How to get of these large horizontal lines

[Heathcliff]

Hello,
I have an avi with a lot of horizontal lines, some are quite tiny, some are large, is theres a way to get rid of them ?
I tried VerticalCleaner, it work not so badly on the tiny lines, but did not affect at all the large ones;

so what do you think about it ? possible ? not possible ?
many thanks !

http://heathcliff08.free.fr/captures/bucheron1.jpg

http://heathcliff08.free.fr/captures/bucheron2.jpg

http://heathcliff08.free.fr/captures/bucheron3.jpg

a 2sec sample - avi -15Mo (http://heathcliff08.free.fr/captures/bucheron_sample_huffyuv.avi)

[Guest]

How did you make the AVI?

[Heathcliff]

it's a capture from a 1'' tape as far as i remember.
but i'm pretty sure it is already like this on the master.

[vcmohan]

You may try DeJitter. The lines appear to be out of sync lines.

[Heathcliff]

The lines appear to be out of sync lines.
it seems to be out of sync lines, yes.
i tried dejitter but i could'nt have satisfying results. maybe the lines are too big, i don't know.
However i verified the master yesterday and it's an 1" tape and the lines are on present the master.

[scharfis_brain]

you could turn it 90° and apply descratch()

[Heathcliff]

you could turn it 90° and apply descratch()
they are still there after that.. i think the problem is the line are not in just one colour :/
i start thinking there is no solution :/

[thetoof]

You could try a motion compensated approach with very restrictive parameters so the motion vectors of the lines won't be calculated and with some function of the MVTools they could be removed...

[2Bdecided]

If the H-shift is constant, it would give an easier way of finding the line - any line2line transition which has far less change when that shift is corrected than when it isn't is probably the start of a bad region; the opposite for the end. Then apply the H-shift over that region.

If it varies, this won't work.

Cheers,
David.

[R3Z]

MugFunky wrote a function called morph that uses motion compensation to replace extreme differences of bad frames.

[Heathcliff]

If the H-shift is constant, it would give an easier way of finding the line - any line2line transition which has far less change when that shift is corrected than when it isn't is probably the start of a bad region; the opposite for the end. Then apply the H-shift over that region.

If it varies, this won't work.

it does not vary. however, i have to say i don't know how to do that :(

MugFunky wrote a function called morph that uses motion compensation to replace extreme differences of bad frames.
i just tried it. it works perfectly when the line(s) are just present during one frame. But when they're present during 2 frames or more continuously, the whole frames become completely blury.

[R3Z]

it does not vary. however, i have to say i don't know how to do that :(


i just tried it. it works perfectly when the line(s) are just present during one frame. But when they're present during 2 frames or more continuously, the whole frames become completely blury.

Thats to be expected with a good - bad - good frame approach.

How important is this clip to you ?
If its really important you could persist with morph and use photoshop to clean up the sucsessive bad frames and feed that back into the morph function.

[foxyshadis]

You can also customize the function for the case of two bad frames - the first frame compensates using mvanalyse 1 back and 2 forward, the next vice versa. It's uncommon but perfectly workable.

[Mug Funky]

i wouldn't use morph if the bad area is only a tiny chunk of the frame... it's meant for whole frame splats like what you get from a digibeta that's been mistreated. it's also meant as a way of avoiding photoshop :). i'm kinda glad i don't do that job anymore.

the problem with something like this is finding a reliable way to detect those dropouts without munging anything else in the frame. the typical way to do this is to check if the same spot is nearby on adjacent frames. unfortunately in your case it is, so there's no safe way for a dumb computer (or a dumb scripter :)) to distinguish with decent confidence between one of these craps or a genuine horizontal edge.

if these things are all through your tape, it's time to track down a new tape or hand over the job to a monkey who can photoshop it all for you. painstaking i know...

[edit]

btw, morph has arguments to replace more than one frame. there's just more chance of weird artefacts. it can compensate as many frames as your machine has memory (you'll know when you've asked too much of it :devil:). it has an argument for inpoint and output, so just go through in virtualdub and write down the framenumbers of the last good frame and the next good frame. then i'd recommend photoshopping the original back in wherever it fits.