JohnnyRoy: Try this method for video noise reduction

JJKizak wrote on 7/27/2004, 2:43 PM

This is just an idea like the old MTI radars. Run a noisy video clip through a time constant circuit that passes only high frequencys (noise)
and shorts to ground all slow stuff (good video). Then take the resultant noise clip and run it through a phase reversal filter to reverse the phase 180 degrees. Apply the reversed phase noise clip on second video track above the first clip and adjust the composite level for perfect noise reduction (theoretical). Since only the noise is present on the top clip it will cancel out all of the noise on the bottom clip thereby giving you pure video. Do you think life could be that simple? Anyway, just an idea. Need some wix-bang electronics type person to set this up and try it. By the way: if you have two separate signals from two receivers receiving the same data and combine them up either at the audio level or the IF level you get a 3 db reduction in noise because the noise in both receivers is random and if the signal strengths are the same a 3 db reduction is realized. If you add two more receivers you get another 3 db reduction. Well maybe but that was a long time ago and remembering this stuff is tuough in retirement.
One catch is the receivers have to have their audio signals adjusted for perfect phasing to have usable data. Do this with dual trace scope
and lengthen or shorten IF receiver inputs to get perfect match. Then you will get the noise reduction with usable data.

JJK

Comments

johnmeyer wrote on 7/27/2004, 4:58 PM

Noise is broadband (basically white noise), so a low pass filter won't yield a video signal that is just the noise.

Coursedesign wrote on 7/30/2004, 2:46 PM

Expressed another way: the noise is in the passband, together with the video signal unfortunately.

If the noise was in a separate frequency band that could be filtered out, life would be too easy.

JohnnyRoy wrote on 7/31/2004, 7:54 AM

Well... it sure sounds impressive if there was a way to only filter the noise. I’m not sure there is. It does sound fun enough to try though. Now where did I place that time constant circuit? (rummage... rummage...) oh yea here it is, right next to my time machine. ;-)

~jr

JJKizak wrote on 7/31/2004, 8:26 AM

Thanks for the humor. Its raining outside and I needed the laugh. I just feel that once somebody licks this problem it will look so simple. The main problem is doing this "after the fact". If they used the TCP-IP methods it would be a piece of cake. A couple of tricks we used to use were to have the best signal to noise equipment you could get then knock the signal down 20 db and the noise goes down 20 db but you still have a usable signal to apply to the receiver. I noticed on some old VHS tapes that if I use a very good deck with good video signal to noise that they are much cleaner with a slight reduction in noise. One other technique would be to have two clips of the same video, add additional noise to one of them, then compare the two signal wise and maybe get a reduction that way. (noise is random and will be different on the two identical clips.) Maybe somebody will figure out some kind of algorithm to determine what is noise and what isn't, kind of like finding a graviton particle with the theory of strings.

JJK

John_Cline wrote on 7/31/2004, 9:53 AM

The best way to reduce video noise is by comparing consecutive frames, it's called temporal noise reduction and there are probably a dozen different plugins for Virtual Dub that will do this. My oersonal favorite is the one written by Jim Casaburi. When tweaked properly for the source material, it works very well.

John

JohnnyRoy wrote on 8/1/2004, 5:53 AM

I agree, Jim Casaburi’s Temporal Cleaner is great. I also use Steven Don and Avery Lee’s Dynamic Noise Reduction all the time on analog source and it really does a good job of clearing it up. I believe that’s temporal also.

~jr

farss wrote on 8/1/2004, 6:48 AM

Even temporal noise reduction carries a price. My ADVC 300 uses it and it does an excellent job in RT. However if you look fairly carfully you notice a ghosting effect on movement. Now this makes sense, as an edge moves between frames the circuit has no way of knowing that it isn't an interframe change because of noise or because something moved. Now I guess with enough signal processing you could use motion vectors to determine what was a moving object versus what really is noise.

Bob.