[mythtv] New Method for Commercial Detection.

Robert Dunn anakinberrybread at paradise.net.nz
Wed May 3 04:50:11 UTC 2006

Steven Adeff wrote:
> I wonder if Philips will integrate this into their recent patent to 
> further prevent us from skipping commercials...
Yeah, thats something I will deal with when I have to.

> By the way, besides the presence of the paper describing how to 
> effectively take audio fingerprints, there is nothing that would stop 
> you from combining these with a video fingerprint right?

No there is nothing stopping you from doing that, however, consider 
doing this in real time. With the audio I can work off 16bit samples. 
Even a small video frame is (assuming gray scale) 320*200*8bit 
(512kbit), the amount of data processing is orders of magnitudes 
greater, I don't see the need to get involved with a video finger print 
considering the huge CPU processing overhead it would create. Besides 
there are very few instances where the audio of television would be an 
exact match for 2 different video sequences. However, the same paper 
could be relatively easily adapted to a video algorithm but I just dont 
think it would be possible to run it in real time.
> There is a fairly decent implementation of a convolver in lavc I 
> believe.  Check out some of the filters in mplayer for examples of how 
> to use it.
Im unsure, searching large databases is not a strong point of mine 
(rarely do I have to deal with that kind of thing in my embedded stuff) 
I deal mostly with real-time processing.

Now i'm in the process of getting myself a PVR-350 and a linux box to 
test on. Does anyone who is familiar with the commercial detection in 
myth feel like giving me a hand by explaining briefly how 
CommDetectorBase etc all fits in with the operation of MythTV, ill get 
it eventually but a quick summary just a paragraph just to point me in 
the right direction would be wonderful.



More information about the mythtv-dev mailing list