Advice on IVTC on a HongKong DVD?

[vybez]

Hi!
Have a HongKong-movie that I'm trying to perform IVTC on which looks like there's only 1 progressive frame out of 5, the other 4 show various degrees of interlacing artifacts.

Have read the forums and the IVTC guide and tried various methods as standard IVTC (Telecide(order=1,guide=1).Decimate(5) (according to the Bad Telecines section of the guide). Also tried two pal to ntsc methods; (Telecide(order=1,guide=3).decimate(6) and repal ( bob() + repal() ) to see if that would make any difference.

So far neither of the three methods have given the results I had hoped for. Maybe I'm being optimistic on what can be done, but would be grateful if anyone with more experience in this area would care to look at a sample and give advice if there might be another method to minimize the artifacts -
Sample removed

Thanks in advance.

[neuron2]

By doing assumetff.separatefields and then stepping through, you can see that it is straight 3:2 pulldown. I applied Telecide as follows:

Telecide(order=1,guide=1,post=0)

But as you observe, some frames show some residual combing. Telecide is locking onto the correct 3:2 pattern right from the start and holding it throughout the clip. So I chose one of the bad frames and did an override to test the 3 matches (p, c, n), and indeed, the best match is the one chosen. Next I separated fields again and looked at the two matched fields. You can see a variation from field to field that is responsible for the residual combing. It doesn't look like field blending. I've seen things like this before.

I don't see much you can do. I did play it at full speed and size, and the effect was inperceptible to me, even when zoomed to full screen.

Maybe scharfis_brain will have an another opinion about it.

[vybez]

Thanks!

Will try that and see how it turns out.

[MOmonster]

Ivct material from Hongkong?
I only heard bad things about. :scared:
I looked at the source.
It doesn't look like field blending. I've seen things like this before.
Oh, for me it looks like field blending. :)
The blendlevels are that small, that the bobbed input looks just blurred on this frames. Try restore24 or Crestore (will be updated next week) to reverse this conversation or use telecide with postprocessing if you can live with the results.
Crestore should detect this softening as blends (maybe you have to higher the bthresh). Just try it out.

Edit: Do you have this combing in all higher motion scenes or only in some parts of your clip?

[scharfis_brain]

Maybe scharfis_brain will have an another opinion about it.
Yes. A brute force one:
blur(0,1) after telecide.decimate

Oh, for me it looks like field blending.
I thought about this too. But throwed away this thought.
There are only two possibilities, that can create such semi-smeared images:
- too strong fieldwise temporal softening (denoising)
- too strong interlacer compression.
- maybe both.
But it is NO fieldblending.

My attempt on this source was

doubleweave().selectevery(5,0,2).blur(0,1)

fast.
quick.
dirty.
and non-adaptive ;)

[vhelp]

slight OT here..

fwiw, I'd like to add that, for anyone who wants to see a good example
of this, with a full source (ie, dvd) you can take a good look at the
DVD movie, "Event Horizon".

This one has those blened-like frames. The video is not sharp, and looks
like those in the sample vob that vybez U/L 'ed here.

-vhelp

[MOmonster]

@scharfis brain
Yes, you are right. I just thought it because my function is able to avoid this strange frames and it seems to show something like a pattern (but the output of Crestore is jerky).
It seems to be a problem because of interlaced encoding I think.

I don´t like the blur(0,1) solution.
My Cdeblend function should be able to catch most of these bad frames.

Use a smartbobbed (best also weaved) input for my Cdeblend (for example tdeint(tryweave=true, mode=1) and than decimate the result to 23.976fps to get a fluid motion.

[vhelp]

IMO, after oversving the sample vob, I believe it has to do with Field
Switching or Field Line is moved up/down by one or two scanlines, in those
frames, which are patterned inside the 3:2 sequence.

(because of where the first frame begins, in vybez 's vob) ..
Lets start with Frame #0 and move onward.


1 1 1 1 1
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 ..
I I Pb Pb P I I Pb Pb P I I Pb Pb P ..


* where++ P=Progressive and b=blend, and I=Interlace
* ++note, where Pb = Progressive_blend
* where you see 'P' only, this is a clean and harp frame.

Perhaps, there is a filter to account for the two Pb 's within
the 3:2 patter sequence that can find these two in each sequence
and then 'undo' them, and make them look sharp, like they are
suppose to ??

-vhelp

[vybez]

Edit: Do you have this combing in all higher motion scenes or only in some parts of your clip?

The combing is like that in the whole movie, with one progressive, followed by two heavily interlaced and the two frames with small combing artefacts as vhelp described. I'm gonna try the other suggestions tomorrow (2 a.m. here right now..)

Thanks to everyone who has replied!

[MOmonster]

@vhelp
This is what I meaned with something like a pattern.
Perhaps, there is a filter to account for the two Pb 's within
the 3:2 patter sequence that can find these two in each sequence
and then 'undo' them, and make them look sharp, like they are
suppose to ??
Trying to undo them is not the best idea (a hard job), just using the right fields for the restoring is fully sufficient. :)

@vybez
So here is my function for you.
off=0
a=tdeint(mode=1).selectevery(5,off)
b=doubleweave().selectevery(5,off+2)

interleave(a,b)
Simple, or? Set the off value (between 0 and 4) till you see no more combing in the output. If the offset change really to often to set them by hand, ask for a automated version. If I have some time, I could do it for you. ;)

[Mug Funky]

i'd like to know what created such a source. i saw another just like it the other day - a bollywood film (Raincoat) . i figured it was PAL->NTSC conversion, but it seemed to keep too strong a 3:2 pattern, and didn't seem to be blended enough.

i've seen other sources where chroma (but not luma) has been deinterlaced, so to get correct IVTC i had to run decimate(5) on the chroma channels and merge it with an IVTC'd luma, otherwise the chroma would comb and exhibit a kind of delay that was impossible to get rid of. it took a while to figure that one out.... very very weird stuff.

[edit]

- too strong interlacer compression.
i don't quite get what you mean there. interlaced picture encoded as progressive? that would ring the opposite way, and there'd be stronger blocking. but i can't figure out the cause :scared:

[scharfis_brain]

it's a typo.

I meant too strong interlaced compression.

[Mug Funky]

that's what i thought. though, i think if alternate scan is enabled, there'll be no crosstalk between fields. and if alternate scan is off, you'll get an inverted first field on the second field (like sharpener overshoot), plus much stronger ringing and blocking.

looks like the quality varies a bit with hong kong DVDs - i've only seen good ones, but all i hear of here are bad ones. perhaps there's some DVD producers with less scruples than others.

[vybez]

big thanks to everyone who replied and gave advice!!
i decided to go with scharfis_brain's suggestion as that was the most pleasing to me.

mug funky: about the quality of hk dvd's. they're definitely better mastered now, but this one was released back in 1999.. don't get me started on the english subs.. ;)