Clean up darker horizontal lines from VHS

[RangerCat]

Hello everybody,

I'm trying to restore capture from an old VHS tapes. I already done some steps and experiments with various AviSynth and VirtualDub filters in order to restore picture. Basically I have an idea how to process the whole thing with the major exception of those darker horizontal lines:



I'm newcomer to AviSynth scripts and will appreciate every help. What is done on the video so far is capture to YUV2 in Lagarith codec and QTMC deinterlacing. The goal is to make DVD and x264 video later. My wish is to use as less damaging tehnique as possible. I dont't like to use heavy manual intervention, but time for processing is not critical.

I have played with Smart Skating Rink, Camcorder color denoise, Color Mill filters on VirtualDubto to achive better results, but those are not applied on samples. If we can come to entire AviSynth chain, it is even better.

[2Bdecided]

I don't have a fix (though remove dirt helps when it's just on one frame), but I'm sure someone is going to ask for raw samples (i.e. before you deinterlaced using QTGMC).

Cheers,
David.

[sanlyn]

Your samples have been encoded to lossy h264. h264 is not considered an editing format, but is a "final delivery format". If you have cleanup cleanup to be worked from analog source, you should be capturing and working with losssless media. There is no information how how the source was captured or the software you are using. The source is not tracking properly and/or is damaged. Colors and bright highlights are clipped to the point where some of the brightest colors are being changed or otherwise corrupted. There also appear to be something that looks like improper colorspace conversion.

If your final output target is standard definition BD or AVCHD or DVD, those SD formats are usually interlaced. In any case you will get more precise advice if you submit a sample of the original captures, before processing.

[RangerCat]

Your samples have been encoded to lossy h264. h264 is not considered an editing format, but is a "final delivery format". If you have cleanup cleanup to be worked from analog source, you should be capturing and working with losssless media.

I'm using Lagarith internally.

There is no information how how the source was captured or the software you are using. The source is not tracking properly and/or is damaged. Colors and bright highlights are clipped to the point where some of the brightest colors are being changed or otherwise corrupted. There also appear to be something that looks like improper colorspace conversion.

I used VirtualDub to capture from Asus TV 7134. The attached bellow sample is what I can get, because no access to pro equipment. Also it is not possible to recapture anymore. However I'm convinced that the main damage is from the tape/recorder itself.

If your final output target is standard definition BD or AVCHD or DVD, those SD formats are usually interlaced. In any case you will get more precise advice if you submit a sample of the original captures, before processing.

I agree. The main concern here is that I evaluate computer format the most. I don't like to process twice. Also I'm not sure that the most filters will work properly than. But You can convince me about that.

...but I'm sure someone is going to ask for raw samples (i.e. before you deinterlaced using QTGMC).

New raw input sample is attached bellow.

Thanks!

[jagabo]

Try capturing the same scene twice. If the dark lines aren't in the same place in both caps make a third cap and use a "median of 3" filter to remove the dark lines.

https://forum.videohelp.com/threads/340963-Best-quality-and-speed-video-denoisers-2011?...=1#post2122313

[2Bdecided]

If it's tape damage, it'll always be in the same place though.

EDIT: OK, many of the lines don't look like tape damage, but like a faulty VCR. Parts obviously are tape damage though.

[RangerCat]

Try capturing the same scene twice. If the dark lines aren't in the same place in both caps make a third cap and use a "median of 3" filter to remove the dark lines.

Great advice! Unfortunately neither the tape, nor the video are available any longer. Shame on me.

The question is can I squeeze a bit more. Otherwise I will process with the current quality.

[sanlyn]

You can clean of some of those rolling gray bars, and clean up the chroma bleed, oversatauration, invalid luma and chroma levels, bring the black levels down to normal, etc. The saturation and contrast levels are too high, losing a lot of bright detail and giving an unnuatuiral appearance overall (the yellow building looks washed out, for example, and ye;llow is so hot that some of the lighter whitish lines are blurring). A few more of the rolling lines might be cleared with some very painstaking work using mvtools -- I've seen it done, but not with the frame doing a war dance. Many of the bad lkevels and color problems could have been fixed during capture. Obviously, many of those lines nand the vertical distortion and ripples could have been minimized with a line tbc -- but, just as obviously, that solution is not available. You'll need a high bitrate to manage that camera motion.

Some of those rolling bars were cleaned up in the sample below, but many remain.


[Attachment 21817 - Click to enlarge]

[Attachment 21818 - Click to enlarge]

[jagabo]

I found:

SeparateFields()
evn=SelectEven().RemoveSpotsMC()
odd=SelectOdd().RemoveSpotsMC()
Interleave(evn,odd)
Weave()

removed some of them (especially in the blue sky) without damaging the rest of the picture much.

I tried using McTemporalDenoise() but you have to use strong settings to remove the dark bands -- and that damages the rest of the picture a lot.

Since the dark bands usually only appear for one field I'm looking into using GeneralConvolution() to build a mask that highlights areas with alternating bright/dark scan lines. Then one might be able to use that with to overlay of a strongly filtered video with the original.

[sanlyn]

That's what I did, but with QTGMC and SelectEven, then ran even and odd frames from that progressive batch with RemoveSpotsMC and interleaved the result. I also used RemoveDirt, which seemed to clean up at least a little of those wiggly verticals.

A mask technique could likely do even better. This is why I'm embarrassed to admit that masktools and the like are still a mystery to me. I've seen a script from poisondeathray a while back that used mvtools and masktools to clean up even worse than the sample, but there was very little motion involved. There are surely more effective ways of getting there. Share the results, if you would (it'll be yet another addition to my very extensive jagabo folder!).

[RangerCat]

Some of those rolling bars were cleaned up in the sample below, but many remain.

Even a little is a lot better. Also most of the videos are moving slowly and have less damage. But those lines are common, the only difference is that sometimes they are thicker. In fact I was so upset with the colors and other things at the beginning, that I didn't recognize them as so irritating.

Since the dark bands usually only appear for one field I'm looking into using GeneralConvolution() to build a mask that highlights areas with alternating bright/dark scan lines. Then one might be able to use that with to overlay of a strongly filtered video with the original.

I will greatly appreciate that! I was thinking about the same, but don't know how to achieve that.

[mammo1789]

He can also use DeVcr (best if used multiple times like run 1 =0,run 2=20/30 or Descratch after the removespotsMC or even RemoveDirtMC

SeparateFields()
evn=SelectEven().RemoveSpotsMC().DeVCR(0)
odd=SelectOdd().RemoveSpotsMC().DeVCR(0)
Interleave(evn,odd) Weave()

In cases like this when "colored lines" appear it is best to use camera color denoise vdub after. I think Neat video can also help.

[sanlyn]

Sad to say, NatVideo has only a minimal effect on this kind of disturbance. However, whatever is does even if mild is usually visible. I used it here, but at low power (something like 30%) because the vid is so soft to begin with -- and it was used mostly for its sharpener. In all fairness to NV, the clip doesn't offer a suitable frame for an effective noise sample. I used a sample from another vid.

[jagabo]

Since the dark bands usually only appear for one field I'm looking into using GeneralConvolution() to build a mask that highlights areas with alternating bright/dark scan lines. Then one might be able to use that with to overlay of a strongly filtered video with the original.

I will greatly appreciate that! I was thinking about the same, but don't know how to achieve that.

It didn't work out. It mask marked the interlace comb artifacts much more than the dark bands.

[RangerCat]

It didn't work out. It mask marked the interlace comb artifacts much more than the dark bands.

I see.
I was trying to use DeScratch, but have kind of similar results - it is either too much sensitive or doesn't detect them. Further more, even it "sees" defects it can't repair them well.

inputClip = AVISource("video.avi")
lowColorClip = ConvertToYV12(inputClip, interlaced = true)
rotatedClip = TurnLeft(lowColorClip)
descratchClip = DeScratch(rotatedClip, mindif=1, maxgap=12, minlen=510, maxlen=770, blurlen=0, keep=95, border=0, modeY=1, maxangle=2, mark=true)
resultClip = TurnRight(descratchClip)
StackHorizontal(lowColorClip,resultClip)

Sad to say, NatVideo has only a minimal effect on this kind of disturbance.

I didn't like NatVideo much. It once was able to do good job about these lines, but I never remake it.

He can also use DeVcr (best if used multiple times like run 1 =0,run 2=20/30 or Descratch after the removespotsMC or even RemoveDirtMC

My initial effort with DeVcr and RemoveDirtMC does not came to good results. Did't try removespotsMC so far.

In cases like this when "colored lines" appear it is best to use camera color denoise vdub after.

Yes, I think CCD is worthy. I plan to use it.