[MPlayer-users] new inverse telecine filter!

[Brian J. Murrell]

On Wed, 2003-04-30 at 04:17, D Richard Felker III wrote:
> OK, I've looked at the files. The problem is that you have a very low
> quality TV capture with lots of noise,

I was wondering if that was going to be the issue.

> so the filter can't accurately
> measure whether to merge frames. Fixing this is difficult, since
> making it less sensitive will break handling of complicated telecine
> patterns at scene change and such.

I see.

> IMO, the only way vf_ivtc will ever
> handle bad noisy input like this is if I find a significantly better
> metric for measuring whether fields match...

OK.

> BTW, a workaround is -vf denoise3d,ivtc (denoise the image first).

I did tried that at your suggestion and it certainly does seem to help. 
The ivtc seems to work pretty decently.

> It
> does fairly well with the default values, but you can turn up the
> spacial denoising if you find it's still messing up at some parts.

OK.  Noted.  So would that be increasing the spatial luma  strength or
the spatial chroma strength or both?  Equally or does one have more
impact than the other?

> Some segments definitely are, but the beginning is telecine. The parts
> meant to look like someone is recording them with a video cam are 60
> fps.

Bah!  Stung again.  I only looked at the first few pulldown sequences to
determine that the ivtc was not working.  If I'd have seen that "video
cam view" section, I would have picked a whole different clip.

On that point, it is not always easy to give a stream a complete overall
visual scan for ivtc suitability.  What would be nice is if the ivtc
filter could give some indication as to how successful it was.  i.e.
give some "percentage" perhaps of "sequences" it found out of sync, etc.
and exit mencoder with some "non-successful" return value if the failure
rate is too high.  If the input was too noisy mencoder "fails" (i.e.
exits non-zero), or if the input had to much non-telecined frames,
mencoder "fails", etc.

Thots?

> Yes, and for this same reason, vf_detc fails on lots of 'clean' input
> when the pulldown pattern is tricky or irregular... :(

Hmmmmm.

b.

-- 
Brian J. Murrell <brian at interlinx.bc.ca>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-users/attachments/20030430/815f4eb2/attachment.pgp>