Ghosting appearance after deinterlacing

[dragon_warrior]

Well, I have some music videos in MPEG-2 format that I want to encode to h264 for some reasons. One of which is not easy to deal with, as it contains both telecined and interlaced materials, however, those interlaced frames cannot be reverted to progressive properly by such good bobber like QTGMC.

Here are some screenshots showing 2 consecutive frames passed through tfm(pp=0) and different deinterlacers, say, vinverse and qtgmc. (Click on the thumbnails to view the full-size images)



If any of you don't mind giving a look into the source, then here is a 10-sec sample clip: http://www.multiupload.com/NHME94RX5G

Any comments would be greatly appreciated. Thanks

[Shirtfull]

Blended frames

[Guest]

Originally Posted by Shirtfull

Blended frames

Nope, they're blended fields.

[Shirtfull]

Not when more than two fields are involved

[dragon_warrior]

Thanks for all the comments. Personally I think the video is at its native framerate, but how come it appears to exhibit kind of blended fields/frames, which are the results of framerate conversion?

[Guest]

Originally Posted by Shirtfull

Not when more than two fields are involved

Two fields are not involved. Step through the fields and you will see single blended fields.

[manono]

Originally Posted by dragon_warrior

Personally I think the video is at its native framerate,

If you're saying the native framerate is 29.97fps, then that's not correct. It was shot on film and edited by idiots. It can't be unblended because in most of the frames with blends, both fields are blended. There's not even much you can do with the car part as it's already missing frames and plays jerky. I don't think you can improve much over a simple:

TFM()#with a better deinterlacer, if you like
TDecimate()

But maybe someone else has a better idea.

[dragon_warrior]

Originally Posted by manono

If you're saying the native framerate is 29.97fps, then that's not correct.

Yeah, my bad. The video was originally 24p telecined to 30i, but the process is irreversible due to the presence of blended fields. Well, as you said, there's nothing I can do with this source, even a qtgmc followed by srestore could hardly produce any noticeable results. I'd better skip it and move on to the rest on my HDD.
Thank you all!

[jmac698]

I think that can be fixed, these are called double-blends. There's 4 fields in a row but anchors before the start and at the end, to work your way inwards to subtract the blends.
Don't give up - can't do it right now, I've got a long list of things...

[dragon_warrior]

Originally Posted by jmac698

I think that can be fixed, these are called double-blends. There's 4 fields in a row but anchors before the start and at the end, to work your way inwards to subtract the blends.
Don't give up - can't do it right now, I've got a long list of things...

Yo, I'm so glad to hear it. Thanks mate

[jmac698]

I've recovered it, but the blended frames look bad. It's not totally done, I'm having a problem with wrong field order, but the end result should be smooth motion and no blends, but artefacts for those two frames. I'm also finding srestore crashes with anything else so I had to separate to two passes.

[jmac698]

Try this, run the two scripts p1even and p1odd and save to lossless (ffdshow ffv1 is one of the best codecs). Then run p2. Scroll to frame 130, you can see the doubleblended reconstructed fairly well. The other sections aren't blended and are now screwed up as a result of processing. You could probably trim sections to finish it now, but I'll see if I can do it in one script. Anybody want to make suggestions feel free, I'm not good at this type of thing!

#p1even.avs
dir="E:\project001a\double blend\"
MPEG2Source(dir+"x.demuxed.d2v", cpu=0)
assumetff
separatefields
selecteven
#pp2 and pp3 are much better than other modes
srestore(omode="pp3.blur(1)", cache=5)

#p1odd.avs
dir="E:\project001a\double blend\"
MPEG2Source(dir+"x.demuxed.d2v", cpu=0)
assumetff
separatefields
selectodd
#pp2 and pp3 are much better than other modes
srestore(omode="pp3.blur(1)", cache=5)

#p2.avs
dir="E:\project001a\double blend\"
e=AVISource(dir+"p1even.avi")
o=AVISource(dir+"p1odd.avi")
interleave(o,e)
assumefieldbased
weave
telecide(guide=1,post=2)
decimate(cycle=5,quality=0)
#works for blended sections, i.e. frame 138-
#in other places, it's deinterlacing and causing (near) dups

[dragon_warrior]

thanks very much, i'll give it a try