Can anything be saved from this source?

[Piper]

An acquaintance in the UK sent me a source file of a comedy gig he did in 1978. The video was recorded by some students who used some borrowed some video equipment from their university. Since then, various attempts to preserve the video have been made which has ultimately resulted in at least 2 VHS generations and then ultimately the DVD which I ended up receiving. Because the source is so noisy, the DVD version itself introduces additional compression artifacts to an already brutal source.

On the upside the video is Full D1 PAL (Interlaced), and the audio is also surprisingly good.

So far, I've thrown a slew of filters at this source, but I can't seem to find a good balance between not enough denoising or completely destroying what ever detail is left. This is easily one of the most challenging sources I've dealt with...

Would anyone be able to suggest what if anything can be done with this source? Please only download the source if you think you can do something productive with it (Only 10 downloads, free account and all that).

Rapidshare: jsc.avi (http://rapidshare.com/files/193395859/jsc.avi.html) (No Audio)

http://farm4.static.flickr.com/3447/3250092629_73e232685c_o.jpg

[2Bdecided]

It can certainly be improved.

When you start to work on it, I think you'll find the MPEG-2 DVD encoding is a real problem. I'm guessing many of the analogue problems were in the original recording, subsequent VHS generations made this slightly worse (plus occasional nasty tape drop-out related issues), and then the DVD encoding obliterated what was there.

(When uploading snippets from a DVD, use DGindex to cut a bit of native m2v - no need to decode to lossless.)

It's not interlaced. The fields are different - one field is much poorer quality (worse encoding artefacts) than the other - but the picture content is identical - even the tape damage is identical in both fields, which is quite strange. It wouldn't look like that if it was a faithful transfer from VHS to DVD - someone has already deinterlaced it.

A decent capture of the surviving VHS would be great. If this really isn't possible...

Most important: cut into sections, use appropriate levels commands to get the video level correct in each section.

A quick way to find out the best that you'll get is to look at the output from this script at points where there is little movement...
avisource("jsc.avi")

#bob()

bicubicresize(720,288)
bicubicresize(720,576)

levels(70,1.0,200,16,235,coring=false)

temporalsoften(2,255,255)

#histogram(mode="levels")

levels(0,0.95,255,0,255,coring=false)
levels(20,1.0,220,0,255,coring=false)

sharpen(0.5)...the levels commands work for the second part of your clip - they're useless for the first part of your clip.


If this is good enough for you, you need to replace the temporal soften with a decent denoiser. The problem is stopping the motion compensated denoisers from tracking the noise!

Throw a deblock in at the start.

Cheers,
David.

[Piper]

Thanks for the reply and my apologies for the slow response. Your input is much appreciated.

The levels gives me a good starting point. I had initially come up with something similar to your 3rd example using the histogram, though it didn't occur to me to use 3 iterations of that filter. Your example shows nice improvement.

For denoising, I'm currently tweaking MC_Spuds to see how far that will take me. Like you say, trying to keep it from tracking the noise will be the trick.

Cheers.

[2Bdecided]

Yes, I usually use mc_spuds for VHS, but it's not doing so well here.

Smarter people than me will be along in a minute to suggest trying the fft3d thingy, or some mvdegrain version fed with a temporalsmooth denoised clip for the motion estimation. Or something.

I'd only normally do levels once, but with so much noise, it seemed to make sense to get it vaguely in range before the denoiser, without clipping the excessive noise (which might yet average out to the correct value when denoised - so no point clipping it first). Then with the excessive noise removed by the denoiser, the range can be tweaked properly. I separated the gamma levels from the low/high levels for convenience of tweaking, that's all.

Cheers,
David.