modifications made by Frederic Briere:fbriere at fbriere.net
Date: Tue, 6 Jan 2004 11:35:37 -0500
Subject: Tentative patch for ivtc

Hiya!  Here's something I whipped out in trying to improve ivtc a bit.
My motivation was that de-interlaced frames never look as good as "real"
ones, and if you're working with material that was telecined the "other"
way (repeating the top frame instead of the bottom frame, or vice versa,
I forget), then you end up with only two good frames out of four,
instead of three.

The easiest solution is to patch filter_decimate.c to reject the second
frame instead of the first, and that's what I've been doing for a few
months.  However, it seemed that the smartest way would be to adjust
filter_ivtc.c itself, which I finally got around to doing.


The field parameter specifies whether you want to work on the top field
(by default) or the bottom field.  Basically, when you enable
verboseness, if you see lots of "using 0" and "using 1", then you stand
to benefit from working on the opposite field instead.  (If ivtc is
using 1s and 2s, you've got it right.)

I also added a bit of "magic", to give ivtc a little nudge towards the
current field instead of the previous/next one.  When looking at the
verbose output, you'll see that ivtc often favors the "wrong" field, ie.
if you work on the top field, and have the following frames:

  Frame:          A  B  C  D  E
  Top field:      1  2  3  3  4
  Bottom field:   1  1  2  3  4

Then there is a good chance that frame D will end up with C's top field,
for reasons that are beyond me.

The values I chose are completely arbitrary, but they seemed to yield
good results for me.  Maybe they should be configurable, or maybe
there's a much better way to do this.  In any case, I made it an option,
so it's not in the way if you'd rather not use it.