Avisynth (or similiar software) - Correct for Frame Drops

[Iceblade 423]

Needing some questions answered about the potential of solution I am considering.

Background on the Transcription Process:
Tapes: Analog High8 Tapes 1991 to early 1998 - Sony Hi8 Metal P120 [P6-120HMP] with anti-static LID (NTSC of course). My dad always bought the "better quality"/more expensive high8 (and later miniDV) tapes.

I'm using a like-new (after-market refurbished from a reputable seller) Digital8 Sony Camera DRC-TRV820 using TBC. The camera has a pretty good analog-to-digital converter within itself. I'm pretty happy with the output AVIs anyway. Certainly no shortage of data when 20GB video files for 2-hours.

I'm using the built-in plug-in-play Sony transcribing software and firewire on a 2008 computer (running windows 10 with an 2010s graphics card)

Issue
While for the vast majority of the videos I have transcribed; 90% of the frames have been of good quality, so noise in the form of static or lines is quite rare. Transcribed videos that have dropped frames is uncommon. When it does occur, it has generally accounting for <1 minute of video spread across the whole 2 hours.

I do however have two tapes from 1995 that are showering considerable dropout for multiple seconds at a time. Total video loss is closer to 30 minutes. All or almost all of the audio is intact. The point at which frames start to drop occurs mainly in the 1-hour to 1.75 hour region of the tape. The frames that are dropped appear to be random and not tied to particular frames.

Here is an example. VOLUME WARNING: Loud SeaGulls: https://www.youtube.com/watch?v=KYFVSjmV_nY

Questions
What I am considering is performing multiple transcriptions of the affected segments. Then using AVISynth to splice the frames together that weren't dropped to recover as much of the video as possible. I've researched into AVISynth's functions and I wonder if it possible to combine transcribe the tape 5 or 6 times and use something like the median function to combine all of videos together while ignoring the frames that are just grayed out (dropped) as shown in the attached image.
[Attachment 60443 - Click to enlarge]

Is there a way to use the Median function or some other function to merge the videos into one video while ignoring any frames that are explicitly dropped? For instance, have AVISynth ignore frames that are clearly all one color as it merges the videos.

Do you have any other suggestions about what might be causing the frame dropout and how to correct for it?

[Quint]

Subtract actual frame from grey (dropped) frame. If threshold is very low (frame grey), then take next possible clip.
You can do this with:
-Loading clips in different variables and cutting each beginning so that they are temporally all aligned
-Using ConditionalFilter several times, if one "strikes", give it the frame of the next next clip, if not, keep frame.

Maybe someone with more experience with ConditionalFilter can help for more exact script.

[ajk]

Running a median of multiple captures should be able to recover the frames without much manual intervention. This is assuming a few things; firstly the timing of the captures must stay consistent and secondly the number of "good" frames must outweigh the dropped ones. Easy enough to give it a try, just capture a problematic section three times (or five, or more), line the starting points up and run a median. Maybe provide such short samples here.

If good frames are in the minority, some other form of processing must be looked into (e.g. searching for grey frames as you suggest).

Of course, ideally you would try to get a more complete capture to begin with, but aside from trying other camcorders/players that may be easier said than done.

[nicholasserra]

I don't have any comment about your proposed solution. But I just want to make sure you know you're solving a hardware problem with a software solution.

The DCR-TRV820 is a Digital8 cam. Yes it has the ability to play HI8 tapes and it does it well, but it's not always the case. The D8 cams are in fact my goto for first capture.

But the firewire out capture is not ideal, as you can see with your dropped frame. It usually gives a higher buffer around a dropout and shows a more significant gap.

Additionally, the firewire capture isn't an ideal capture method in the first place.Those HI8 tapes are analog, and a proper transfer would be S-Video out into a capture card like a ATI TV Wonder or Blackmagic, etc etc.

Following up on my comment about the D8 camcorder Video8/HI8 conversion: It's not always perfect, and when it's not working right, you should be reverting back to an older Sony CCD camcorder (or similar from other brands) to play the tape. In the case of the older analog cameras, they will give a significantly better result during the dropout then the D8 cams, because there's no conversion going on.

My usual flow for a trouble tape is this: Capture via S-Video on my D8 TRV-730 > capture via S-Video on my D8 TRV-740 > capture via S-Video on my CCD-TRV65.

IMO the D8 cams give a better result WHEN there are no dropout issues.