Info Wanted: Averaging the various versions of the 2004 master?

I was writing a post about how the 2004/2011 transfer and how it could be surpassed in detail by a modern transfer from the same sources and how on top of that, we only have a compressed version of it at our disposal and an idea struck me:

People who do LaserDisc transfers often do several captures and then average those captures to gain as clear an image as possible, free from artifacts not present in the source but introduced randomly during the capture - now if one were to take all the available versions of the 2004 transfer (the various HDTV captures, the Blu-Ray, the digital versions from various providers) and average them in the same manner, could one get closer to the uncompressed master that way? It obviously wouldn’t work for the shots that have been changed between the 2004 and 2011 versions but those could be replaced with an average of only the HDTV sources.

Now, I’m not necessarily proposing this be done, and I’m definitely not saying I would do it, because I don’t have the know-how nor do I have access to all those sources; I’m just asking if it could be done and if there would be any merit to doing it.

This is pretty much what Mike Verta is doing with his restoration. He puts all of his IB prints in and averages them out and, according to him, you should have an image that closely resembles the original negative.

Well sure, although the IB prints all had the same print master, so by averaging them, you can at best arrive at the detail of the master, not the negative, but I guess if you throw in a couple of Eastmans, you can get something even closer to the o-neg, though again, those still have IP and IN grain in them but if the prints came from different IPs, I guess in theory the info from the o-neg could be there, but I think you’d need a lot of prints for that, but that is a discussion for another thread.

But what I’m suggesting here with the 2004 version is more similar to the LD capture averaging - these are digital encodes that all come from the same common source, only with a layer of artifacts on top, so there would be no frame by frame image registration needed, or color matching or anything - just stack them on top of each other and average them.

Ah I see what you mean. Just to finish up the MV discussion, I know he has some other prints as well (97 and perhaps some Eastman prints) and he may have used those but I’m not sure.

Right back to the 2004 idea: I think the main problem is that they all use the same digital master. The only difference is the level of compression. The bluray is already the closest we’re going to get. The advantage to using this technique for laserdiscs and film prints is that they are analogue copies and this averaging technique is trying to peel away the analogue generational anomalies.

Yeah but what about the digital versions, like iTunes and such? They and the BD are all encodes from the same masters with digital artifacts on top, which will be unique to each encode - just like the LD captures are all captures of the same master with analog artifacts on top, which are unique to each capture, no?

I’ve done this to some degree in my HDTV project. My experience with this is that averaging will not improve the detail in any way, and will certainly not remove encoding errors/issues.

The benefit of averaging in an LD capture is that you’re combining 3-5 copies of the exact same master through the exact same capture protocol. The goal is to eliminate analog noise, not increase detail. That’s not really possible. If you’ve tried capturing LD sources, you’ll notice that if you use only a single copy of the film, any defects due to the physical structure of the disc will remain even after averaging. This include laser rot, or dropouts, etc. The only way to remove those types of defects is by capturing from three or more copies of the exact same LD, with the same mastering, and using the median, rather than the average. Using the Median allows the video processor (AVISynth) to retain the most common or middle value for each pixel, which will not include random dropouts and laser rot speckling, that multiple copies of the disc are unlikely to have in common.

As far as the HDTV and Blu Ray versions, the first step is to register the frames, which are not always lined up, even though they hail from the same master. They are different down to the subpixel level, and can require multiple rescaling procedures to register perfectly. This can have a minor negative effect on the detail, in and of itself. Once the frames are properly registered, any averaging will just have the effect of lowering the detail and lessening the impact of encoding errors, not eliminating them. In an average of three sources, the master with the least detail still has one third of the input into every pixel, this applies to frames with encoding errors as well.

In working on version 2 of my HDTV project, I’ve found the best results come by using the highest quality source as the backbone of the project, and replacing damaged frames with the next best source. Ideally, you would get the best results using a method like Mike Verta’s, where you stack the layers and erase damaged areas to reveal info from the frame(s) below. Again this works well where the damage is dirt spots, but takes a lot of skill to match the sources when there are large parts of the frame damaged by the encoding/capturing process. Replacing frame sequences has proven to be much less labor intensive than that ideal, though color correction is sometimes required.

I’ve also seen that the Blu Rays are not more detailed by default. For instance, the Blu Ray of Episode III has more DNR applied and has much less grain than the German HDTV source. Sometimes, especially for the opening crawls, I’ve found that I get good results when I combine the Luma channel from the Blu Ray with the chroma channel from the UK SkyTV source.

Additionally, the colors in different captures are not identical, especially between the blu ray and hdtv versions, though to some extent, different captures are different as well. Averaging and regrading would make this not so big an issue though.

Chewtobacca said:

schorman13 said:
As far as the HDTV and Blu Ray versions, the first step is to register the frames, which are not always lined up, even though they hail from the same master.

Yes; this the problem. It would be nice if different encodings from the same source lined up perfectly, but they rarely do.

I just learned something new today. You’d think the opposite was true

I agree that it probably wouldn’t do very much. The only benefit would be something more along the lines of what schorman was saying by taking individual elements that are better in one source and then applying them to another.
(Nice point about ROTS…and to think that was the one BD I thought would actually be an improvement! Haha, guess they never do learn…)

If the separate sources are close enough then you can motion compensate the frames instead of trying to achieve a perfect alignment. I do this regularly with laserdisc caps. They never align perfectly even using the same machine with the same settings.

schorman13 said:
[snip]
I’ve done this to some degree in my HDTV project. My experience with this is that averaging will not improve the detail in any way, and will certainly not remove encoding errors/issues.

Thank you for your detailed analysis, great info.

About 3 years ago I experimented with producing a median average of the 4 different DVB releases of the 97SE of Star Wars. It had been suggested at the time that the smearing in these releases was the result of the capture cards/equipment used, rather than DVNR in the broadcast master.

I seem to recall the averaging did improve the picture, just not very much. 😦

I spent a lot of time trying this with PSB, and eventually gave up. Several forum members worked on a variety of avisynth scripts, but in the end the result never looked as good as the best single capture. I’m not saying it isn’t a good idea, it’s just that you need to have some really sophisticated software to align everything perfectly.

I think in theory it would be good if you had at least three versions, and could eliminate the most disparate of the three (frame by frame), averaging the remaining two. That way you could remove noise. Otherwise, by averaging all the takes, you’re essentially incorporating all the noise from all the takes.

It’s also a very different situation depending on if you’re working with video or film.

Puggo - Jar Jar’s Yoda said:

I spent a lot of time trying this with PSB, and eventually gave up. Several forum members worked on a variety of avisynth scripts, but in the end the result never looked as good as the best single capture. I’m not saying it isn’t a good idea, it’s just that you need to have some really sophisticated software to align everything perfectly.

I think in theory it would be good if you had at least three versions, and could eliminate the most disparate of the three (frame by frame), averaging the remaining two. That way you could remove noise. Otherwise, by averaging all the takes, you’re essentially incorporating all the noise from all the takes.

It’s also a very different situation depending on if you’re working with video or film.

This is why I moved to motion compensating instead of registration. With laserdisc, it does seem to improve the pic quality by removing analog noise. Like others have stated, I’m not sure that a digital master would benefit a whole lot from this. I’m not home right now but I will give it a try. I have the HDTV caps and blu-ray. Will post any positive results here if there are any.

pittrek said:

Can some of you guys recommend some good registration techniques / software? I always did it in Avisynth but it rarely was “perfect” during the whole video.

The auto-align feature in Photoshop works pretty well, but it would be unwieldy for more than a few frames here and there without creating a script.

I have the sky(UK) and Premiere(German) caps. If anybody knows where to find the others please pm me.

There are torrents out there for Cinemax rips of the OT, but not the PT. I think I deleted them, but I’ll check when I get home from work. I think it was discussed in Andrea’s thread about HDTV sources.

schorman13 said:

There are torrents out there for Cinemax rips of the OT, but not the PT. I think I deleted them, but I’ll check when I get home from work. I think it was discussed in Andrea’s thread about HDTV sources.

I’m only interested in the OT anyway. I looked in Andrea’s thread and used that information to try and find the cinemax or any of the other older mpeg2 rips for that matter but had no luck in finding them. They are probably bad sources for this kind of thing anyway.

They’re not that bad, but not as good as the two AVC versions. They are still available on rutracker. Just look for the ones that are mpeg2 encoded. Star Wars is 15.36 gb and Empire and Jedi are both about 16.5 gb. There are comparable versions of AOTC and ROTS, but no Episode 1. The audio is Russian, though, not that that matters.

Thanks for the info. I had looked there but didn’t see them at first. This time I added hdtv to the search and they popped up right away. Like you said, they don’t look too bad.