De-interlacing or not for this anime (PAL version) ?

[Music Fan]

Hi,
someone sent me this manga extract coming from a PAL dvd, it looks like a bad Ntsc to Pal conversion ;
http://www.mediafire.com/download/6w6h5qbgmr7o9m2/VIDEO_TS.zip
We found some scripts that help but none totally removes blending.
Yadif or QTGMC followed by SRestore helps, or also TFM and SRestore.
I prefer to avoid de-interlacing when possible thus I hope TFM can do the job, this script is not bad but not perfect ;
MPEG2Source("P:\VTS_01_1 honor pull.d2v")
#
assumetff()
#
Interleave(TFM(field=1),TFM(field=0))
#
SRestore(frate=23.976)

There are so many functions in TFM, I don't know what to do and I'm not familiar with anime and IVTC, so thanks if you can help.;)

[MWilson]

From what I remember (could be wrong), isn't PAL bottom field first?

If so, that may very well be your problem.

[Music Fan]

isn't PAL bottom field first ?
It can be either BFF or TFF, but most of my PAL dvds are in TFF.
When I convert DV to mpeg-2 for dvd, I let it in BFF.

edit : with the script I posted above, frames 109, 111, 285 ... among many others, show blending with the previous frame. Same problem if I write assumebff instead of assumetff.

[Music Fan]

I already tried bff (I was editing my previous post when you posted yours), it does not help. I guess the (v) in your script is an error.

[MWilson]

Sorry, yeah it was an error. Deleted my post.
:o

[FranceBB]

If it's a 23.976 source telecined in 29.970 NTSC which has been converted to 25 PAL by blending, it means that it has probably been converted via hardware assuming that it was interlaced instead of telecined. Unfortunately, even though there are many filters to fix blending issues, it really depends on your source.
Try these filters and play with settings:

removeblend(threshold=10, dthresh=0.03, pixels=50, info=false, interlaced=true, mode=0, show=false, cthresh=0.05, bthresh=1.1, othresh=0.8, decomb=true, pt=16)

Unblend(threshold=5, dthresh=0.03, show=false)

deblend()

FixBlendIVTC(sbd=true)

p.s add "SRestore(23.976)" at the very end.

Good luck!

[thecoreyburton]

I'm definitely not an expert like a lot of the other members here, but I live in a PAL region. I've had to deal with hundreds of PAL DVDs and it's never a pleasant experience figuring out exactly what steps to take to fix the source, but I'm yet to run into a problem I haven't eventually solved.

I had a quick look at your source and noticed a few things. The good news is that traditionally field-blended sources have visible combing on almost every frame. They're easy to spot with a quick bob because every second frame is a blurry, blended mess. This isn't the case with your source, which I'd consider a blessing. I did a quick pass of QTGMC on the very slow preset (you're welcome to use whatever settings you prefer) followed by SRestore on settings I use for somewhat troublesome sources and the results look good.

QTGMC(Preset="Very Slow")
SRestore(FRate=23.976, Speed=-1, Thresh=12)

The good news is that after applying this, the source becomes watchable. The motion is about as smooth as you could hope for and the blended frames are removed to a point where they're barely noticeable. Given the smooth results, I removed SRestore from the equation and had a look at the deinterlaced source at two sequential moments in time:

http://i63.tinypic.com/314sfio.jpg
http://i63.tinypic.com/51xze1.jpg

The problem becomes immediately apparent. SRestore has two frames to choose from for the final output and both of the frames have some degree of blending. This is something I usually come across on interlaced sources that have been encoded as progressive material at some point after the interlacing was added. This usually means that the luma for each moment in time is stored on alternating lines, but the chroma is stored on the frame as a whole, making it impossible to restore to it's original state.

Other more experienced members may be able to better explain my point or may even have information that completely counters what I've said, but I sincerely think this is the case with your source. The counterpoint here is that if you do apply the script I've quoted earlier in this post, the result is very smooth, completely watchable and unless you're specifically going frame-by-frame and hunting for blended frames, you won't notice them on playback.

Best of luck!

[Music Fan]

Thanks guys.
Actually the result with de-interlacing does not seem better than with TFM (with the script I gave), and it's much more slower than TFM.
But I hope that TFM experts know special settings that may help.

@ FranceBB : the functions you mention seem exceeded by SRestore. Anyway I tried FixBlendIVTC(sbd=true).SRestore(23.976) but it didn't help.

[FranceBB]

@music fan... they didn't do anything because as theCoreybutton said it's another kind of issue. (I didn't download the video at the very beginning as I don't have a good connection at all this very moment). Anyway, I had the same issue on some old transfer made by TV Tokyo (shugo chara). It indeed is an interlaced anime which has been converted by assuming it was progressive, so the chroma expanded in the fields and disaster happened. Unfortunately, I tried to fix it in several ways but I didn't manage to get a good result. (It was about a year ago). I tried several different approaches: different deinterlace methods, different deblending filters and even to desaturate and to apply some strong denoise to the chroma only, but none of them helped... at the very end, I did something "horrible" and I deinterlaced them, removed duplicates to bring everything to 23,976 and then blended them to 29 to "cover" chroma artifacts, hoping that they wouldn't have been noticed by users during the playback. Even though my transfer were really bad, I don't actually recommend that approach. A colleague of mine, instead, did something crazy yet powerful (but incredibly time consuming). He separated chroma and luma; thrown away the chroma, he took the luma from the "damaged" source, the chroma from a Transport Stream (another source) and he made them fit perfectly, then merged them together to get a good final version.
That was crazy, yet it worked.

[Music Fan]

Ok thanks for the explanation, but are you sure my video is coming from an interlaced anime ?
If yes, I guess Srestore shouldn't be used.

[FranceBB]

@music fan... no it's not, it's native in 23.976.
Probably the "workflow chain" was:
1) native 23.976 fps video telecined to 29i NTSC.
2) 4:2:2 planar (yv16) to 4:2:0 conversion assuming it was progressive (chroma expanded in the field).
3) then they finally converted to PAL assuming it was interlaced instead of telecined (bad things happened again: 29i images blended together to get 25i).


You have to use SRestore to bring back everything to 23.976, which is the original frame rate, and it will remove blended frames added during the PAL conversion, but you can't fix the chroma expanded in the field. ;)

[Music Fan]

Ok thanks, thus my first script with TFM was actually enough, it seems de-interlacing isn't really better.

[FranceBB]

Well, it really depends on what you wanna do.
Since deinterlace introduces a light quality loss (no matter what), you shouldn't generally deinterlace unless you have to. ;)
At the very end, it seems you did the right "choice" since the very beginning xD

[Tormaid]

That's Eva, right? Why on earth are you working with PAL DVDs when the Blu-rays are out?

[Squarealex]

That's Eva, right? Why on earth are you working with PAL DVDs when the Blu-rays are out?

Hi.
Because they are very expensive (350€-400€), they are available only in Japan, the movie are not remastered in 1080p, difference between DVD PAL French and Blu-ray is not huge (see attach files) compared to TV series and I don't have a bluray drive in my PC.
That's why I use the French DVD, (with the DTS sound) which is fairly close to bluray sources (film grain less & little less detail on DVD).
Yes, I upload this video. I asked MusicFan see if some member of Doom9 was more or less successful in solving the problem of ghosting PAL. Because I'm not very good in English and I do not have an account yet. :o

Final render :
http://forum.doom9.org/attachment.php?attachmentid=15521&stc=1&d=1471038180
http://forum.doom9.org/attachment.php?attachmentid=15525&stc=1&d=1471038538
http://forum.doom9.org/attachment.php?attachmentid=15523&stc=1&d=1471038180
http://forum.doom9.org/attachment.php?attachmentid=15524&stc=1&d=1471038180