Can anyone repair a rolling VHS tracking band in an MP4-only source?

[edwebster3]

Hi everyone. I'm learning but I think this is over my head and beyond the tools I have. Anyone willing to assist?

[johnmeyer]

This is similar to the noise bars I removed from an amateur movie film of a TV set. Here is the the thread at doom9.org where I developed the technique:

Bad 1950s Kinescope

You can see the rather remarkable result here (I still consider this one of my best restorations):

Noise Bars Before-After

Unfortunately, it is going to be more difficult to detect your noise bar when it becomes partially transparent.

[edwebster3]

Thank you, but I don't think that technique is going to work for me because it's not just a matter of changing the shading... I need to reconstruct the pixels somehow using the other parts of the video as a reference... something I think AI can do but I'm not sure what toolset is advanced enough to do it... Especially because the band runs right through the actor's face.

[johnmeyer]

Thank you, but I don't think that technique is going to work for me because it's not just a matter of changing the shading... I need to reconstruct the pixels somehow using the other parts of the video as a reference...

A true reconstruction will certainly require AI. There are, of course, all sorts of fixes which involve using adjacent pixels to fill the problem area. The various "delogo" tools usually do this. You could do this here as well because, if you look at how I did it, the final fix was actually done in Vegas using a mask created by AVISynth. In the final step. instead of changing the exposure, you would use the adjacent pixels in Vegas, or feed the mask to one of the delogo filters.

However, the result might actually be more distracting that the semi-transparent moving bar you have now.

If I were you, and you still want to pursue this, I would first search out what AI tool can reconstruct an area in a still photo, using adjacent pixels. See if you can get a good result on one single frame by designating the section you want to replace and then feeding that to the AI. If you can get one good frame, there are often ways to deconstruct the video into individual frames, and then automate the single-frame approach to fix ALL frames.