Averaging multiple VHS captures technique

[JasonCA]

Hi Everyone,

I came across an old but interesting thread in regards to averaging multiple VHS captures which is posted on another thread here:

http://www.digitalfaq.com/forum/vide...-captures.html

I was going reply there, but it said, "This thread is more than 643 days old. You should only reply if you were part of the original conversation AND need follow-up advice or want to give an update. Make a new thread if you have a new question." So I'm starting a new thread since I have a few questions and am very interested in this technique.

In reading the original thread, it's not clear to me how the hardware is setup to capture multiple VHS captures? My understanding is that a VHS tape is captured at least 3 separate times. Then, the Avisynth 'Average' plugin is then used to average 3 separate but similar captured VHS videos of the same source. Correct? Can someone please clarify how to setup for this technique?

However, to me it would seem that the video frames would never align on separate captures? I would imagine that eventually, one of the similar but separately captured videos would drift whereby the video frames would be out of sync with the other captured videos from the original source REGARDLESS if you first aligned the frames in Avisynth at the begging of all the 3 video captures.

There has been some time that has passed now since the original post. How has this technique faired? Has anyone used this technique? Has this technique helped in enhancing the final video product once all the videos have been averaged?

[robjv1]

Quote:

Originally Posted by JasonCA

Hi Everyone,

I came across an old but interesting thread in regards to averaging multiple VHS captures which is posted on another thread here:

http://www.digitalfaq.com/forum/vide...-captures.html

I was going reply there, but it said, "This thread is more than 643 days old. You should only reply if you were part of the original conversation AND need follow-up advice or want to give an update. Make a new thread if you have a new question." So I'm starting a new thread since I have a few questions and am very interested in this technique.

In reading the original thread, it's not clear to me how the hardware is setup to capture multiple VHS captures? My understanding is that a VHS tape is captured at least 3 separate times. Then, the Avisynth 'Average' plugin is then used to average 3 separate but similar captured VHS videos of the same source. Correct? Can someone please clarify how to setup for this technique?

However, to me it would seem that the video frames would never align on separate captures? I would imagine that eventually, one of the similar but separately captured videos would drift whereby the video frames would be out of sync with the other captured videos from the original source REGARDLESS if you first aligned the frames in Avisynth at the begging of all the 3 video captures.

There has been some time that has passed now since the original post. How has this technique faired? Has anyone used this technique? Has this technique helped in enhancing the final video product once all the videos have been averaged?

I've never used this technique, but I often manually combine parts of multiple captures. I imagine the biggest issue would be dropped or repeated frames. This isn't much of an issue if you have the right capture device and a TBC, but it can be an issue with tapes that have physical damage, as you can't guarantee that the VCR will play it back exactly the same way.

[JasonCA]

Quote:

Originally Posted by robjv1

I've never used this technique, but I often manually combine parts of multiple captures. I imagine the biggest issue would be dropped or repeated frames. This isn't much of an issue if you have the right capture device and a TBC, but it can be an issue with tapes that have physical damage, as you can't guarantee that the VCR will play it back exactly the same way.

Exactly! I like you, would assume this too, "I imagine the biggest issue would be dropped or repeated frames." So that leads me to asking the question: how does one create multiple captures of the same source so that frames are not dropped or repeated? The technique of combining multiple captures exists, but it's not really clear how the single source is captured multiple times without inducing dropped or repeated frames?

[robjv1]

Quote:

Originally Posted by JasonCA

Exactly! I like you, would assume this too, "I imagine the biggest issue would be dropped or repeated frames." So that leads me to asking the question: how does one create multiple captures of the same source so that frames are not dropped or repeated? The technique of combining multiple captures exists, but it's not really clear how the single source is captured multiple times without inducing dropped or repeated frames?

I don't think you can ensure it 100%. The best you can likely do is use tapes that are in good physical shape and use good practices (external TBC, fully working VCR with clean heads, etc).

Unless the software is able to detect frames that aren't quite the same (which in theory is all together possible) and alert you to them, then you'll have to keep all of your source files on hand, watch your output from end to end and spot problems and mistakes and fix those manually.

[meson1]

I've done this once. I captured an obscure movie from a commercial ex-rental tape 3 times using an ADVC-100.

I used AVISynth to stack the three captures side by side and view the result in Virtualdub. I could then go through all three captures simultaneously to identify mismatching frames. This isn't as laborious as it sounds. You check say 5000 frames in. If one of the frames mismatches, you search for when the mismatch begins by using a binary chop. So you go back 2500 frames, if they match, you jump forward 1250 frames and so on until you track down where the frames become mismatched.

I then altered the AVISynth script to copy the missing frame from the good clip and insert it into the bad one(s). Then I reload the script in virtualdub and search for the next mismatch. At the end you end up with an AVISynth script with all three clips that line up frame for frame. Then I comment out the horizontalstack instruction and uncomment the averaging part of the script.

This technique does depend on your not having too many dropped frames.

That said I found the practice of averaging to be of limited value if you have a good quality workflow i.e. VCR, procamp, tbc, cables and capture device. Averaging only really eliminates noise generated by your workflow. If the noise is in the recording, then that noise will be present on all your captures regardless, and that can only be addressed by post-processing your capture.

In the end I abandoned my averaged captures, not least because said obscure movie was released in the U.S. on DVD, so I was able to import a legitimate copy.

[jmac698]

Several people on the net have thought of this technique before, I have also used it. It reduces noise from the video head all the way to the file, but of course not noise in the original master. One important use case is when there's brief horizontal white lines throughout the video. I discovered that these will appear in different places, so 3 passes will virtually guarantee getting rid of them.

I should also mention that this technique is developed well in other fields; in astrophotography, we are dealing with sensor noise. We will "stack" hundreds of snapshots of stars in order to average out the noise. Some common options include min, max, average, median, and sigma. Median is great for impulse noise and works better than average, in other words for dealing with the white lines. Average is good for usual noise, called guassian noise. The noise reduction is based on the square root of the number of captures, so doubling the captures reduces the noise one step further. The most caps I'd go is probably 5, but I have tests with 9.

Now, the big part, aligning. This was a sub-problem I had to work on. In fact, there's all the scripts you need out there to automatically align all the captures. There's even options to motion compensate around damaged sections. You can find such threads on videohelp (restoration) and doom9 (avisynth usage). Search for multiple capture, dropouts, comets, white lines, badframes. I don't have time to do the search for you right now.

[JasonCA]

I'd like to throw another idea out there as a possibility: simultaneously capturing from a single source with multiple capture cards?

For example:

[VCR playback]->[Full Frame TBC]->[4 Capture cards]

So, what if I played a VCR tape back, simultaneously captured the stream with 4 different capture cards at one time, and then did some type of merge using (min,max,average,median, ...etc)?

The question is, would there be any benefit from doing it this way? I would imagine since the output stream is analog, the capture cards would each capture the same data but in a slightly different way during the analog to digital conversion. No? I would imagine this would also help avoid the frame's out of sync issue.

Thoughts?

[jmac698]

I've done this too for various reasons. In that case, you are only denoising the effects of the capture cards themselves. If you have a good capture card, the difference is quite slight. Using two of the same capture cards for example will barely make any difference, especially compared to the quality of VHS.

There's all kinds of differences between cards like capture window, clock phase, filtering quality, sampling rate, levels. You can show a difference capturing from dvd. But for vhs, nah. As for catching dropped frames, yes I've thought of that before, can't remember if it works though, I suppose it would.

There's another problem, the TBC itself creates dropped frames. I've proved this by looking at the noise between two duped frames. The small amount of noise shows me that the dupe happened before being digitized. What you really need is a TBC that captures to digital as well.

It's like this, the vcr has unstable timing, the TBC creates a stable output. If the vcr isn't playing back fast enough, even slightly, the tbc *must* create a dupe frame. That dupe frame will come out at a stable frame rate, but is a dupe none-the-less. The capture card then has it's own clock, which will be much closer to the tbc but not quite. It might also add it's own dupes.

What we really need is a complete *unstable* capture device, that works at a variable frame rate, matching the incoming speed as closely as possible. Then we need to create a file with *artificial* timing in it, at the proper speed. Then the vcr which varies 26-32 fps will be artificially set to 29.97fps. Finally we need a playback software that plays at your monitors refresh rate if it's close to the playback rate. In this case it might play at exactly 30fps. Your movie may end a few seconds sooner but you will never drop frames. In other words we need to practically ignore framerate within a few percent.

[JasonCA]

Quote:

Originally Posted by jmac698

I've done this too for various reasons. In that case, you are only denoising the effects of the capture cards themselves. If you have a good capture card, the difference is quite slight. Using two of the same capture cards for example will barely make any difference, especially compared to the quality of VHS.

As of right now, I'm seeing two ways of capturing the VHS source (I'll stick with 3 captures for discussion):

1) Capture the VHS source three separate times (which is what has been discussed a lot in other threads). This means then using AVISynth to offset the frames until all the frames line up for each VHS source.

or

2) Capture the VHS source three times simultaneous with three different but same capture cards. Would this have the same outcome as capturing the VHS source three separate times?

For either of these methods, you are saying the end results would be that I would effectively only be denoising the effects of the capture cards themselves and not necessarily improving the resulting VHS capture?

By the way, New Years is approaching. So I wish jmac698, everyone else, and all in Smurfdome a Happy New Year! May we all have the greatest video captures in the months to come!

[jmac698]

Capturing once with 3 different cards may cause the frame drops in different spots in each recording, which is good if you can merge them later. It also reduces capture noise itself. But the white lines or other noise from the tape deck are only played once, so you can't reduce that.

Happy New Year and may your captures not drop frames