Difficult Interlacing

[ZacharyGeorgeNN]

I'm trying to rip this very rare DVD, but the interlacing in it looks like nothing I've seen before. Even after using QTGMC, it isn't removed, but instead looks like a set of Venetian blinds almost. What would be the best way to handle this?

With interlacing:

http://i.imgur.com/hZkJ3e0.png

After QTGMC:

http://i.imgur.com/qcEBdlT.png

[johnmeyer]

You have to upload a few seconds from the original. The second picture looks like interlaced video that was badly resized.

[luquinhas0021]

Try eedi3 + nnedi3

[jmac698]

None of these will work, this is a combination of clean and blended interlacing, in a pattern which repeats every 12 lines.
The pattern shown in this frame is like this:
line mod 12,effect
0,this field
1,next field
2,blend
3,blend
4,next field
5,this field
6,next field
7,this field
8,blend
9,blend
10,this field
11,next field

I could write a script to fix that now, but there's only 4 of 12 lines that are clean, each field would be 1/3 height, and the pattern might reverse on frames. You need to use unblending scripts for this. I think it's called depal? There's one brilliant one that can figure out blending patterns. I could puzzle on it and make a good guess how it was created, but it would be much easier to have a sample. Pick an area with very strong contrast and movement. His white robe against the dark background was a good choice.

Btw the robe is about 210-217, the coat behind is 99-118, and the blended areas are 150-179. I've found that the blends are about 50/50 between this and next frames.

[wonkey_monkey]

Resize it to 720x480 with something other than pointresize, then QTGMC. You'll have some blending, but no venetian blinds. What you do next depends on the rest of the clip, since as johnmeyer points out, a single frame is not much to go on.

[Sharc]

What is your playback device? TV or some PC SW player?

A sample of the original would be helpful for getting better advice.

[jmac698]

huh? It's 720x576, and he already showed what qtgmc did to it.

[Sharc]

SRestore() may help, but one would need a sample to try it out.

[wonkey_monkey]

huh? It's 720x576, and he already showed what qtgmc did to it.

To whom do you "huh?"? My suggestion is to resize to 720x480. Doing so in Photoshop and deinterlacing the odd and even frames gives a fairly clean result. There's a bit of blending of the two frames, but applied to the video as a whole it should be watchably smooth, and srestore might be able to something about the blending.

[TheSkiller]

It's interlaced video which was resized in the vertical direction. This mixes the fields partly. It's not possible to reverse this, but if you're lucky it can be helped a bit.

[ZacharyGeorgeNN]

Here's a sample of the DVD: https://mega.nz/#!k5w3EbiZ!cTHHwwdK2hja8OAcdMIp553I029CCkWwhzFom3uHvAw

The frames are also pretty screwed up, there doesn't seem to be a discernible pattern in regards to the duplicates. What would be a good script to use here to make this look at least a little better?

[johnmeyer]

Wow, that is a very strange video clip. Here's what I found.

1. Mediainfo (and other tools) report the clip as 23.976 progressive. However, I went ahead and used separatefields() on the clip and found that it actually IS interlaced, TFF.

2. The interlaced cadence is all over the map, with lots of blends in the individual fields.

3. The herring bones that appear in individual fields are not of uniform width, and many of the lines are several scan lines tall.

4. There are lots of dropped frames. You can easily see this by walking through the clip one frame at a time. You will see sudden jumps in motion.

So, as others already surmised from your single frame capture, this is going to be a very, very difficult task to get a decent result. I don't know if SRestore can deal with the uneven vertical widths of the "interlace" artifact that you are trying to remove. In addition, you will need to have some sort of dropped frame detection and replacement, something I have posted about several times before.

No easy answer to this problem.

P.S. I just looked at the footage, full speed, and it plays way too fast.

[wonkey_monkey]

Hmm, okay, it actually is 720x480, unlike the image in the original post. So the original problem was because it'd been resized to 720x576.

That said, even at 720x480 the interlacing is a mess. Maybe it's an interlaced telecine that's been encoded as progressive? But even then field priority seems to jump around...

P.S. I just looked at the footage, full speed, and it plays way too fast.

It looks okay speed-wise to me.

[ZacharyGeorgeNN]

I think some of the confusion regarding the size of the first pictures was that I saved the screencap that had loaded in avisynth. I didn't resize it, though.

[Sharc]

You may try
FieldDeinterlace(blend=true)

It doesn't fix the messy source, but the playback looks better (means less annoying) to my eyes.
You should also check how it looks on TV, finally.

[ZacharyGeorgeNN]

That worked well. It doesn't seem there's much that can be done about the duplicate frames since there's no pattern to them, but at least it looks more watchable now.

[AzraelNewtype]

That worked well. It doesn't seem there's much that can be done about the duplicate frames since there's no pattern to them, but at least it looks more watchable now.

If you're feeling lucky, dedup with the right threshold can nuke them for you without ruining the content you want. Of course, finding that right threshold will be a hilarious guessing game, and it's probably not particularly safe, but if you're willing to do a shit load of trial and error (and maybe segmenting to cut down on the problems) you can do it in a way slightly less gross than manually excluding frames. I mean, that's technically possible too, but you don't want to do it. Nobody wants to do that.

[kuchikirukia]

2 pass?

Index with dgindex. Run analysis pass with:

spline36resize(720,480)
tfm(d2v="D:\input.d2v",output="tfm.txt")
tdecimate(mode=4,output="tdecimate.txt")

then encode:
spline36resize(720,480)
tfm(d2v="D:\input.d2v",input="tfm.txt")
tdecimate(mode=2,input="tdecimate.txt",tfmIn="tfm.txt"")

Still combed to hell, but seems to be less nausea inducing.