Thread: VirtualDub2
View Single Post
Old 16th September 2019, 17:28   #904  |  Link
nji
Registered User
 
Join Date: Mar 2018
Location: Germany
Posts: 201
Having had a closer look on the matter, I'm suspicious
if the duplicate frames actually are simple added,
or if they stand for a kind of "dropped frame" the
camera hardware wasn't quick enough for, and so inserted
a duplicate to stay in sync.
(Maybe there is someone here who knows if that was done sometimes?)
I tried to estimate it by the motion (and a possible "jump" next
frame of a dupe), but I'm not sure.
If that should be the case then removing the dupes would be wrong
(Except at the simple pattern of a dupe every other frame).
An improvement in that case would still be possible
by replacing the duplicates by a "temporal mean" of it's neighbours.

In both cases (true dupe or compensation for dropped frame)
there remains the problem of identifying the dupe positions.

It can't be possible that the positions are arbitrarily.
So I had a closer look to that too.
And indeed, not for all duping movies, but for the majority I found
that the period p (dupe at last pos) of dupes is not constant but more complex.
It is n times p, then (p+1), then n times p again, then... etc.
I.e. the "real period" is n*p + (p+1).
Starting at an offset.

If (!) the task is to remove the frames this can't be removed by jpsdr's Remove Frame.
In ONE go.
But in (n+1) goes!
First go: offset, period: n*p + (p+1), frame at pos. p
Second go: offset, period: p - 1 + (n-1)*p + (p+1), frame at pos. (p-1)+p
(n+1) go: ...
Not very cool I admit.
And one has to find out the pattern, and if it holds until to the movie's end
by having a close look at the frames.
(I really hope not be completely wrong with all that ;-)

Last edited by nji; 16th September 2019 at 17:32.
nji is offline   Reply With Quote