Filter for 'weird' interlacing?

[bartel75]

Hello,

Could anyone suggest a nice filter to make the following video smooth again? The regular deinterlace filters (tomsmocomp, yadif, etc.) don't seem to handle this one...

http://4.bp.blogspot.com/_OdiNwEAc3js/S_WXRLRof1I/AAAAAAAAAXc/2UPgBF2eYjw/s1600/snap.jpg

Video sample (6 mb vob):
http://rapidshare.com/files/389671750/VTS_01_1.VOB
or
http://www.megaupload.com/?d=ZKRYGNXQ

Thanks in advance, and sorry for the rather vague title; I don't know how I'd call this phenomenon, so it was a bit hard to search the forum for similar cases. If you know what this is actually called, thanks for letting me know...

[LoRd_MuldeR]

The obvious question: Did you try YadifMod+NNEDI2 or TempGaussMC yet ???

[poisondeathray]

it's actually not interlaced , and those horizontal lines are encoded into each separate field

I don't know how to improve it without blurring the crap out of it - But I would wait for some of the experts to chime in

[LoRd_MuldeR]

it's actually not interlaced , and those horizontal lines are encoded into each separate field

Now that I actually have the sample, it looks like the individual fields already contain something that looks like re-sized interlace.

I doubt there is a way to "fix" this mess :rolleyes:

http://i45.tinypic.com/15hhpqv.jpg

[bartel75]

I doubt there is a way to "fix" this mess :rolleyes:

I guess I'll just leave it like it is, then. Thanks for your input, guys...

[wonkey_monkey]

I think it's an old 405-line broadcast. You might get something slightly nicer by downsizing vertically (but to what size is a matter for experimentation) and then bobbing/reinterlacing. I'd have a go myself but it's bedtime...

David

[LoRd_MuldeR]

A very quick experiment with this code crudely smoothed out the interlacing artifacts:

MPEG2Source("C:\Temp\!_Downloads\VTS_01_1.d2v")
SeparateFields()
Merge(Last.SelectEven(),Last.SelectOdd())
NNEDI3(field=-2)
Merge(Last.SelectEven(),Last.SelectOdd())
NNEDI3(dh=true)

[wonkey_monkey]

I tried:


lanczosresize(720,320) # guesswork; another value for height may work better
bob # also use a better bobber!
lanczosresize(720,576)


My guess: film recording telecined to 405-line videotape, then converted to PAL in a quick-and-dirty fashion.

David

[bartel75]

A very quick experiment with this code crudely smoothed out the interlacing artifacts:

Very acceptable! :)
Very slow (0.12 fps)! :(

[bartel75]

I tried:


lanczosresize(720,320) # guesswork; another value for height may work better
bob # also use a better bobber!
lanczosresize(720,576)


Not as nice as LoRd_MuldeR's solution, but at least the speed is within reason. Thanks!

[10L23r]

It looks like a 1.5x upscale, so it should be
lanczosresize(720,384)

[2Bdecided]

It's a telerecording (kinescope) from a 405-line TV broadcast. They would have pointed a film camera at a TV screen, and captured the image displayed. The camera would have been synchronised to the video signal. This is quite high quality, and has captured both fields without blurring.

The resulting film has just been transferred to DVD like any other normal feature film - the "interlacing" you see is on the film print - the transfer from film to DVD isn't "interlaced" as such. (I know it's flagged interlaced, but so are many PAL film transfers).

It's quite rare for the original interlacing to be so well intact - it would have made this film very difficult to re-transmit, because there's no way the interlacing can "line up" again, so you'd get very strange effects watching the film on a 405-line interlaced CRT. Usually the fields are blurred together a little using spot wobble, either on the TV screen as the film is being captured, or in the film scanner when the film is being converted back to video.

It must be possible to detect the original scan lines, recover the original interlaced video, deinterlace it, and upscale it (this will have about 375 visible lines - you need 576 for PAL). Deinterlacing and upscaling can be done very well in AVIsynth, but I don't think anyone has ever recovered the original scan lines from a film print like this.

yet.

...but given that the far harder problem of colour recovery from a B&W film print (http://colour-recovery.wikispaces.com/) has been acheived, this should be a walk in the park for some clever person! ;)

Cheers,
David..

[2Bdecided]

Without anything clever, my suggestion would be...

mpeg2source("405-line interlaced VTS_01_1.d2v")

bilinearresize(720,188)

nnedi2(field=0,dh=true)

spline36resize(720,576)


Adjust the number 188 - higher to maintain more detail (but leave more "interlacing").

You could make it adaptive and only process the areas with movement. That would keep full resolution on stationary parts (where there's no "interlacing") - though adaptively blurring like this make look even stranger than having the whole thing look soft.

Cheers,
David.

[Terka]

dont know if its suitable, but yuo may have a look at:
http://compression.ru/video/old_film_recover/field_shift_en.html

[LoRd_MuldeR]

Very acceptable! :)
Very slow (0.12 fps)! :(

NNEDI has some "speed -vs- quality" options you could tweak for speed. Also NNEDI2 is a lot faster than NNEDI3...

[bartel75]

Without anything clever, my suggestion would be...
(...)
Adjust the number 188 - higher to maintain more detail (but leave more "interlacing").

I put the number just a little higher (204) and used spline36resize(720,528) to get the AR right. I'm currently encoding (at 4 fps, which is acceptable for me).

The quality is not great, but at least it's better than the original dvd! :)

Thank you (all of you) for your suggestions!

[bartel75]

Also NNEDI2 is a lot faster than NNEDI3...

I just noticed, based on 2Bdecided's suggestion! :)

[2Bdecided]

btw, what is the DVD?

[bartel75]

btw, what is the DVD?

http://www.amazon.com/Beethoven-Cello-Sonatas-Rostropovich-Richter/dp/B000092T5P (European R2 version)

[zilog jones]

It must be possible to detect the original scan lines, recover the original interlaced video, deinterlace it, and upscale it (this will have about 375 visible lines - you need 576 for PAL). Deinterlacing and upscaling can be done very well in AVIsynth, but I don't think anyone has ever recovered the original scan lines from a film print like this.

yet.

...but given that the far harder problem of colour recovery from a B&W film print (http://colour-recovery.wikispaces.com/) has been acheived, this should be a walk in the park for some clever person! ;)

Field restoration from telerecordings has been done: http://en.wikipedia.org/wiki/VidFIRE
I have not seen any of the restored programmes mentioned (they don't seem to broadcast them very often) but it sounds promising.

[Undead Sega]

VidFire can actually be done in Avisynth.

[2Bdecided]

Field restoration from telerecordings has been done: http://en.wikipedia.org/wiki/VidFIRE
I have not seen any of the restored programmes mentioned (they don't seem to broadcast them very often) but it sounds promising.No, VidFIRE may sound similar, but the fluid motion it restores is completely fake i.e. re-created. AFAICT it is normally used with telerecordings where the two fields are completely inseparable / blurred, or even where there's only one field present (skip-field telerecording).

On this specific telerecording, both fields are present, and they're wonderfully clear. So a suitable technique could recover the actual original video motion, not some re-creation of it.

Cheers,
David.

[2Bdecided]

VidFire can actually be done in Avisynth.Yes, you can convert 25p>50p>50i using MVtools using AVIsynth.

IMO VidFIRE is a slightly more subtle, and far less artefact-prone process than what you get from MVtools. It's likely though that the worst artefacts of VidFIRE are simply edited out manually; if so, it's not possible to make a fair comparison.

Cheers,
David.

[zilog jones]

No, VidFIRE may sound similar, but the fluid motion it restores is completely fake i.e. re-created. AFAICT it is normally used with telerecordings where the two fields are completely inseparable / blurred, or even where there's only one field present (skip-field telerecording).

I didn't know that, thanks for clarifying. I thought it actually recreated something using what original fields were left on the film (so assumed it wouldn't work with skip-field, spot wobble, etc.) but I guess not.