Which passthrough devices straighten out the bottom of VHS signals?

[Brad]

I've seen some recordings from VHS where the "head switching noise" (as I've seen it called) is perfectly corrected except for a few lines maybe 8-10 lines up from the bottom.

I believe these were done using DVD recorders, but which ones? Any TBC that corrects these lines as well?

Meanwhile, this image is from an official DVD release and it has errors at both bottom and top. Pretty sad.

[robjv1]

There is no TBC or device to correct these errors directly. The 'head switching noise' as it is called normally falls into the overscan of a TV and is only visible on non-overscan devices such as computer monitors or on televisions where the overscan has been turned off in the settings.

The variablity in the amount you see is due to a possibility of a few different things. Typically it is the VCR -- some hide it by masking or cropping, either blacking out or cutting off the bottom of the image. Some VCRs seem to have more overscan than others -- the one in your image doesn't look too bad to be honest.


Anyhow, if you have a playback device where it is showing up, you have a few options.

1. Mask it out (don't crop!): An easy fix with VirtualDub. The disadvantage here is that you have to re-encode the footage to do it, which will cause at least slight deterioration in the source.
2. Deal with it on playback: If you're using an HTPC or a computer, VLC for example will allow you to add a mask to your video on playback, so you don't have to re-encode the video. As long as you computer is fast enough to use VLC, you should be gold.
3. Hide it with a black subtitle: You'll have to search the forum or I'll have to find a link to it after work, but a clever way to add a mask is to add it as a subtitle that covers the head switching noise. Once again, no re-encoding of the video itself, but it can take some fiddling around with to get it right

[Brad]

The variablity in the amount you see is due to a possibility of a few different things. Typically it is the VCR -- some hide it by masking or cropping, either blacking out or cutting off the bottom of the image. Some VCRs seem to have more overscan than others -- the one in your image doesn't look too bad to be honest.

I said nothing about varying amounts. I talked about videos I have seen where the head switching noise looks almost like normal video -- except for some lines midway through. I'd like to know how to accomplish that with my own transfers.



I would post a video clip but my upload speed is soooo sloooow.

Maybe it has nothing to do with the capture end and everything to do with the player, I don't know. But I think the guy who was writing a software TBC also found that he was able to un-stretch (or whichever term) this area of the video almost back to normal.

[TimA-C]

Read what robjv1 said as he's spot on with his explanation of what you're seeing and what you can do about it.

[jagabo]

I talked about videos I have seen where the head switching noise looks almost like normal video -- except for some lines midway through. I'd like to know how to accomplish that with my own transfers.

Get a different VCR.

[Brad]

Sure. Which?

[robjv1]

You can't fix it with software either in terms of straightening it out as you said, because it isn't the case that the picture is twisted or contorted. There isn't any information to recover -- it's not a TBC error. No VCR is going to "recover" it either, it'll just hide it. You'd be better off doing it yourself though, because even VCRs with the reputation for having little overscan (see GShelly posts on the JVC SR-W5U) do vary from tape to tape.

[jagabo]

No VCR is going to "recover" it either, it'll just hide it.

It depends on the individual recording. On some there is more signal there than others. Some players are better at playing that signal. I think it has to do with how far around the drum the tape is wrapped (both during recording and playback).

[robjv1]

No VCR is going to "recover" it either, it'll just hide it.

It depends on the individual recording. On some there is more signal there than others. Some players are better at playing that signal. I think it has to do with how far around the drum the tape is wrapped (both during recording and playback).

I agree, that's a good distinction to make, there are small improvements to be made even if finding one deck that will play everything you have perfectly is futile. Another thing I've noticed is that I have a couple of different decks that are of the same model #, one I actually used to record some of my tapes on and another I bought years later. The one I recorded my tapes on seems to play those tapes back with less distortion -- it could just be coincidence, but it's interesting.

[sanlyn]

Hmm. Learning more about capture, getting better stuff (there's a lot out there), and working on your 'puter...offers many possibilities:


[Attachment 15789 - Click to enlarge]

(Couldn't resist. Sorry. Too much coffee today.)

[Brad]

You can't fix it with software either in terms of straightening it out as you said, because it isn't the case that the picture is twisted or contorted.

http://forum.doom9.org/showthread.php?p=1534487#post1534487

Oh, and that 'head switching noise' that everyone crops? No need to, I can recover quite a bit of it.

He posted screenshots and videos, but they are dead now. Where were you last year?

[Brad]

More evidence to counter the claim that head switching noise is just garbage video that can't be massaged back into fine form.

http://www.youtube.com/watch?v=sB5w2cXU-jU

The couple lines that remain unrecovered are ridiculously high in this video. Lines below it only suffer from chroma issues.

[Cornucopia]

No, vaporeon800, the first link talked about a post-capture, software/algorithm-based TBC for correcting interline jitter. Head switching is MORE than just interline timebase jitter, it's actually often 2 different signals placed onto the same line with WAY more jitter than what would be normal, so a soft TBC would have to work on both the normal Interline Jitter, plus it would have to work with different variables on the segment encompassed by the head switching. and even then, you'd have a horizontal portion THAT WOULD NOT BELONG to that original line. Could be from the previous line, or 2 or 3, or could be from the next line, or 3 or 4. And it would vary RANDOMLY from line to line and from field to field in strength, positioning, etc. So there's no real way to account for that and fix it. Hence: it's garbage.

That 2nd link is not "evidence" of anything except that the deck shown must have better interchange than most. The isolated problem mid-screen is likely evidence of a problem with the tape (wrinkle, etc) not head-switching.

Scott

<edit>...so the answer to the original question is: NO device does.

[Brad]

You are directly contradicting the author of the software TBC based on what evidence?

[robjv1]

I'm not seeing anything other than theory and testing in the first link. It is promising work for a software TBC in general, but not all of the answers are there and it's not a useable general purpose solution for this problem, and certainly not for footage that is already captured.

As far as the second link, I have captures from my decks that look exactly like that in terms of the noise, so that is certainly no evidence that any software fix is at work there.

[Brad]

You're not seeing anything else because the image and video links are down. Is his word really that unreliable? He says there is no need to crop the head switching noise because he can recover quite a bit of it.

Here is another.

http://forum.doom9.org/showthread.php?t=162953

Finally, the bottom lines of the video look normal. In fact in the orignal, it was
badly slanted and had head switching noise. This part is normally cutoff, but now we
don't need to!

Regarding the second one, fair enough, I believe you.

[robjv1]

No, I wouldn't go that far. If we're still arguing about the general principle of whether or not it is possible, I stand corrected on making such a broad statement and defer to jmac698 or Cornucopia as they are much more technically inclined in the area. I don't doubt from the other knowledgeable replies in this thread that it is possible to recover some of it, particularly in test cases, but there is no question that it's work that is in a research stage and there are parts of the signal which are not recoverable. I suppose if a method came along that could successfully treat a few of the topmost scanlines in every general video, then it'd give people less to mask out at least.

At any rate, there isn't a general purpose solution for it and there may never be. Although I don't personally see the need for masking it out in the first place, since it's in the overscan.

So just leave it in for now and see if it can be addressed to your satisfaction later, or ask Jmac to take a look at it, I know he takes samples.

[Cornucopia]

It should be masked out precisely because, even if it is in overscan, it can strongly affect bitrate efficiency.

You've all got to remember, what we're dealing with here is the result of MECHANICAL SLACK and RANDOMNESS. That's what head switching is: the heads "catch hold" of a signal for the length of the tape diagonal until the next head catches hold of the next diagonal strip. But there's always error at the beginning of the "catching hold" where the device has to work out (through azimuth phase-feedback alignment) whether it's actually got the correct track or not. And OFTEN it doesn't have quite the correct track, which is why you'll see the "tearing" and jumping around of a seemingly different signal in the area of the head switching.

I believe that such "software TBCs" can do a good deal to improving/minimizing the problem, but they won't ever ELIMINATE it. I read through most, if not all, of that thread, and it doesn't contradict what I'm saying. He is being overly optimistic when claiming his method will take care of all those problems, but his examples (and yes, I got to see them) bear out that he doesn't actually have the magic wand you're expecting it to. You need to get over it. There are just some random-type errors that cannot be accounted for. If there were, we'd probably have "restore resolution to what it MIGHT have meant to have been" filters.

Scott

[2Bdecided]

A decent VCR with a built-in line TBC can, with a bit of luck, deliver stable video in this region (i.e. the last 8 lines). However, there are always a couple of lines missing during the head switch: 2 lines, sitting about ten lines up from the bottom. I've had no luck with video inpainting to repair them. Since those two lines are gone, I can't see much use in grabbing the final 8 accurately. Hence I choose the VCR which gives the best results on the majority of the picture, rather than the VCR which gives the best results on the last 8 lines.

FWIW I've seen one VCR (and I mean one out of maybe 120 I've played with over the years) that put the head switching noise at the top of the active picture. No idea how or why it did this, and it died before I had chance to investigate further. Maybe this was a symptom of it dying.

Cheers,
David.

[Brad]

Isn't each line in a different field? NNEDI on each?

[2Bdecided]

You could use NNEDI as an interpolator (it might be two lines per field, so it would get very messy using it in this way).

You missed the line "with a bit of luck". At best this is luck. I'll see if I can find a captured sample to show the best I've seen. I might not have kept it though. It's on an external HDD somewhere!

Cheers,
David.

[Brad]

Hmm, darn. The images that I posted have two garbage lines per field and they represent the best I've seen besides that PAL YouTube video that has been discounted.

[jagabo]

If the bad lines don't move up and down it's pretty easy to fill them with data interpolated from the lines above and below. Individual fields of the original on the left, bad lines interpolated (via AviSynth) on the right:

[jagabo]

By the way, if you want something to play with, someone just posted some PAL Hi-8 caps (DV AVI) with just a few lines of head switching noise near the bottom of the frame (like the sample image above).
https://forum.videohelp.com/threads/354425-Hi8-capture-using-Digital8-camcorder-Edge-co...=1#post2228842

[manono]

If the bad lines don't move up and down it's pretty easy to fill them with data interpolated from the lines above and below.

I can use something like that. How'd you do it?

Off topic, but did you see the post at Doom9 where the guy updated/modded a lot of the frame interpolators using MFlowInter and MVFlowFPS:

http://forum.doom9.org/showthread.php?p=1620974#post1620974

[jagabo]

If the bad lines don't move up and down it's pretty easy to fill them with data interpolated from the lines above and below.

I can use something like that. How'd you do it?

I stacked the line above the bad lines and the line below the bad lines into a two line image, stretched it to four lines, cropped away the top and bottom lines, then overlaid the new two lines onto the original image (twice, once for each field).

ImageSource("VTS_01_1004.jpg")
AssumeTFF()
SeparateFields()
top=SelectEven()
bot=SelectOdd()

# fix top field
p1=Crop(top,0,235,-0,1)
p2=Crop(top,0,238,-0,1)
StackVertical(p1,p2)
PointResize(720, 4).Blur(1.0).Blur(0,1.0) # cheap interpolation, or use another resizer
Crop(0,1,-0,-1)
newtop=Overlay(top,last,0,236)

# fix bottom field
p1=Crop(bot,0,234,-0,1)
p2=Crop(bot,0,237,-0,1)
StackVertical(p1,p2)
PointResize(720, 4).Blur(1.0).Blur(0,1.0) # cheap interpolation, or use another resizer
Crop(0,1,-0,-1)
newbot=Overlay(bot,last,0,235)

Interleave(newtop, newbot)
AssumeTFF()
Weave()

For YV12 video you'll have to modify this because you can't have single pixel thin lines. And, as I recall, BilinearResize() etc. don't work with two line tall videos. So you would need to modify the script to use them.

[manono]

Thanks. I was wondering if you had done it with some sort of a delogo filter.

[Cornucopia]

A bit of a cheat, jagabo, using interpolation & inpainting. Though a nice one.

Heck, if the OP wants to use a Svhs vcr + line TBC followed by a fullframe TBC then capture uncompressed and use that SoftTBC script and your infilling/interpolation script, he can probably get 99.99% of it cleaned up (rarely do those few remaining lines stay put), and more power to him. But its still not a complete fix magic wand. And since that stuff is mainly only in overscan areas, it certainly is more straightforward and quicker to just mask the garbage. There is that law of diminishing returns rearing its head again.

[jagabo]

A bit of a cheat, jagabo, using interpolation & inpainting.

But if that's all you have to work with...

[Brad]

Cornucopia: the only thing I'm missing from that equation is the SoftTBC script.

This thread isn't about what's good for Joe Q. Public but for Brad D.R. Private of Vancouver who has been on VideoHelp for nearly 12 years.

Diminishing returns is my middle name...