PDA

View Full Version : filtering bands/scratches


AlanHK
4th October 2007, 05:37
I've got an old B/W video, which has some vertical lines or scratches. (Eg, the dark band through the man's eye.)

The bands are about 9-10 pixels wide, and are throughout most of the video, moving only slightly from frame to frame.

I tried to use Fizick's descratch, like this:
descratch(mindif=2, minlen=200, maxgap=20)

But it wasn't very successful. It detected and filtered the scratch about 20% of the time, leaving a flickering effect even worse than the original.

In the docs, it says:
maxwidth - maximal scratch width (1 or 3, default=3)

Does this mean it only detects scratches up to 3 pixels wide? As mentioned, those in this video are 9-10 pixels wide.

Any suggestions on tweaking descratch's parameters, or other filters to try?

Spuds
5th October 2007, 15:41
Can you post a short sample we can work on?

Mug Funky
5th October 2007, 17:06
get the original camera neg and make a wet-gate print of it :) j/k

you might be able to use a mask and brighten under it to match what's around it, but it'll be tricky tracking it when it moves.

something like this is pretty tricky to automate, but i don't think it's impossible.

Didée
5th October 2007, 17:35
Something along the lines of

- reduce width to (a bit less than) 1/3
- descratch
- get difference (narrow-descratched)
- scale difference to original size
- apply scaled difference to original input

should do.

AlanHK
7th October 2007, 08:27
Something along the lines of
- reduce width to (a bit less than) 1/3
- descratch
- get difference (narrow-descratched)
- scale difference to original size
- apply scaled difference to original input
should do.

So that means descratch defines a scratch as no more than 3 pixels?

Anyway, I reduced the video to 25%, but couldn't get descratch to reliably detect the bands even then.

When the band was over a dark background, and thus just a little darker, it seemed to lose it (using the "mark" parameter to show what it had detected), as well as detecting lots of false positives.

Can you post a short sample we can work on?

Thanks, try http://www.badongo.com/vid/494259
5 MB.

steptoe
8th October 2007, 19:05
That looks like the very early steptoe and son episodes that I also have, I've tried doing some alterations to it, but its not going to be easy and will probably need some very specific filters and functions

Main reason is I think the early B&W episodes are converted from 405 line recordings that were in the personal collection of Galton and Simpson, they had the only copies in existence until converted from 405 to 625 line, hence the atrocious quality

You might be able to improve them slightly, but I gave up as they were already a conversion from a very poor and very old master so thought it was better to leave them as they were

I didn't have time time needed to mess with filter settings, nor the knowledge to write fuctions purely for these



You could try RemoveNoiseMC, its a very good function that I had played with on a very old and very poor quality laserdisc to DVD conversion I aquired and it cleaned and sharpened those pretty well

http://forum.doom9.org/showthread.php?t=110078

It did a very good job on the Laurel and Hardy box set that had a lot of seriously poor quality films and shorts that were untouched and have very bad grain, scratches and spots 'to preserve the original format', in other words they really mean 'we can't be bothered spending the time and money needed to restore them for the return we would get selling the box set to a very specific target'

Be warned its a very, very slow function as it calls a few slow filters to do the work, along with even slower sharpening functions, but the wait is worth it

Even with a dual core AMD 6000+ running at 3.2ghz and 4GB of RAM it chugs along at about 5-7fps

AlanHK
9th October 2007, 01:53
That looks like the very early steptoe and son episodes that I also have, I've tried doing some alterations to it, but its not going to be easy and will probably need some very specific filters and functions

Yes. this is series 3, ep 1. Most of the rest are not bad. The very first, pilot episode, was pretty grungy, RemoveDirt helped.

Also the soundtracks on several were helped by Audacity's noise removal filter.

But the persistent bands mentioned are only in this episode (so far).

Slowness doesn't matter, I can run it all day in the background.

2Bdecided
9th October 2007, 13:00
It did a very good job on the Laurel and Hardy box set that had a lot of seriously poor quality films and shorts that were untouched and have very bad grain, scratches and spots 'to preserve the original format', in other words they really mean 'we can't be bothered spending the time and money needed to restore them for the return we would get selling the box set to a very specific target'

Have you seen this?
http://www.laurel-and-hardy.com/html/res.html

Cheers,
David.