How to remove horizontal white lines in VirtualDub/AVISynth

[FrankDrebin]

Hello everybody.

I have an old VHS tape which I captured today. Everything is fine except for the fact that, sometimes and randomly, there is an horizontal white line in the top area of the video which lasts about 5-6-7 seconds This happens only with this tape, so I think it has nothing to do with the VCR heads. Here are a few frames taken from the video:

http://img266.imageshack.us/img266/5586/e7w8.jpg
http://img203.imageshack.us/img203/2394/6n2q.jpg

How can I get rid of it? Is there a VirtualDub filter/AVISynth script which could allow me to do that?
If so, since this white line is not always present throughout the video, could it be possible to specify only the video segments on which to apply the filter/script?

Thanks in advance.

[jagabo]

Does the line move? Is it the same every time you capture? Provide a video sample.

Use Trim() to select portions of a video. Then append them with + or ++.

part1 = Trim(0,99)
part2 = Trim(100,199).filter()
part3 - Trim(200,0)

return(part1 ++ part2 ++ part3)

[johns0]

That line is caused by tape skew,parts of it has been creased due to a poor tape path such as bad pinch roller.Gonna be very tough to filter out.

[FrankDrebin]

Does the line move? Is it the same every time you capture? Provide a video sample.

Use Trim() to select portions of a video. Then append them with + or ++.

part1 = Trim(0,99)
part2 = Trim(100,199).filter()
part3 - Trim(200,0)

return(part1 ++ part2 ++ part3)

The line doesn't move, it always appears in the same place.

[jagabo]

You could try replacing the bad scan lines with data interpolated from the lines above and below:

function InterpolateScanLines(clip src, int first, int lines)
{
     top=Crop(src,0,first-1,src.width,1)
     bot=Crop(src,0,first+lines,src.width,1)
     StackVertical(top,bot)
     BilinearResize(src.width,lines+2)
     Crop(0,1,-0,-1)
     Overlay(src, last, 0, first, opacity=1.0) # change opacity to suit
}

ImageSource("p1.jpg")
ConvertToYUY2(interlaced=true)
SeparateFields()
InterpolateScanLines(last, 35, 5)
Weave()

[poisondeathray]

In addition to using trim() or similar methods to apply the filter to specific frames, you can also apply filters to a particular section of a frame by using a cropped filtered overlay using overlay() or mt_merge() , or by using masks

Search for some example of scripts from user "pirej" , he had some that worked ok, or at least improved significantly these kinds of static location defects . IIRC, He combined some of videofred's script with a bunch of other things

If you post a sample , you might get more specific help

[FrankDrebin]

Thanks a lot everybody for your precious help!

If you post a sample , you might get more specific help

You can download a short clip of my video from here: https://forum.videohelp.com/attachment.php?attachmentid=19286&d=1376141541
It has no audio track and it's compressed in MJPEG format.

[poisondeathray]

I'm going to assume that you've tried the tape in another machine?

There are other problems with field alignment and warping, but regarding the line defect:

This one is tough, because it's essentially an opaque line underneath ; there are no good pixels "underneath" or on adjacent frames to salvage in that area in the affected frames - so that essentially rules out any avisynth motion compensated or mvtools2 related fixes . The reason why pirej's (and related) scripts worked ok - there were large white or grey lines, but unlike here there were at least a few good pixels "underneath" the noise pattern . That's why it was important to see a clip to analyze the pattern

So you're left with either inpainting filters, or blurring specifically that section in some manner (e.g. vertical blur confined by masks) .

Impainting usually leave alot of artifacts (only does ok on some types of patterned backgrounds).

A variant on the blurring method is the "mirror inward" approach. Basically , the line defect is divided in half and edges adjacent on the "good" part are mirrored inward to cover the defect with a slope blur to blend it. Tough to do in avisynth, unless the defect is completely static - which it's not. This is what some "wire removal" software does in FX packages (think of actors suspended on wires) . It's typically done with some motion tracking and keyframe interpolation (so you don't have to adjust it frame by frame ) - that's the "easy" way. The "hard" way is to manually paint and borrow textures from other frames , but it's very tedious

[sanlyn]

The bad lines (well, it's a scratch, abrasion, or crimp in the tape) does change position, and it comes and goes. In some sequences it's thicker and lower, in others it's thinner and higher.

[jagabo]

The bad lines (well, it's a scratch, abrasion, or crimp in the tape) does change position, and it comes and goes. In some sequences it's thicker and lower, in others it's thinner and higher.

Yes, but it does so slowly. So you can't use data from an adjacent frame to fill in the bad lines.

[sanlyn]

Yep. That's what I mean. You'd have to change the formula over time, work different segments, patch them together. And it still wouldn't be a clean fix.

[lordsmurf]

You can "fix" this (not really), but it blurs other things.
The easy way to "fix" this is find another copy if you can.
If it were home movies, nothing can be done. (But it doesn't look to be the case here.)