I would vote for jumpy subtitles over jumpy objects.
I would really like to help you, because you've helped me so much with your contributions to this forum. Unfortunately, I don't think I can take the pebble from your hand on this one. This is quite an advanced exercise! Whew!
Your knowledge of avisynth plugins and the built-in features of avisynth exceeds mine, so I don't pretend I'm saying anything here that hasn't already crossed your mind. But perhaps my thinking out loud here provides some small spark of an idea. But for anyone else that may get something out of this, the only trick I've used in simple cases to help clean up selective parts of a frame when something like the ReplaceXXXMC scripts doesn't do the trick (because multiple frames had the same defect), is to use a very rudimentary trimming of a block of pixels from a frame before or after and overlay it on the target frame. Once there is motion within that part of the frame, this method breaks down rather quickly. But if there's no motion, it would seem possible by determining the maximum size "box' that the captions drift around in, and passing those coordinates to a function like this repeatedly frame by frame.
Code:
function blockreplace(clip c, int targetframe, int refframe, int left, int top, int right, int bottom, int xoff, int yoff)
{
f1=c.trim(targetframe,targetframe)
f2=c.trim(refframe,refframe)
f2b=f2.crop(left,top,(right-720),(bottom-480))
newf=Overlay(f1,f2b,x=(left+xoff),y=(top+yoff),opacity=1.0)
result=c.trim(0,targetframe-1)++newf++c.trim(targetframe+1,0)
return result
}
# Example usage
# Take pixels inside "box" defined by {x=40, y=300} and {x=680, y=420} from reference frame 99, then overlay on target frame 100
blockreplace(100,99,40,300,680,420,0,0) # set x and/or y offsets to non-zero to shift box relative to initial left/top coordinates
# same call with frames incremented
blockreplace(101,100,40,300,680,420,0,0)
#copy/paste/adjust previous line to exhaustion, or perhaps create some kind of "loop" that iterates over a range of frames calling blockreplace with frame numbers incremented by 1
I'm not of the experience level to create an avisynth plugin. I'm just brainstorming how I might try to do such a thing with the built in tools avisynth provides. The devil is in the details. The vision or theory of the idea is easy.
In very high level abstract terms, perhaps there's a way to just replace the bottom of the frame with a repeating still image without captions (using the blockreplace method above or some other kind of overlay of a constant image), then use tools mentioned in
http://avisynth.nl/index.php/Interna...d_Mask_filters section of the avisynth page to create a layer that is caption only that can be overlayed on the captionless layer. The caption could be snapshotted/isolated from a single frame and repeated on this new caption only layer. Perhaps some hue and/or saturation based filter/keying could help isolate the caption from the background. The picture/examples on
http://avisynth.nl/index.php/Overlay is how I see it working in theory, at least.
I don't know what methods were used. But it seems you've been able to solve both problems in isolation from the other. It's easy to say, well just take the two, isolate the caption part and then re-overlay it back on the other version. But, of course, it wasn't that easy. So I hope all this rambling on about theory isn't a big waste of time. I guess talking things through is the only help I can give at this time with my level of experience. Sometimes talking through the "obvious" stuff can spark a new idea or uncover a blind spot we might have.
I wish you the best of luck. I understand how much labor/love goes into restoring even a small bit of video. What we do is very painstaking, but also extremely noble and rewarding.
If you or anyone else does figure this out, I would really enjoy/appreciate learning what is done. I'm naturally curious, and I may have a similar problem to solve some day.
