Re: Filter (or anyxthing else) to stabilize shaky videos?

--nextPart1581530.XN5YeCXl39
Content-Type: text/plain;
  charset="iso-8859-15"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline

On Thursday 25 February 2010 21:59:27 Al Bogner wrote:
> Am Donnerstag, 25. Februar 2010 18:09:31 schrieben Sie:
>
> Hi Georg,
>
> > At the moment it looks like you have to handtune the parameters for some
> > clips.
> > However the fieldnum=3D40:mincontrast=3D0.3 does not hurt for the other
> > movies as well. So I would see how far I get with this setting.
>
> I am testing with this:
>
>
> 	transcode -J smartdeinter -i "$FILE" \
> 	-x ffmpeg -y ffmpeg,tcaud -F ffv1 -o "$DEINTERFILE"
>
>
> 	transcode -J stabilize fieldnum=3D40:mincontrast=3D0.3 -i "$FILE" \
> 	-x ffmpeg -y null,null -o dummy
transcode -J stabilize=3Dfieldnum=3D40:mincontrast=3D0.3 -i "$FILE" \
 the "=3D" is important!
<copied from other email>
> [transcode] warning: unused command line argument detected (11/12)
> [transcode] warning: argc[11]=3Dfieldnum=3D40:mincontrast=3D0.3 (unused)
>
> So if these options are unused, why was the boat video better.
</copied from other email>
I don't know. Did you use maxangle=3D0 as well?
>
>
> 	transcode -J transform=3Dzoom=3D10:smoothing=3D20 -i "$FILE" \
> 	-x ffmpeg -y ffmpeg,tcaud -F ffv1 -o "$STABLEFILE"
>
>
> Am I right, that fieldnum=3D40:mincontrast=3D0.3 is _not_ needed in the l=
ast
> part?
>
> I tried another test-file, which was not uploaded till now and incredible,
> it got fantastic with this options, but it is slow now, 5fps with my
> Dualcore 5050e. No question, it is worth the time, if it works as with th=
is
> very shaky new sample from a dark room.
>
> It's hard for me to understand, why some samples are stabilized and others
> not.
This is due to the fact that the stabilize algorithm is using small=20
measurement fields that are placed on the frames. The river for example cau=
ses=20
the illusion (for many local fields) that the camera is moving. Since this =
is=20
not the same on all frames the stabilizing is not so well. A whole picture=
=20
version is used with stabilize=3Dalgo=3D0. This is very slow and does only =
do=20
translations and no rotations.
> The ship-sample is better now. Probably this one belongs to the clips,
> which are better not stabilized. The statues in the mountain are more
> stable now, but there is a sharpness problem, the flowing river still is
> more shaky than the original and with the 4th sample there is no big
> difference with the original and the stabilized version.
>
> I have another important question:
>
> Kdenlive has a preset for 768x576 non-interlaced only, the original is
> 720x576 or 704x576 interlaced. What do I have to do to get 768x576 as
> result?
Mh, maybe just adding again a black border to it? Doesn;t Kdenlive do that?=
=20
Otherwise I know that ffmpeg can do that nicely.
>
> When and where to do I add filters for denoising and sharpening? The
> quality is always similar to the files you have. S-VHS tapes need not to =
be
> denoised, but a little bit sharpened.
>
> Which options do you recommend?
>
> Without stabilizing I use this sharpening options as last filter:
>
> -J unsharp=3Dluma=3D0.8:luma_matrix=3D7x5:chroma=3D0.1:chroma_matrix=3D3x3
This you can add to the transform call simply as
=2DJ=20
transform=3D.....,unsharp=3Dluma=3D0.8:luma_matrix=3D7x5:chroma=3D0.1:chrom=
a_matrix=3D3x3
>
> Al

=2D-=20
=2D--- Georg Martius,  Tel: +49 177 6413311  -----
=2D------ http://www.flexman.homeip.net ----------

--nextPart1581530.XN5YeCXl39
Content-Type: application/pgp-signature; name=signature.asc 
Content-Description: This is a digitally signed message part.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (GNU/Linux)

iEYEABECAAYFAkuHnKYACgkQ7Y/IAAz6KBPH4gCfUmZYDcQlRHMDcl37aWCHtrzm
MpgAn2JxFHaRerOauBbYyGc3FIbudNvT
=CmxS
-----END PGP SIGNATURE-----

--nextPart1581530.XN5YeCXl39--