How to IVTC a field-blended source

[FlimsyFeet]

I have some cartoons on a PAL format VHS tape. I think it's been converted from NTSC due to the presence of blended fields in my capture.

You people who regularly deal with anime probably see this sort of thing all the time, but it's new to me. I tried AnimeIVTC(mode=2, normconv=true) but that didn't remove all the blends.

Doing a DoubleWeave, it looks like all the complete unblended frames can be recovered from the video (it also looks like their are only 12 complete frames per second, I believe this is called animated in 2's). I imagine the answer lies in a DoubleWeave and some long convoluted SelectEvery command, but how do you determine the pattern to go from 25fps to 23.976fps? Or is there an easier way?

I can upload a sample if anyone is prepared to take a look at the source.

[Didée]

Here you go (http://forum.doom9.org/showthread.php?t=97447&highlight=manual+deblend).

[mandarinka]

(it also looks like their are only 12 complete frames per second, I believe this is called animated in 2's)

The problem is that pans, zooms, credit rolls and similar kinds of motion will still be "animated" at 24fps and you won't have a clean field to pick there.

Since it is a conversion, you could use qtgmc (bob deinterlacer) and srestore on top of that to bring it to 23.976fps.

Personally though, if it is possible (also, this would apply to NTSC source), I prefer to do IVTC with tfm and tdecimate and after that postprocess with daa(). After tfm, you will naturally get interlaced ghosting caused by the blended fields, but daa treats that well. In motion, the blends aren't *that* noticeable, and if you use daa with nnedi3 instead of the old nnedi, it won't harm detail much - usually much less than bob-deinterlacing would.

In some cases, the blends are too strong and the interlacing is too much for daa, though. In such case, the qtgmc option is probably better.

[Didée]

@mandarinka - Most of that is wrong, but I'm too tired for explaining. Zzzzz...[snore]...zzzzZ

[FlimsyFeet]

The problem is that pans, zooms, credit rolls and similar kinds of motion will still be "animated" at 24fps ...
Well, this bit was correct at least I've now discovered.

[mandarinka]

I'm aware that it's like fixing stuff with a hammer. But if the result is better than bobbing, what can I do...

[Didée]

If it's a "standard" normconversion done by the usual converter box at the TV station, then

- 24fps sections are no problem, since one clean field of the original frame is always present

- bob+Srestore is the best (automated) choice anyway, however QTGMC not necessarily (technically not well suited for fieldblended sources, not always good for animations, and even less for a combination of them both)

- if it's a fieldblended normconversion, then starting with TFM is a really poor choice. Basically you get 12 frames blended, 12 frames clean, 12 frames blended, 12 frames clean, etc. Daa is not related to that. A blended frame is a blended frame, no matter if you call daa a hundred times.


If there's something "technically unusual" with the source, then posting a sample would be good.

[mandarinka]

- if it's a fieldblended normconversion, then starting with TFM is a really poor choice.

Yeah, I probably shouldn't have mentioned that given that the original poster didn't have NTSC source (the ivtc mention tempted me).


Daa is not related to that. A blended frame is a blended frame, no matter if you call daa a hundred times.
Yes, daa won't remove any blend. However my problem was not with the blends themselves (since without bobbing - and in a lot of cases even with bobbing - you can't get rid of them completely), actually. My problem was the combing and aliasing they cause in the field-matched clip, and Daa is great at polishing these issues and making the afflicted frame watchable.

[FlimsyFeet]

I tied scharfis_brain's method, and bob + srestore, and neither gave a clean output.
Looking at the bobbed video it looks like there might be some sort of chroma blending present, chorma from one frame is carried over into the next.

I'll post a sample.

[FlimsyFeet]

Sorry about the delay, I've uploaded a 30MB sample here:
http://www.sendspace.com/file/erjt15

I'd appreciate some advice on how to deal with this blending.

[dragon_warrior]

- bob+Srestore is the best (automated) choice anyway, however QTGMC not necessarily (technically not well suited for fieldblended sources, not always good for animations, and even less for a combination of them both)

So what's the most suitable bobber to be used here?
As a rule of thumb, I always apply qtgmc() followed by srestore to handle such type of NTSC -> PAL footage.

edit: I never think this framerate converted stuff could be IVTC'd. It could be restored to original framerate, however.

[jmac698]

Blended Chroma? That smells like DOCI - a script I wrote to solve this problem. In that case, I'll check it out. Also look at exblend.

[FlimsyFeet]

So what's the most suitable bobber to be used here?
As a rule of thumb, I always apply qtgmc() followed by srestore to handle such type of NTSC -> PAL footage.

edit: I never think this framerate converted stuff could be IVTC'd. It could be restored to original framerate, however.
I have been trying srestore after a plain vanilla Bob() - my reasoning is that the more advanced bob filters usually opearte temporally as well, and obviously I want to keep the clean fields clean without any elements of adajcent blended fields creeping in. Happy to be corrected if my assumptions are wrong.

And yes, I was using the term IVTC, when resotring the original progressive 23.976fps from 25i is what I'm trying to do.

[retartedted]

I have another example of the bad 24fps -> 30fps -> 25fps conversion footage from the same cartoons.

http://www.sendspace.com/file/ze0vi1

Using scharfis_brain's method can sometimes clean up the footage when on a stationary background, but when it pans there is a stutter effect.

Changed the multiplier from 10 to 25 to give more options, and still had problems.

AviSource("pal_test.avi")
tdeint(mode=1,full=false)
changefps(23.976 * 25,linear=false)
selectevery(25,x) # x= 0...25 / try those values until the output is smooth.

I still haven't found a good method for getting the footage back to 24fps.