VCR Capture with NTSC->PAL Conversion

[manono]

Hi-

I and others were trying to help a guy over at Videohelp.com with a video he captured from VCR. He captured it in PAL DV format and posted a short clip from it that he had reencoded to XviD with no filtering, in order to keep the size down. It's 1.36 MB in size.

It's an animation that's been put through a couple of conversions, I think, with a final NTSC to PAL. He wants to make it progressive again on the way to making a PAL DVD from it. After making a quick diagnosis based only on his description, I later downloaded it and discovered that it's really messed up. I suggested he ask about it here, but he's only now registered and has to wait out his 5 days, so I said I'd post for him.

I originally suggested this script:

AssumeTFF() #if that's what it is
Telecide(Guide=2)
Decimate(25)

which works, sort of, except where the field reversals take place. He told me later via PM that while the entire video is greatly improved with that script, it's nowhere near good enough.

For those that may get the clip, I have some questions:
1. Would he have been better off to have captured it as an MPEG, so that DGIndex's FixD2V could straighten out the field reversals? If it were in MPEG format, could FixD2V correct it?
2. Short of that, is there an IVTC that can work with the sometimes back and forth movement to decimate it back to 24fps, so that it plays smoothly without duplicate frames and without dropped frames. Ignore the ghosting in every single field.
3. Would his best bet be just to leave it alone and encode for PAL DVD as is? That is, can nothing be done? I have seen similar things on NTSC DVDs, and am usually at a loss for what to do.

Any other observations and advice will be much appreciated. Here's the clip:

http://users.tkk.fi/~akorhola/weird.avi

Right-click on it and "Save Target As...". And here's the original thread:

http://forum.videohelp.com/viewtopic.php?p=1451917

Thanks

[Mug Funky]

could they be dropped frames? or even dropped fields?

a straight assumebff().bob(0,.5) showed it wasn't too bad, and when it jittered, it looked to be from a whole, repeated frame.

perhaps EEDI plus blendbob could help sort it out?

[edit] EEDI2 and blendbob did indeed help, but at what cost? :) the source is hella blurry to start with.

btw, look at frame 43 - it shows signs of a really poor resize (like the crappy bilinear that premiere does on low quality mode - not quite pointresize, not quite bilinear, all bastard). looks like some kind of IVTC was done, then frameduping up to 50 fields, then re-interlacing.

i've seen worse :) bootlegged VHS's are not fun (don't worry - no rule 6'er here, it's all legal). only good for audio, really. the video on these things needs to be killed with fire.

[edit 2]

try this:assumebff()
separatefields().eedi2(field=-2)
blendbob()
lanczosresize(704,576-16,22,7,720-34,576-(20))
addborders(8,8,8,8)

[edit 3]

btw, the chroma ghosting is most likely from the DV encoder used (a camera? a DV box? i find a lot of stuff shot on a DV cam will have similar effects. possibly you could fix it by messing with cedocida's forced chroma sampling options).

[manono]

Thanks for taking the time, Mug, to look at it and work on it a bit. Your script does improve it. It still has a couple of jumps in it, and only the guy, after applying it to the entire video, will be able to tell if it's good enough. Pans like that are kind of a torture test, and maybe there's not a whole lot of them to show up the video's deficiencies.

I've never seen crop/resize done with that notation. For me anyway, when I used the script as you wrote it, it gave a bunch of brown and black garbage along the bottom maybe 60 or so pixels thick. Heck, I was surprised it even opened. That's OK though, as I or he can do the crop/resize/addborders the "old fashioned" way.

I'll go PM him to find out exactly how he capped it, and explain the script to him, and then find out what he says. Thanks again.

[Mug Funky]

ah, it doesn't look like that on my end :)

try avisynth 2.5.6, but i don't think the notation has changed.

unless you swapped the "lanczos" for "bicubic", in which case it's taking my cropping args as a and b parms for the bicubic filter..

[manono]

Hi-

No, I just copied and pasted your script, after adding the required .dlls. I'm using an AviSynth ver 2.5.6 from Oct 7,2005. But it doesn't matter, I got the garbage cropped away and the video centered.

He says he captured it using an Aver capture card and Virtual VCR. No camcorder involved. He said that although there were a few initial dropped frames, as far as he knows, the video itself doesn't have any. He also said that he's pretty sure, but not positive, that it's Australian in origin, which, if true, might explain the double conversion as PAL2NTSC2PAL. I may have the name of it next time, to see if it rings any bells with you. He'll also check out your script later on today.

[Revgen]

Perhaps you guys could try using or tweaking Didee's DeHalo Alpha script. It was orginally created to get rid of ringing but it may be useful to get rid of the faint blends. With alot of tweaking of course.

[manono]

Hi-

I know the name of the video now. It's an animated version of King Solomon's Mines that seems to be of British origin, as near as I can tell:

http://www.imdb.com/title/tt0462384/

Since he's in Europe himself, and the VCR tape he captured is PAL, I don't know why it has those conversions on it. Anyway, he's reasonably happy with your script, Mug Funky, and posted a small sample in the thread over at Videohelp.com. It's 1.66 MB, and this is it, if you're interested:

http://forum.videohelp.com/images/guides/p1452315/better.avi

Revgen, thanks for the advice about removing or lessening the blends using DeHalo. He is interested in cleaning it up some more, either with an unblender such as Restore24 (most of the fields aren't blended, as they were in the earlier example), or some filter such as the one you suggested. Thanks.

[Mug Funky]

if it's old, it may have been peculiarblended (though it didn't seem so). conceivably an analog setup could have produced those blends though (a prism-style telecine with a CCD instead of a tube camera maybe? clutching at straws here...).

[ajk]

Hey guys,

And thanks for the help everyone, I'm the one who came up with this capture. Had to wait the 5 days before replying here myself (thanks manono!).

Upon closer inspection of some other tapes I found a second film that had similar blending issues, albeit much less noticeable since it was a "real life" film. Most tapes have been fine so it can't be due to my current VCR / other setup, so what I'm guessing now is the old VCR used for taping these two films had this blending "feature". Hopefully it won't come up too often in the more interesting recordings I have left.

Anyhow, thanks to all the tips&scripts here and on the other forum I have managed to fix almost all of the interlacing mismatches and also filter out the worst noise, so the quality right now is already very watchable.

While fixing all of the blends would be impossible without manual retouching, I think up to 70-80% of them could be removed if there was a way to replace blended frames with clean ones. Since a lot of the time the film only uses around 12, 8, or even fewer frames per second there are often good ones that could be used. Only some of the pans and other faster moving scenes use all 24 fps so those would have to remain, but in those scenes the remaining "motion blur" isn't that bothersome anyway.

Now, since the clean frames follow no specific pattern some way to automatically detect and replace the blended ones would have to be devised. The problem is that the difference between a clean and a blended frame is often quite small, and so far I haven't found a working way to go about this. Does anyone happen to have more experience in this sort of scripting or shall I just let the blends remain?