Re: gst-thumbnail issue



Dnia 07-09-2004, wto o godzinie 06:52 -0700, Paul Ionescu napisał:
> > >  I guess not,
> > > because it gives always the same output. Another
> > question is if this
> > > behavior can be modified, for instance we can
> > select to re-thumbnail a
> > > file if the first thumbnail was not acceptable, or
> > even we can make it
> > > recognize bad frames, and seek further.
> > 
> > It is theoretically possible, but it needs quite a
> > bit of logic to
> > figure out "bad" frames.  Think about it - when is a
> > frame "bad" ?
> Yeah, I assume a lot of math is needed. A basic test
> would be that a frame should have less than 95% same
> color, but this is an ideal case. Usualy bad frames
> are not just one color like black or whatever, but
> they have dots, or other things.
> So, now I think is too much for a thumbnailer to make
> intelligent decisions about a frame.

First thing - there's difference between "bad" as in "too dark to see
anything", and "bad" as in "broken". Latter is undoable in general
(different codecs break in different ways), and former is, IMHO, not
worth doing it. If 1% of thumbnails ends up being mostly black, that's
still going to give distinctive info, while avoiding possible
spoilers ;).

> > I'll probably change it to "somewhere in the middle"
> > or even "take four
> > and put them in one".
> 
> Well, first they have to be runtime configurable, in
> order to let the user choose what it wants.

What for? I can't imagine single user wanting to adjust that.

> I think the "somewhere in the middle" scheme is going
> to solve most of the situations of black frames.

That's roughly what current algo does in practice. Works OK for me

Cheers,
Maciej

-- 
"Tautologizm to coś tautologicznego"
   Maciej Katafiasz <mnews2 wp pl>
       http://mathrick.blog.pl




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]