I am capturing some old VHS tape and have noticed a strange noise / pattern on my captures.
It resembles a vertical herring bone pattern and is worse on some colours than others. The VHS tape used is not the best but even so the noise/pattern is frustrating.
My hardware is as follows:-
AJA Kona LSe.
Panasonic SVHS deck 7350.
Final Cut Pro X
+ Reply to Thread
Results 1 to 19 of 19
You appear to be using a Mac. Maybe this should be moved to the Mac forum?
If Mac is the case, I'm afraid you're limited when it come to the kind of video repair needed here. But it could also be a fault of your capture device.- My sister Ann's brother
I believe the "resolve it" part was the point. The software (and possibly hardware) available to a Mac user is extremely limited compared to that available to your average Windows user.
To me it does seem to be some kind of interference but I am no expert hence asking the question.
This conversation should be taking place in the Mac section.
The card is designed for use on the Macintosh platform, making it an excellent choice for Final Cut Pro users.
Does your card really have BNC connectors as inputs?
I didn't imply that using a Mac would be a cause, although the capture device might have something to do with it. The noise is a type of RF effect. Filters that clean it usually make use of convolution matrices and masking techniques.
One filter often used to clean this particular type of RF noise, which can take many forms such as floating gray bands or moire as well as fine checkered grids, is Avisynth's DeStripe plugin script. One example of its use is here: https://forum.videohelp.com/threads/333982-Bizarre-vertical-lines?p=2072013&viewfull=1#post2072013. Another example here: http://forum.doom9.org/showpost.php?p=1449523&postcount=9 . The script is called DeStripe.avs and originally appeared here: http://forum.doom9.org/showpost.php?p=1408276&postcount=31. DeStripe requires that video is de-interlaced, inverse telecined, or otherwise without field or frame blending. The colorspace must be YV12.
Another filter that often works a little better on fine-grid hash or herringbone is VirtualDub's fsn.vfd filter ("Frequency Noise Suppressor). An example of its use is here: https://forum.videohelp.com/threads/362889-Advice-on-a-VHS-transfer?p=2326764&viewfull=1#post2326764 . The fsn.vdf filter is listed among many VirtualDub noise suppressors on this web page: http://www.infognition.com/VirtualDubFilters/denoising.html. A copy of the vdf filter and brief documentation can be downloaded here: http://www.infognition.com/cgi/getfilter?id=137 . It originally appeared on a tiny and rather uninformative Russian website (the page is in Russian but can be read via Google Translate): http://acobw.narod.ru/fsn.html. The filter works only on deinterlaced video in RGB.
A long time ago I noticed a similar filter plugin for After Effects Pro, but it was part of a package that cost well into 4 figures. Perhaps someone makes an RF noise suppressor plugin for FCP ? ?
The problem can often be caused by faulty connectors or bad grounding somewhere in the capture circuit.- My sister Ann's brother
Yes the Kona LSe has BNC connections same as my VTR, It also has a breakout cable with :-
3x BNC R,G,B in / out.
3 x BNC YPbPr in / out.
4 x XLR connectors Audio in / out.
+ Sync cable.
NeatVideo both of which do a pretty damn good job at removing the noise but I'm determined to prevent it at source. Will probably end up being a dodgy cable or something equally as silly
It might be a lack of comb filtering when separating composite into luma and chroma. Without a comb filter you will get herringbone noise anywhere there's color. The more saturated the color the more herringbone noise. Or it could be some other source of high frequency noise somewhere in the chain.
A short non-reencoded sample would make analysis easier.
The noise pattern is not on the tape because it is too high of a frequency for VHS to store it like that. It looks like a grounding issue between VCR and capture device to me. You didn't mention anything about S-Video inputs on your capture device, just YPbPr and RGB. I just looked it up – the Panasonic 7350 has S-Video out and Composite out (via BNC). Which are you using and how?
It's not a true "fix" of the problem directly, but a multiple capture+merge would greatly alleviate the problem, as it is extremely unlikely that the noise pattern is regular and in the same spot(s).
Lot of effort, though.
This is also not a true fix but an inverted high pass of 0.4 pixels should remove the pattern.
Last edited by newpball; 11th Nov 2014 at 13:07.
The noise pattern is not on the tape because it is too high of a frequency for VHS to store it like that. It looks like a grounding issue between VCR and capture device to me
Maybe Defreq (avysinth) http://avisynth.org.ru/defreq/defreq.html can be more suitible for this kind of problem
Depends. If the noise is already captured/recorded on tape, Median/TimeAveraging won't work. If it's a playback-only error, it could work. Both are possible.
Looking at what Defreq does from a visual standpoint, doesn't seem like it would be that helpful, but I haven't scoured the algorithm either.
@newpball, an "inverted HP filter" = a LP filter = blurring. It can work, but would "throw the baby out with the bathwater".
I suppose you think the results I posted are not very good, so if you think my method is so bad why don't you put an image of your superior method next to the image I posted? Or even better if the poster can provide a 10 second video we can put both our results here.
Last edited by newpball; 14th Nov 2014 at 00:13.
You do understand how time averaging works, right? It needs 3, 5, 9, etc. separate captures/passes of TBC'd identical-length frames in order to provide the variety needed for the Median to work. Yes, I'm confident that my suggestion would work better than yours given what you showed (it would retain noise-free smoothness without exhibiting the blurring, plus is ~noise-type agnostic), but I no longer have the equipment to do those kinds of transfers (long since transferred all my VHS -> digital and don't do this as a service anymore, even though I did it for over a decade at my last production company).
So, no, I don't have any material to show you. Perhaps others do, such as Lordsmurf or Orsetto. However, time averaging is quite labor-intensive and so is not meant (or necessary) for your everyday job. Clearly, the Median method cannot make use of a single clip from a single pass. That's why I wrote the "lot of effort" bit.
What I can show you (and others) is that you are throwing out more than just noise with your suggested filter/setting. Here is the (normalized) difference between the original and your filtered version. The normalization (aka expansion of contrast) exaggerates the differences, so those differences aren't as obvious when used normally. However, it makes clear the fact that WHAT your filter is filtering goes well beyond the noise, to the point where you can recognize the train + windows + clouds that have had elements removed. One of the prime tenets of restoration/enhancement/noise-reduction is to err on the conservative side of processing.
<edit>I do have a link to something similarly done in a research paper (with pix): https://hal.archives-ouvertes.fr/hal-00510866/document</edit>
Last edited by Cornucopia; 14th Nov 2014 at 01:18.
Last edited by newpball; 14th Nov 2014 at 01:37.
Both methods are valid, one way or the other, more or less. Would be more interesting and effective if we had some real video to work with instead of chopped up, resized stills. Can't do a damn thing with stills.
Combination of resize and/or Virtualdub's frequency noise suppressor. Some real, unprocessed video please? Then we can debate to our heart's content. I resized/added borders to have 8x8 blocks and mod-8 to work with. Were these cropped in YUV using odd pixel numbers?
Last edited by LMotlow; 14th Nov 2014 at 08:26.- My sister Ann's brother