MJPEG Tools Svn

Uncluttering lavtools (in preparation for adding several more utilities) by
relocation to the new y4mutils directory

OK, this is a big quality improvement... The image quality now is better
(by far!) than by any other deinterlacer (motion-compensating or not) I
know. I should disable the lowpass-filter in the postprocessing-stage,
too (This then gives an image which nearly does not show any loss in
sharpness...)

Bugs: first two fields garbage. Need to interpolate spatialy.
not as fast as it could be...
does not compile cleanly

further quality improvement

0 is a valid option and does the expected thing for -t (some or all of the
temporal filtering can be disabled). Also, document that 0 works for -g
by disabling some or all of the gaussian blur processing.

gcc complained thusly:
imageplanes.cc:82:2: warning: no newline at end of file

so add a \n :)

Unused variable dotptr.

Ugly hack time. In the CVS version of libquicktime the quicktime_t structure
has become an opaque data type and the underlying filedescriptor must be
obtained by using the accessor function lqt_fileno instead of the ".streamer"
member (which is no longer available). However, libquicktime 0.9.8 has not
been released yet. IF libquicktime is being used then test for lqt_fileno()
being present and define HAVE_LQT_FILENO for use in 1 place.

When libquicktime has a release this can all go away when the required version
of libquicktime is bumped to 0.9.8

Argh - can't commit this module's change until libquicktime releases 0.98

Enhancements by the author were submitted. He recommends reading the
notes/comments for more details ;)

Change to a better (AKA cleaner) approach towards the motionsearch-problem
in interlaced streams. The trick is to shift bottom and top-field a little
up (resp. down), to avoid up and down jumping Motion-Blocks...

This fixed a lot of bugs, introduced some new and is far from being complete.
After the deinterlacer has syncronized (after two fields) the image-quality
is far better than before. These first fields only can be interpolated with
a sinc-filter... but are currently not reconstructed. -> Bug Nr.?

Stefan