Can much be done about this transfer?

[houstonian]

I just got the transfer files back from Film Rescue International for some very old film I sent in. One of the reels was shot by my late grandfather and the other was shot by me back in the 90's, but it was my grandfather's stock and it was already very old at the time. The film was in very poor condition and you can see the complete transfers (joined) at https://youtu.be/dS2zmvlacp0. Please view in 1080p.

Can much be done with Avisynth or some other software to improve the appearance of this footage? It's very important to me, especially the first reel. What are some things that would be suggested to try? I've tried a few different ideas including temporal smoothing (VirtualDub) and temporal degrain (Avisynth), but without much improvement.

Thanks for any help anyone can provide. I realize it's still going to look quite bad no matter what I try, but I would like to improve it.

By the way, the YouTube video may contain duplicated frames but the original (right now in UT Video lossless compressed format) is currently at 14.985 fps after having removed all duplicated frames.

[jmac698]

Could you put the uncompressed clip on megaupload? I'm getting too many compression artefacts

[johnmeyer]

Have you looked at the original film with a magnifying glass? Is it really that dark? Can you see more details? If you can see more than what is shown on your YouTube version, then it would be worth spending the money to get it transferred using a Rank or Spirit machine. They can extract every last bit of detail that is on the film.

There is no magic to extract details that aren't there. The best you can do is reduce the film grain noise, just to make it a little more pleasant to watch.

[jmac698]

It's already a high-quality scan:
http://www.filmrescue.com/home-movie-transfer/

[houstonian]

Apparently megaupload is now defunct. However, I have uploaded the files to my GoDaddy hosting space which has unlimited storage and bandwidth.

Here is the UT Video lossless AVI converted from the original .mxf (DNxHD) files I got from Film Rescue (>5 GB):
http://hfrvideo.com/Film_Rescue_Transfers.zip

Here is a short lossless sample of the above file:
http://hfrvideo.com/Film_Rescue_Transfers_sample.zip

Here is the original .mxf (DNxHD) file of the first reel that I got from Film Rescue:
http://hfrvideo.com/Reel1.zip

And here is the original file for the second reel:
http://hfrvideo.com/Reel2.zip

The AVI files have had duplicate frames removed, but the .mxf files contain duplicate frames (apparently due to frame rate conversion). Both filetypes are at 29.97 fps and I was mistaken about the 14.985 (half speed) fps for the AVI file. I made an H264 version for YouTube that was 14.985 fps but the AVI files are at 29.97 fps so are double speed.

I originally tried to convert the .mxf files to a more useable format using ffmbc and other software but it didn't work, so at the recommendation of Cory from Film Rescue, I downloaded a trial of Premiere Pro CC and used the included Media Encoder CC to convert the files to lossless AVI via the UT Video codec.

[poisondeathray]

neat video can probably improve it slightly, at least make it more watchable

DNxHD is a decent intermediate codec for "normal" conditions, but when you have pure noise like yours it will fall apart . For best practices, and best results -you need to start out with a higher signal to noise ratio than what 100Mbps intraframe at 1440x1080 (of any codec) can provide. It probably won't make a huge difference in the end given the condition of the source - but if you look closely you can see compression macroblocks on top of the noise - you're starting off already at a disadvantage. On the website they say "master image sequence" - if that is available I would start with that.

[jmac698]

Hi,
I've been having computer problems this week. I got as far as looking at your sample and finding potential fixes (research level MATLAB software, movie production software, and consumer level software). To try each of these, requires the solving of several sub-problems, such as converting DNxHD to frames, which as you know is a pain. I can try the sample in UTVideo, and creating frames shouldn't be hard. If anyone wants to post the ffmpeg command line, that would save me a few minutes. I can use frames in movie production formats, like sgi, tiff, png, targa. I also need to clear a lot of space, so give me some time.

[Mounir]

With Neat you can have something like this (640x480 version, ut codec): https://mega.co.nz/#!xhByVTZZ!ZaDe6fAHzNYnnDPzChdYdn0boiJXGqwHi1z2tPGaEb0
Is it worth the trouble i don't know

[houstonian]

With Neat you can have something like this (640x480 version, ut codec): https://mega.co.nz/#!xhByVTZZ!ZaDe6fAHzNYnnDPzChdYdn0boiJXGqwHi1z2tPGaEb0
Is it worth the trouble i don't know
I think that's actually better than I was hoping, thank you! What settings did you use? And does anyone know if other software might do noticeably better?

Sent from my LG-H443 using Tapatalk

[Mounir]

These are the presets i used: https://mega.co.nz/#!soIV0BCK!vMO4LD_0SZBFtsVYtViJy7wUp7f7ZxTsoQlICQ20VCk

radius: 5 ,adaptive filtration: yes / Neat v3.0

[geometer]

Is there a spectral adaptive denoiser for video, as they have in audio engineering? This would also use a second of (almost) black and another of white signal (lead-in, camera blend up or something) as a noise footprint.
I think this should be executed softly in maybe two stacked steps, and only then we can throw it into one of the sophisticated motion-adaptive denoisers. Or even have a split control channel in the motion-aware denoiser, that derives the motion-detection from the denoised signal, but as its data input uses the original signal.
One of the tricks in a multistep process might work with the fact that denoising tends to create some steady or very slowly moving patterns of garbage, but the motion detection will rule them out because they are steady. Then, with a spectrally different treatment of that kind of noise at those locations, the garbage can be evened out a lot better.

[houstonian]

These are the presets i used: https://mega.co.nz/#!soIV0BCK!vMO4LD_0SZBFtsVYtViJy7wUp7f7ZxTsoQlICQ20VCk

radius: 5 ,adaptive filtration: yes / Neat v3.0
Thank you Mounir. Definitely worth the trouble to me. Should I build a different noise profile for the second reel? With the current profile, the second reel doesn't look as good as the first but maybe that's because the film was already old when the second reel was shot.

Sent from my LG-H443 using Tapatalk