MJPEG Tools Svn

Described the new -s option (to skip frames), and fixed some typos.

Added a new option, "-s num", to allow beginning frames to be skipped.
The first num frames are written to stdout unmodified. This is useful
for resuming an interrupted process without duplicating work.

Changes to the man page, to correspond with my recent modifications to
yuvmedianfilter (and to fix some typos).

Added several options.

-I num: override the interlacing value read from the YUV stream.
-s num: Allow the first num frames to be skipped. Those frames are
written to the YUV output stream unfiltered.
-f: "Fast" filtering. The threshold-based averaging is not done;
instead a simple mean is calculated, much like what happens when
not enough values in the radius are within the threshold.
-w num: In fast mode, the weight to assign the current value, relative
to the values within the radius. Default is 8, just like in
the threshold-based version when not enough values are within the
threshold.
Note that certain combinations of radius/weight have been optimized
to be even faster -- radius 1 and weight 2.667, radius 1 and weight
8, and radius 2 and weight 8.

Code freeze is over (release yesterday), add the spatial filter submitted by
Dan Scholnik.

This commit was manufactured by cvs2svn to create branch 'MJPEGTOOLS_1_6_2'.

1.6.2 version number bump, require jpeg-mmx 0.1.5 for RPM build

There were actually several fencepost errors in mpegconsts.c (and in
mplex as well), due to inconsistencies in how the mpeg_aspect_ratio
and mpeg_framerate arrays were constructed/used.

I restructured the code a bit to clean up the interface, so that I
could fix the bugs (without going cross-eyed) and to make it less
susceptible to such errors in the future.

Andrew: I do indeed step most gingerly when I walk into your code!
I'm not married to this revised interface in mpegconsts.[ch],
but it's a little cleaner and it fixes some plain old bugs.

revert to rev.1.24.

- image variance fixed
- libmotion working again
- still too slow