Comet samples requested

[jmac698]

Hi,
I'm working on a better de-comet plugin, and I'm looking for some samples of comets on VHS tapes. That's when random white lines appear on old tapes, which change position with each capture.

The ideal sample:
-a flat area, such as a wall, blackness, or colourbars, which is dark
-comets present in some of the frames
-about 1s, uncompressed video
-at least 3 captures of the same spot
-TBC was used preferably
-proper capture levels were used
-ideally no camera movement (such as animation or tripod mounted camera)

Thanks

[ChiDragon]

I don't get it. If blackness works, why does it matter whether it's 3 seconds of uninterrupted blackness or the same 1 second played back 3 times?

Here's 3 x 100 frames. Portions of the test pattern are flat grey. I have no option to correct the levels; whites are unfortunately blown out.

Comets - 3 VHS samples for jmac698 [Lagarith].zip on Mega.co.nz (https://mega.co.nz/#!YIEQgKIK!FuV8-HOgjAoAWsXu8b_JRqsMy2SkhrFh_HzEka2Nw0g) (77.1 MB)

[jmac698]

Thanks for your sample! That's a good question, normally you'd be trying to recover the underlying image, it just happens that in this case you don't care because it's black and all the same. So, you're right, I don't think 3 passes matter :( Now I realize that things like 12fps animations basically have a built-in 2-3 passes that you could use in that special case.

I wrote a program to do some statistics. One of the things I wanted to look at was, does the underlying video affect the comet in any way? So far it seems that comets are opaque. As for why dark video, I want to see the range of luma values the comet can take up. Also, I looked at if comets affect colour and so far they don't. I love that you have flat grey; the point here is to make my statistics against a known stable signal. I can measure the amount of noise in the signal.

Part of the reason for all these statistics is that I'm looking at a guassian mixing model and cluster analysis. https://en.wikipedia.org/wiki/Cluster_analysis#Density-based_clustering

The hardest part to mask of comets is where the underlying video is close and overlapping in luma. To separate that, I have to go beyond 1 dimension and look at the geometry of comets. I find that it's not just a length but can have several pulses with heads and tails.

I'm thinking of some kind of technique where you calibrate/learn a pattern on one frame of the video that could be applied to the rest. The idea is to do better than median and use less passes. Even median can still show some remaining, but faded comets with just 3 passes. It needs to be more intelligent.

Having more passes lets me average them out to see what the "true" pixel luma should have been, and more accurately shows the statistical spread; they I can see how much worse off an estimate would be with few passes. etc. :)

[jmac698]

In this case 3 passes helped, since there were very few comets and with more passes (or length) I get to see them against different backgrounds and more of them.

[Mounir]

Video with comets:
https://mega.co.nz/#!J9xTEZbA!RRBnR-HztdN6MTe9q966yBoL1wcvWJOm8ZUv_4vWgzc

[scharfis_brain]

@Mounir: these are comets from an analogue sattelite-TV transmission. They are not caused by VHS.
the comets clearly vary in amount depending on the color saturation of the image.