[mythtv] Commercial flagging paper in December IEEE Computer

[Daniel Walton]

On Thu, 4 Jan 2007, Lucas Meijer wrote:

> Daniel Walton wrote:
> >> doesn't the new commdetector2 code use a plugin style architecture so
> >> other methods could be created and easily added?
> >>
> >>
> >
> > yep.  The experimental commercial detector is currently using a blank frame
> > detection plugin and a logo detection plugin.  The code is there for scene
> > change detection but that plugin isn't being used.
> Unless things have things have changed radically and I haven't noticed
> any of it, this is not true at all.
>
> There are two "plugins", (which aren't even plugins as they aren't
> loaded dynamically).
>
> ClassicCommDetector is what is used by default, which uses a combination of
> blank frame, scene change , aspect change and logo detection.

Steven asked about commdetector2 so that is what I was referring to in my reply.
I haven't done any work on ClassicCommDetector so I'm not very familiar with it.
commdetector2 uses a combination of logo detection with blank frame detection to
determine where the commercial breaks are.  Using just logo detection alone
isn't enough because

- we can't detect with 100% accuracy when the logo is and isn't present in a
frame.  If the logo doesn't have enough contrast against the background of the
frame then it is very difficult to detect the edges of the logo so sometimes we
don't think the logo is present even though it is.

- On many stations the logo disappears a few seconds before the show cuts to
commercial and reappears a few seconds after the show restarts after commercial
break.

So commdetector2 will adjust the commercial breaks as identified by logo
detection to begin and end on nearby blank frames.  I've been using/tweaking it
for about six months, it works really well for me.

Daniel