Image "Swimming" when deinterlacing camcorder footage

[P-Rock]

I'm in the process of transcoding a bunch of DVDs of old home movie captures for use on a Plex server. The first half of DVDs contain footage from a camcorder that recorded directly to VHS tapes, circa 1989 to 1998. The second half are filled with footage from VHS tapes which themselves were captures of DV cam footage (from about 1998 to 2005). I know this capture approach was not optimal for a variety of reasons, but I wasn't involved in the initial captures, so this is what I'm stuck working with. All footage is interlaced, but only the second set looks terrible when any sort of de-interlacing is applied to it. I tried every de-interlace option in VLC, as well as playing with interlace options when I transcode in Handbrake, and every single one of them looks terrible--the non-deinterlaced footage looks much better. The reason is that whenever there's any sort of fast movement in the camera short (a quick pan or someone moving quickly), the de-interlacing makes the image look like it's "swimming":


[Attachment 54247 - Click to enlarge]

If I turn off de-interlacing, it looks like this:


[Attachment 54248 - Click to enlarge]

Still not great, but considerably better than the first image. My question is: how do I de-interlace this footage so that it doesn't look terrible?

According to VLC, the original files are .VOBs encoded with MPEG-2 at 720x480 at 29.970029 FPS. Since it's about 4-5 VOBs per DVD, I merged all those into a single MKV file, which is what I'm doing my transcoding on. These files reports as MPEG-2, 720x480, and 59.941257FPS in VLC (not sure why the framerate doubled). De-interlacing is a problem for both the original VOBs and the merged MKVs.

[jagabo]

Those artifacts are usually the result of interlaced video being resized vertically without consideration for the interlacing. The two fields are no longer separate images and can't be deinterlaced properly. You should upload a short sample of your MPEG 2 video for further analysis. Use a program like DgIndex or Mpeg2Cut2 to select about 10 seconds with the problem (medium speed horizontal panning shots are good for this). Upload the sample here.

[P-Rock]

Here's about 10 seconds of video. I cut this from one of the MKV merges of my VOB files, not the original VOB file itself. I'm assuming this won't make a difference, but if I does I can re-upload a sample from the original VOB file.

[davexnet]

If that the way it looks in the mpeg-2, then something went wrong during the capture or in the DVD creation

[manono]

I'm assuming this won't make a difference...

You assume wrong. jagabo didn't suggest DGIndex or MPEG2Cut2 for no reason at all. You create extra work for those willing to help.

As jagabo speculated, it was resized while still interlaced at some point, perhaps from an original 352x240.

Using this script:

MPEG2Source("test.d2v")
BilinearResize(360,240)
separatefields().SelectEven().AssumeFieldBased()
Weave()
QTGMC()
BilinearResize(720,480)

I got this (which can probably be improved upon):

[jagabo]

Yes, the MPEG 2 video was resized vertically while interlaced. My guess is somebody removed the head switching noise at the bottom of the frame and resized what was left to refill the screen. About the best you can do with video like that is fully blend the two fields together (leaving what looks like double exposures). Here's what I did in AviSynth.

Mpeg2Source("test video.d2v")
Spline36Resize(width,height/2).nnedi3(dh=true)
deblock_qed(quant1=50, quant2=50) # reduce MPEG blocking
Crop(6,0,-10,-0)
QTGMC(InputType=2) # reduce edge noise, mosquito noise

[P-Rock]

Thanks! This looks pretty good. Is there any way to tell what the original resolution was? Like I said the before, the VOB shows a resolution of 720x480 in VLC.

Jagabo, how much fidelity am I losing by doing that blend you're suggesting? It's possible that I may try to get all these videos re-captured from their original source, and if I do it this time, I want to make sure it's done right.

[jagabo]

Jagabo, how much fidelity am I losing by doing that blend you're suggesting?

You're losing a little vertical resolution, losing half the temporal resolution, and getting frames that look like double exposures -- which is essentially what they are.

The original video was almost certainly 720x480. I suspect someone enlarged it to 720x540 (a 4:3 frame size) then cropped away 60 lines (some from the top, some from the bottom) leaving 720x480. If you have them re-captured you should get 720x480 (or 704x480), 29.97 fps, interlaced. That's the international standard for NTSC capture. That can be encoded directly to MPEG 2 for DVD (if you filter you need to be sure to use interlace aware filtering). When done properly you will have two clean fields in each frame. You will not see comb artifacts on playback, you won't have those scalloped edges, and you won't have images that look like double exposures. If you don't want DVD, but rather progressive video for upload to youtube (or whatever) you should use AviSynth and QTGMC to deinterlace the video.

johnmeyer has referenced some things he's done with that sort of video. But I don't think it will work well with your source:

https://forum.videohelp.com/threads/398084-Field-Blending-Fix-in-AviSynth-(Script-Advi...s)#post2589431

[P-Rock]

johnmeyer has referenced some things he's done with that sort of video. But I don't think it will work well with your source:

https://forum.videohelp.com/threads/398084-Field-Blending-Fix-in-AviSynth-(Script-Advi...s)#post2589431

Why exactly do you think it wouldn't work well with my source? Is it because the problem is fundamentally impossible to "properly" fix short of doing a recapture of the VHS source?

[jagabo]

The basic technique he uses is to cut the video into horizontal bands then reassemble them (with some motion interpolation). The source he was working with had only 21 bands with a little fuzziness at the edges. Your video has a lot more bands and more fuzziness. And your source has terrible DCT blocking from bad MPEG 2 compression. I don't think you will get good results.

Since you have the original tapes re-capturing them is almost certainly the best option.