Re: [Discuss-gnuradio] Analog video restoration

[Dave Dodge]

On Wed, Nov 02, 2005 at 02:53:24PM -0500, Marcus Leech wrote:
> So, this whole MPAA nonsense got me thinking about technology for 
> converting old VHS tapes into
>  modern formats, and doing some quality restoration in the process.
> 
> Imagine, for a moment, that you have several 1st or 2nd gen copies of a 
> given VHS tape, and you're able to synchronize them.

I suspect the synchronization is going to be a significant problem,
especially if you're talking about running a couple of VCRs at once.
VCRs are notorious for producing inconsistent and badly-timed output,
even for the same tape in the same machine.  It may look okay played
directly into a TV, but when you try to do anything that requires
precise timing, you start running into trouble.

I've been exploring a similar idea myself recently.  I took a
commercial VHS tape of an obscure movie (that was not particularly
well-mastered).  I used a fairly high-end consumer VCR, a dedicated
outboard comb filter, and lossless capture from the S-Video input of
my HD3000 card.  Capturing the same clip multiple times, I find:

  - Occasional frame dropouts, with different fields or frames being
    dropped each time.  I assume this is because the VCR's output
    timing is globally a bit too fast or slow.

  - Subpixel (or greater) shifting of a given scanline between
    captures.

  - Subpixel (or greater) wavering of scanlines within in a single
    field, such as vertical lines becoming slightly wobbly.  It's
    different every time, and in fact is different for two fields
    sourced from a single film frame.

Another thing to bear in mind is that the VHS tape has the Y and C
signals recorded separately, so it's really better if you get a VCR
that dosen't mix them before giving you the signal (turning composite
back into clean Y/C is known to be a difficult problem).

I've since obtained a low-end broadcast VCR.  It has Y/C output, and a
built-in time-base corrector that will hopefully reduce the scanline
alignment problems.  I haven't had a chance to repeat my capture
experiments with it, so I don't know how much of an improvement it
might be.

In any case, stacking/layering/averaging multiple analog captures to
reduce noise is a well-known technique.  I believe it's very common in
amateur astronomy, with the goal being a single clean image:

  http://www.skyinsight.net/wiki/index.php?title=Image_Stacking

In the video processing field it's a bit more controversial.  Here's
some discussions:

  http://forum.doom9.org/showthread.php?s=&threadid=42733
  http://forum.doom9.org/showthread.php?s=&threadid=28438

I believe there are people who've experimented with mixing not just
VHS, but also multiple Laserdiscs of a film (ideally from different
transfers) in order to produce a high-quality DVD.

In my own case I might also try messing with the capture driver, since
supposedly the cx88 internally captures at a high sample rate and can
output more than 640 per line if you ask it to (I think folks on the
Windows side have found the sweet spot to be somewhere in the 660-670
range).  I also hope to be able to blend repeated fields from the 3-2
pulldown to reduce noise.  Probably the holy grail would then be a
practical video super-resolution algorithm:

  http://ask.slashdot.org/article.pl?sid=03/06/19/2216246

                                                  -Dave Dodge