Deblending again, need some help

[dragon_warrior]

So I've just came across another badly-made DVD footage. The video was originally shot at 25p and the vob is 25i. That sounds great, but because of many blending I doubt that the video has been framerate converted at least once before being put on DVD.

Here is the vob http://www.multiupload.com/T13PMHKTCC

A simple QTGMC will result in a video like this
http://thumbnails26.imagebam.com/13202/abf10d132018815.jpg (http://www.imagebam.com/image/abf10d132018815) http://thumbnails28.imagebam.com/13202/086e14132018830.jpg (http://www.imagebam.com/image/086e14132018830) http://thumbnails14.imagebam.com/13202/959f42132018866.jpg (http://www.imagebam.com/image/959f42132018866) http://thumbnails21.imagebam.com/13202/1316e9132018896.jpg (http://www.imagebam.com/image/1316e9132018896)

I wonder if there are better ways to get rid of these blending (or at least make them less annoying).

Thanks

[Ghitulescu]

Concert videos and videoclips are overwhelmingly shot in 50i (25 fps interlaced) for PAL. Are you sure yours isn't native 50i?
I'm not a big specialist, but I don't see any NTSC->PAL artefacts (framerate conversion) in these 10 seconds of video.
I also assume you see the blending on the PC only.

[dragon_warrior]

Concert videos and videoclips are overwhelmingly shot in 50i (25 fps interlaced) for PAL. Are you sure yours isn't native 50i?
I'm not a big specialist, but I don't see any NTSC->PAL artefacts (framerate conversion) in these 10 seconds of video.
I also assume you see the blending on the PC only.

Thanks, I'm sure the conversion is PAL -> ..... -> PAL again, that's why the vob introduces severe blending.
I'm gonna encode this for PC playback, not for TV or any handheld devices

[manono]

It looks to me like it's doubly blended, possible evidence of PAL->NTSC->PAL. That's the chance you sometimes take with music videos. The people that make many of them don't know any better and care even less.

You might be able to get rid of one layer of blending by bobbing followed by SRestore.

[dragon_warrior]

It looks to me like it's doubly blended, possible evidence of PAL->NTSC->PAL. That's the chance you sometimes take with music videos. The people that make many of them don't know any better and care even less.

You might be able to get rid of one layer of blending by bobbing followed by SRestore.

You mean SRestore(25) or SRestore(23.976)?
And the rest layer of blending, would it possible to be removed?

[Didée]

There is no sign of NTSC<>PAL conversion in this sample. Hard to tell what happened exactly, but it seems the basic problem is that a partial vertical blur was applied to the source while it was in phase-shifted interlaced stage.

Examine the result of this:
mpeg2source("sample_(Shakira)(dragon_warrior).d2v")
o=last

bob1 = o.nnedi3(field=-2)
bob2 = o.sharpen(0,1).nnedi3(field=-2)
max = bob1.mt_logic(bob1.selectevery(1,-1),"max",U=3,V=3).mt_logic(bob1.selectevery(1,1),"max",U=3,V=3)
min = bob1.mt_logic(bob1.selectevery(1,-1),"min",U=3,V=3).mt_logic(bob1.selectevery(1,1),"min",U=3,V=3)
clamped = bob2.mt_clamp(max,min,0,0,U=3,V=3)

x1 = bob1.selectevery(1,-1)
x1 = merge(x1.selecteven(),x1.selectodd())
x2 = clamped.selectevery(2,-1)

stackhorizontal(x1,x2)

return(last)

What happens: the blending basically can be undone/reversed by applying vertical sharpening ["sharpen(0,x)"] to the interlaced source. However this can't be tweaked to a "sweetspot strenght", it's always either to weak or too strong. Therefore a full sharpen(0,1) is used, which definetly is overdoing the reversal. Then the resulting pixels are clamped by [min,max] range of the original input (despite the blending, the valid pixel range is present in the source, and can be used to limit the overshooting of the blend reversal).

"x1" is more or less "the best" that you can get from this source without any sophisticated filtering, by blending the corresponding bob-frames together.

"x2" is the result of blend-reversal plus clamping.

Some high-frequency trails from the blending are remaining, but I don't see a way to avoid that.

[dragon_warrior]

There is no sign of NTSC<>PAL conversion in this sample. Hard to tell what happened exactly, but it seems the basic problem is that a partial vertical blur was applied to the source while it was in phase-shifted interlaced stage.

Examine the result of this:
mpeg2source("sample_(Shakira)(dragon_warrior).d2v")
o=last

bob1 = o.nnedi3(field=-2)
bob2 = o.sharpen(0,1).nnedi3(field=-2)
max = bob1.mt_logic(bob1.selectevery(1,-1),"max",U=3,V=3).mt_logic(bob1.selectevery(1,1),"max",U=3,V=3)
min = bob1.mt_logic(bob1.selectevery(1,-1),"min",U=3,V=3).mt_logic(bob1.selectevery(1,1),"min",U=3,V=3)
clamped = bob2.mt_clamp(max,min,0,0,U=3,V=3)

x1 = bob1.selectevery(1,-1)
x1 = merge(x1.selecteven(),x1.selectodd())
x2 = clamped.selectevery(2,-1)

stackhorizontal(x1,x2)

return(last)

What happens: the blending basically can be undone/reversed by applying vertical sharpening ["sharpen(0,x)"] to the interlaced source. However this can't be tweaked to a "sweetspot strenght", it's always either to weak or too strong. Therefore a full sharpen(0,1) is used, which definetly is overdoing the reversal. Then the resulting pixels are clamped by [min,max] range of the original input (despite the blending, the valid pixel range is present in the source, and can be used to limit the overshooting of the blend reversal).

"x1" is more or less "the best" that you can get from this source without any sophisticated filtering, by blending the corresponding bob-frames together.

"x2" is the result of blend-reversal plus clamping.

Some high-frequency trails from the blending are remaining, but I don't see a way to avoid that.

Thanks to your script, the video is now (almost) free from blending.
:thanks:

[manono]

Thanks from me as well. Looking at it again I now see that there's no NTSC in there, which is what I had thought that kind of extra blending implied. While the left side of the StackHorizontal doesn't seem to give a better result than a simple NNEDI3(Field=-2).SRestore(FRate=25) (or even NNEDI3(Field=-2).CDeblend().BlendBob()), once you throw in the Clamp it improves it quite a bit. Now all I have to do is figure out how to adapt it for my own screwy blended sources. Thanks again.

[johnmeyer]

This will be very useful to me. Thank you!