[MPlayer-users] Inverse Telecine on Noisy Material

[D Richard Felker III]

On Wed, Aug 27, 2003 at 08:42:28PM -0500, Angel wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> 	I've been trying to reverse the telecining in the episodes of Seinfeld I've 
> been capturing. Unfortunately, the video seems to be too noisy and the detc 
> filter is not able to pick out the interlaced frames. Are there any 
> thresholds/parameters that can be adjusted to improve the interlaced frame 
> detection? If not, what are the prospects of detc/ivtc etc. working better on 
> noisy material in the future.

vf_detc has some thresholds which can be adjusted by the user. See the
man page or better yet the source for info. However it's somewhat
confusing. Inverse telecine is hard enough to do right without noise,
though, so don't expect great results. One thing you could try is:

  -vf il=d:d,denoise3d,il=i:i,your_favorite_inv_telecine_filter

Optionally increase the strength of the denoise filter. This will
separate the two fields, clean up the noise, and recombine them before
performing inverse telecine.

You might also want to try the new (incomplete, experimental, intended
mainly for G2 so it might not work so well in G1!) vf_pullup filter
instead of detc or ivtc. It's algorithm is simpler and smarter and
looks more for duplicate fields rather than interlacing when
determining how to put frames together, so it might give you better
results. Keep in mind, though, that it will NOT work right with -ofps
23.978, so you'll have to leave -ofps alone and let mencoder duplicate
frames.

Rich