[Subhandler-devel] Experience with SVCD/VCD on videolan

[R. B. <ro...@pa...>]

I've recently been adding SVCD and CVD subtitle handling in VideoLan's
vlc client. I started with subhandler which ahs been extremely helpful.

Although I'm not totally finished with this, folks might be interested
in looking or trying the code. In some small ways, it may be a tad
cleaner; some duplicated code has been made into a macros or a
procedures. One thing that I think has been a win has been taking out
the color palette encoding and using an underlying YUV format. When
trying to scale to counteract the scaling effects of scaled chromas
like YUY2 or I420, having individual pixel entries helps I think.

The routine that does the scaling I think is pretty simple and seems
to work well. It scales down and works for aspect ratios between 1/2
and 1 (the inverse of say 4:3 or 16:9) where you combine at most 3
linear pixels into one. 

In vlc I've noticed as alluded to above that for YUY2 or I420 formats
I've had to reverse scale the subtitles (e.g. 3:4 for a 4:3 aspect
ratio). However for the RGB formats, no scaling should be done. I
looked at the subhandler code and I don't see if it has to deal with
the problem as well. (It does look like there is a bug somewhere in
subhandler at the ends of the lines of the boundaries of the
subtitles. Perhaps the next line data is bleading through.)

One bad thing about vlc in its current form, is that there really
isn't a image library that I can use to do the blending. Nothing as
simple as what you do in xine and say put pixel with color c at
postion x/y. In fact I'm not sure there are *simple* routines exposed
in the the core to say convert YUV to RGB. (There are some
*complicated*, not very well documented, and not exposed routines or
macros though.) So alas I've had to write code blend in all the
different chromas and that's been tough going.

Anyway, to close. If folks have nice non-proprietary subtitle test
cases - e.g. subtitles in different colors, maybe with different
transparancies I'd appreciate of learning. Of course I could over time
create such, but I'd rather be coding and not possibly recreating work
that others might have.

Thanks