[mythtv-users] Low End MythTV

Robert Denier denier at umr.edu
Tue Mar 8 01:46:55 UTC 2005


On Mon, 2005-03-07 at 19:59 -0500, Chris Pinkham wrote:
> > As far as the commercial flagging taking 8hours, the machine is a
> > Celeron 300, which came with the tiny 128k cache. I can only assume
> > that has something to do with the slow commerical flagging time.
> 
> This probably has a HUGE impact. :)  If you figure that at 320x240,
> you're frame has 76800 pixels, but at 720x480, you're up to 345600.
> Commercial flagging usually only looks at the Y part of the YUV data,
> so that would be 345KB per frame, which is a lot more than the amount
> of data that can be cached.  1920x1088 is over 2MB, so flagging those
> shows would be slow even on CPUs with larger cache.  I currently
> don't scan every pixel, as the resolution gets higher, the pixel
> spacing that I use for most things gets higher in order to try to
> keep the speed up.  I plan on consolidating some of the methods so
> that the flagging process doesn't have to scan a frame more than once,
> but this is a little ways off.  Since the various detection methods
> were independently created, the current code scans the frame multiple
> twice (once for blank-frame detection and once for scene-change).   It
> also partially scans a third time for the logo.


One random idea that I'm sure that is not new would be to create code to
remove the logo's.  It would of course be imperfect.  Logo's that use
translucency should remove more cleanly than others.  I think I played
with something like that a long time ago with a piece of labview code.
(Labview is often quick to code things in, but sometimes executes
slowly.)  Of course the little I did was a very simplified case.

At any rate doing this for real would take considerable work.  I wonder
if anyone anyone has already implemented this kind of thing anywhere?
At least code already appears to exist that detect the presence of the
logo so that would save some work.  The decision making on whether or
not to replace pixels and if so how to guess the original value would be
the interesting part.  

I do wish tv had some more entertaining programs on to justify the
effort.







More information about the mythtv-users mailing list