar on some video'd extras in NTSC.

it's probably a PAL source that's been bobbed, changefps'd to 59.94, then re-interlaced to 29.97.

so bob + decimate(6) + reinterlace should do you.

[manono]

Yeah, a bad PAL2NTSC conversion. You can't just IVTC it. Use either RePAL or Restore24 on it. RePAL is way easier to set up than either Restore24 or CDeblend.

LeakKernelBob(Order=1)
RePAL()

[JMP]

I'm not sure... RePAL, if I understand, assumes a certain fixed pattern of original/blended fields in the bobbed stream. It may or may not work on this material.

On top of this I'm a bit disoriented by the chilledinsanity's claim, that in the bobbed stream all the frames are different. This may indicate that it is a PAL->NTSC conversion of an interlaced (50 fps) video - in such case RePAL will do no good...

[manono]

Yes, you're right on both counts, and maybe I shouldn't have made such a blanket statement. I've also seen a few instances of PAL2NTSC where RePAL didn't work. It takes almost no time to find out though, and, in my opinion, it should be tried first. If it doesn't work, then you go on to something else.

However, without having a sample to examine, it does show all the signs of being RePALable. No offense to chilled insanity, but he's said both that 1 frame in 6 is progressive and also that every bobbed frame is different. Do you also see a contradiction there?

[JMP]

No offense to chilled insanity, but he's said both that 1 frame in 6 is progressive and also that every bobbed frame is different. Do you also see a contradiction there?
That's one of the reasons I said I was disoriented :) . Yes, a short sample of the material would be helpful.

[chilledinsanity]

No offense to chilled insanity, but he's said both that 1 frame in 6 is progressive and also that every bobbed frame is different. Do you also see a contradiction there?

Yeah, I just now realized that for whatever reason, Bobbing wasn't doubling the framerate, my bad. I fixed it and now there are repeat frames. There will be 4 non repeating, then 2 that are the same, so kind of like this:

110000110000110000

And there are still ghost frames within these. Sorry about the confusion on my part before.

Edit: Oh and about Restore24, I went through 3 different versions sifting around on the forum and got a different problem with each one. Do you have a link to the good one along with versions required of the .dll's and with an updated readme (one I downloaded only had the .avs and the .dll's).

[JMP]

OK, so the thing slowly becomes clear... Looks like a trurly interlaced PAL converted to NTSC by duplicating every 5-th field. To get rid of the doubling: Bob, Decimate(6) and either leave it at 50 fps, of weave back to 25 fps, as you prefer.

Now, if I knew where the blends come from... Some earlier conversion maybe? How often do you estimate you have blended frames in the bobbed stream?

[scharfis_brain]

provide an unprocessed sample clip!

[chilledinsanity]

JMP:

I doubt it's an earlier conversion since I'm not running much of anything in the script besides Bob(). As for the ghost frames, I'd estimate about half of them are blended in Bob mode.


scharfis_brain:

Yeah, I might later. As it is now I'm lazy because I know that involves me tracking down a VOB splitter, making sure I can split the file alright, find a place to host it, which may mean fileshare or megaupload, which I haven't used before, so then I'll have to register an account, then verify it by email, then depending on how they accept the file I'll have to put on some FTP software, then configure it and hope my router doesn't get in the way.

[manono]

Hi-

Use http://rapidshare.de/ to upload. No registration. Just upload, they give you the link, and you post it here.

As for cutting the video, use the "[" and "]" buttons in DGIndex to isolate a section with motion and then File->Save Project and Demux Video. Upload a 10-20 second section to Rapidshare. Easy as pie.

[chilledinsanity]

Nice, that's a lot easier than what I envisioned. Here's the link:

http://rapidshare.de/files/29569546/Sample.demuxed.m2v.html

I took a different sample with more motion (it's a good test, lots of people in a crowd, then a constant rotation around one person) and now I'm not even sure it's a constant 5i 1p. I'm thinking regular deinterlacing is going to be my only real option with this bastard, but I uploaded it anyway because I'm curious what you guys think the source material of this is.

[manono]

Standard PAL2NTSC stuff. RePAL works fine:

LeakKernelBob(Order=0)#because that clip was BFF
RePAL()

[scharfis_brain]

I'd try this one:

TDeint(mode=1, full=fallse, tryweave=true)
changefps(250)
selectevery(10,?)

replace ? with a number between 0 an 9. THis specifies the pattern offset.
Also TDeint will enhance the image quality a lot here, cause it weaves most of the fields together.