Good deinterlace for PAL?

[AlexeyS]

I have few videos. This is PAL 25 fps MPEG2 videos. The 99% of movie is 1939-1945 videos (plus some computer graphics).

Is this possible to turn this video to progressive without forced deinterlace?

PS: This is example:
http://www.callofdutyserver.net/!tempmov/pal.m2v

[neuron2]

First, get rid of the ! character in your URLs. It makes it a PITA to fetch your clips.

That clip is progressive PAL with a one-field phase shift. You can use Telecide by itself, for example:

Telecide(order=1,post=0)

Always be careful to get the order correct when using Telecide. You can postprocess with a high vthresh if any combs slip through at scene changes.

[eb]

Use VDM,
in Video > Filterrs
set Filter deinterlace, discard field1
set Filter deinterlace , discard field 2
set Filter resize 720x576

save avi or mpeg

see sample
ftp://www.eb.enterpol.pl:eb@www.eb.enterpol.pl/pal+.avi

eb

[neuron2]

@eb

Ha, ha. Funny joke!

[eb]

@neuron2,

It is not joke, for sure

eb

[neuron2]

Justify your solution in technical terms, comparing it to simple field matching (which restores the original progressive frames, without loss of vertical detail).

[AlexeyS]

Originally posted by neuron2
First, get rid of the ! character in your URLs. It makes it a PITA to fetch your clips.

That clip is progressive PAL with a one-field phase shift. You can use Telecide by itself, for example:

Telecide(order=1,post=0)

Always be careful to get the order correct when using Telecide. You can postprocess with a high vthresh if any combs slip through at scene changes.
Wow. Thanks a lot. I've tried to deinterlace this movie for hours, but all videos were with some artefacts. Still can't understand all this order, guide, post etc. :( But I'll try. :)

First, get rid of the ! character in your URLs. It makes it a PITA to fetch your clips.
Ok. I'll make new folder soon...

[AlexeyS]

Another PAL video (http://www.callofdutyserver.net/iii/ast.m2v)

I use this script:
Telecide(order=1,guide=0)

It's working good. But on sometimes I see some interlace artifacts (look at gold thing in this sample for example). Is this script is good or I've made something wrong? :confused:

[eb]

For such cases as your sample I am using TomsMoComp deinterlacing
see sample
ftp://www.eb.enterpol.pl:eb@www.eb.enterpol.pl/ast+.01.avi

eb
EDIT
Use VirtualDubMod for looking for details or MPlayer.

[neuron2]

@AlexeyS

Is that the source video or the processed video? Seeing your processed video doesn't help us at all. We have to see the *unprocessed* source video.

[AlexeyS]

Is that the source video or the processed video? Seeing your processed video doesn't help us at all. We have to see the *unprocessed* source video. [/B]
It's source.

[neuron2]

That is not telecined video, so you should not use Telecide on it. You need a pure deinterlacer, such as TDeint, FieldDeinterlace, TomsMoComp, KernelDeint, LeakKernelDeint, etc.

[AlexeyS]

Originally posted by neuron2
That is not telecined video, so you should not use Telecide on it. You need a pure deinterlacer, such as TDeint, FieldDeinterlace, TomsMoComp, KernelDeint, LeakKernelDeint, etc.
Ok, but how did you know that (just for future)? :confused:

[neuron2]

@AlexeyS

I separated the fields and then stepped through them. I saw that every field was different from the one before. That is the "signature" of pure interlaced video:

a b c d e f g ... (fields)

The signature for progressive video is this:

a a b b c c d d e e ...

The signature for telecine is this:

a a a b b c c c d d e e e ...

SeparateFields() is the key to the kingdom. Make sure to get the field order correct using AssumeTFF() or AssumeBFF() as required.

[AlexeyS]

Originally posted by neuron2
@AlexeyS

I separated the fields and then stepped through them. I saw that every field was different from the one before. That is the "signature" of pure interlaced video:
I see. :)
BTW is there some big guide for 100% sure detection of video type (combined, film, pure interlaced, telecine etc)?

SeparateFields() is the key to the kingdom. Make sure to get the field order correct using AssumeTFF() or AssumeBFF() as required.
I should use AssumeTFF() or AssumeBFF() first? Or I can use only this to detect type of video:
MPEG2Source("d:\Astra\astra.d2v")
SeparateFields()

Also, I've tried to use AssumeTFF()/AssumeBFF() and I can't see any backwards video (field order wrong). I can see it on few movies I've tried, but here AssumeTFF() and AssumeBFF() works similar. :confused:

For such cases as your sample I am using TomsMoComp deinterlacing
What settings you've used?

[neuron2]

I do this:

AssumeTFF().SeparateFields()

Then I step through a high motion scene. If the motion goes forward then backward instead of always forward, then I change it to this:

AssumeBFF().SeparateFields()

If the order makes no difference then you have progressive video.

Then I look to see the "signature" as I described it.

In the case of an MPEG2 you have the D2V file, which can give you extra clues. But the field stepping is the definitive, reliable way to know what kind of video you have.

[AlexeyS]

Originally posted by neuron2
I do this:

AssumeTFF().SeparateFields()

Then I step through a high motion scene. If the motion goes forward then backward instead of always forward, then I change it to this:

AssumeBFF().SeparateFields()

If the order makes no difference then you have progressive video.

Then I look to see the "signature" as I described it.

In the case of an MPEG2 you have the D2V file, which can give you extra clues. But the field stepping is the definitive, reliable way to know what kind of video you have.
Ok. I take movie (Sample (http://www.callofdutyserver.net/iii/shr.m2v)). First of all, I save DGIndex project (Filed Operation >> None). DGIndex shows "NTSC 29%".

Then I use script AssumeBFF().SeparateFields() and AssumeTFF().SeparateFields() and see that video is BottomFieldFirst (order=0).

Also, whole movie is 3:2 (I see this when using AssumeBFF().SeparateFields()).

But, when I use Telecide(order=1,guide=1,post=0,show=true), the video sometimes "in-pattern", "in-pattern*" and "out-of-pattern".

I wonder what I should use: guide=1 or guide=0? :confused:
I've tried both and didn't see difference (but one of this should be better).

PS: Maybe stupid question, but when I need postproccesing (in what cases)?

[AlexeyS]

Originally posted by neuron2
First, get rid of the ! character in your URLs. It makes it a PITA to fetch your clips.

That clip is progressive PAL with a one-field phase shift. You can use Telecide by itself, for example:

Telecide(order=1,post=0)

Always be careful to get the order correct when using Telecide. You can postprocess with a high vthresh if any combs slip through at scene changes.
I've just finished encode my PAL movie (my first post in this thread). And sometimes movie has this artifacts (zoomed 2x):

http://www.callofdutyserver.net/iii/artif.jpg

Is this something wrong with Telecide(order=1,post=0)? :confused:

[neuron2]

Originally posted by AlexeyS
Then I use script AssumeBFF().SeparateFields() and AssumeTFF().SeparateFields() and see that video is BottomFieldFirst (order=0).

Also, whole movie is 3:2 (I see this when using AssumeBFF().SeparateFields()).

But, when I use Telecide(order=1,guide=1,post=0,show=true), the video sometimes "in-pattern", "in-pattern*" and "out-of-pattern". You said the movie is BFF and then you ran with order=1. It is crucial to specify the field order correctly.

So, I wonder what I should use guide=1 or guide=0? :confused:
I've tried both and don't see difference (but one of this should be better). If the whole movie is 3:2 then certainly use guide=1. But don't forget to set the order parameter correctly!

PS: Maybe stupid question, but when I need postproccesing (in what cases)? Alexey, this is described in the documents that come with Decomb. You're just going to have to do some reading.

[neuron2]

Originally posted by AlexeyS
I've just finished encode my PAL movie (my first post in this thread). And sometimes movie has this artifacts (zoomed 2x):
Is this something wrong with Telecide(order=1,post=0)? You turned off postprocessing, so if Telecide cannot match fields for a frame, or misses a match due to noise, etc., it will come through as combed. Turn on postprocessing to catch these stray combs.

Again, all of this basic stuff is described in the documents. I spent a lot of time making those documents, and I'd really appreciate it if you would study them.

[AlexeyS]

If the whole movie is 3:2 then certainly use guide=1. But don't forget to set the order parameter correctly!
I use guide=1,order 0:

http://www.callofdutyserver.net/iii/cat.jpg

This is cat falling. Is this double silhouette really normal? I can see this in all movie (fast movements).

Alexey, this is described in the documents that come with Decomb. You're just going to have to do some reading. [/B]
Yes, I've read it. But I'm asking not "how to do", but "what cases"? ;)

[AlexeyS]

Originally posted by neuron2
You turned off postprocessing, so if Telecide cannot match fields for a frame, or misses a match due to noise, etc., it will come through as combed. Turn on postprocessing to catch these stray combs.

Again, all of this basic stuff is described in the documents. I spent a lot of time making those documents, and I'd really appreciate it if you would study them.
I see. I will make experiments with postprocessing.

[neuron2]

Originally posted by AlexeyS
This is cat falling. Is this double silhouette really normal? I can see this in all movie (fast movements). It's camera motion blur. The exposure for each frame is a fixed time duration. If the objects move during that time, you will get motion blur. It has nothing to do with Avisynth, filters, etc., unless you're asking for blind deconvolution. :)

Yes, I've read it. But I'm asking not "how to do", but "what cases"? If combed frames come through the field matching process, postprocessing is appropriate.

[neuron2]

Originally posted by AlexeyS
I see. I will make experiments with postprocessing. It's described quite clearly in the tutorial document.

[AlexeyS]

Originally posted by neuron2
It's camera motion blur. The exposure for each frame is a fixed time duration. If the objects move during that time, you will get motion blur. It has nothing to do with Avisynth, filters, etc., unless you're asking for blind deconvolution. :)
Yes, but this is computer graphics. ;) That's why I'm asking...

If combed frames come through the field matching process, postprocessing is appropriate.
Ok.

[neuron2]

Originally posted by AlexeyS
Yes, but this is computer graphics. ;) That's why I'm asking... Apparently the content creator wanted to simulate motion blur. It seems to me to be a reasonable thing to do when depicting high speed motion.

[AlexeyS]

Originally posted by neuron2
Apparently the content creator wanted to simulate motion blur. It seems to me to be a reasonable thing to do when depicting high speed motion.
Ok.

By the way, I can't find ideal Telecide parametrs. I still get little (but sometimes annoying) interlaced artefacts even with good postprocessing. I've tried TomsMoComp and it's give me visually better quality with my WW2 black and white video. Is this possible? :confused:

[neuron2]

I can't read minds, Alexey. If you want my opinion on something, you have to post an unprocessed source clip for me to inspect.

Maybe you've messed up again and applied Telecide when you shouldn't, or set the wrong field order, or set up postprocessing wrong. Sure, that's possible.

[AlexeyS]

Originally posted by neuron2
I can't read minds, Alexey. If you want my opinion on something, you have to post an unprocessed source clip for me to inspect.
Here is another sample (http://www.callofdutyserver.net/iii/inter.m2v) of same B&W movie - with interlace artifacts.

I've tried:
Telecide(order=1,post=2,guide=2,show=true,blend=false,vthresh=25)

Also tried vthresh 50...

[neuron2]

It's field-blended video. You can't IVTC that. Use a deinterlacer. And search the forum for "blended fields".

[AlexeyS]

Originally posted by neuron2
It's field-blended video. You can't IVTC that. Use a deinterlacer. And search the forum for "blended fields".
Ok. But how did you know that this video is "blended fields"?

[neuron2]

I already told you how to analyze video. I also told you to do a search. Everything you'd want to know about blended fields is available in the Decomb user manual and existing forum threads.

[Wilbert]

I suggest you read the following docs to gain better understanding

http://www.doom9.org/index.html?/capture/postprocessing_avisynth.html (section 7.2.4)
http://www.doom9.org/ivtc-tut.htm (everything)

[zilog jones]

Originally posted by Wilbert
http://www.doom9.org/ivtc-tut.htm (everything)

That page could do with some updating. There's a few dead links there, and there's no mention of stuff like Restore24, mvbob and other newer de-interlacers. But the information on actual types of video/telecine is still good.

[Wilbert]

That page could do with some updating.
I know that they are working on an update :)