Those will work now, and maybe for a while in Germany. But they are all easily defeated, by: Black frames: don't do them. Some shows right now have essentially zero black frames; I end up skipping the start of the show with my VCR because I don't see it. I've single-stepped my VCR, and there is almost no black frame(s). (A good example: repeats of "SportNights" on Comedy Central (US)). Also, you could easily transition to/from ads instead of adding black frames. Volume detection: obviously, play transition music; don't drop the audio, don't make the start of commercial louder: turn it up gradually. Logo: keep the logo up, or morph it so you can't recognize it, etc. Audio fingerprinting, segment duration: sorry, but ... give me a break! Image and audio recognition is primitive. People have been claiming for 50 years that voice recognition is "just around the corner", and it still is just that. The recognition you are talking about is much more difficult - but by all means work on it; it's useful research if nothing else. And remember: if you are not 100% accurate - say 95% - then you will miss 5% of your shows (false positives), or to be conservative not skip many commercials (false negatives). And do they *have to* always play the same music at the beginning of the Simpsons. Like I say, all of it can work until it is widely used, at which time the advertisers will stop paying, the TV providers will lose all their revenues, and they will then do anything - *anything* - to prevent ad blocking, and they will succeed. We US residents understand the power of the moneyed interests, because they run our country and our government. In the meantime we can all skip commercials the old-fashioned way: fast-forward the VCR, or even easier, your PVR if you have one (I don't yet). Peter K "The power of observation is called cynicism by those who do not possess it". (GB Shaw) Thomas Hargrove wrote: > I know that commercial deletion is possible because I have a pretty good > working codebase right now. My program, Karin, does not work on all types > of shows, but it is extremely accurate on shows that follow the standard > format. (intro, 2 larger segments, credits). Here are some details on what > we actually have working: > > 1) Black frame detection - Used to define video segments > > 2) Volume detection - Used to tell if a black frame occurs between segments > or during the same segment. > > 3) Logo Detection (Both transparent and opaque) - This will watch for the > channel logo that usually shows up in the bottom right hand corner. The > output of this module is a list of video frames that either have a logo, > don't have one, or unknown. > > 4) Audio Fingerprinting - My roommate implemented this using musicbrains I > think. This allows us to have a 2 databases of audio fingerprints, one for > commercials, one for shows. For instance, the intro song to the Simpson's > can be fingerprinted and placed in the "shows" file. The next time you > process a Simpson's episode, the intro will be recognized. > > 5) Segment Duration Analysis - This makes a guess at what a segment is based > on it's duration and position in the file. Most likely 30 second clips are > commercials, especially if it lies between 2 other commercials. For the > most part, any segment over 2 minutes is show. > > So far, this combination has been working quite well. The only tricky part > is detecting the credits of a show, but this should not be a problem once we > get image matching working. We have our first (alpha) release done, but I > just have to package it and put it up on the web. There should be a > Linux/gstreamer port sometime early this summer. > > --Tom > http://toonarchive.com/karin > > ----- Original Message ----- > From: "Robert Siemer" <Robert.Siemer@xxxxxx> > To: <video4linux-list@xxxxxxxxxx>; <peterk@xxxxxxxxxxxx> > Sent: Thursday, April 05, 2001 8:01 AM > Subject: Re: Automatically cutting commercials > > > From: Peter Kaczowka <peterk@xxxxxxxxxxxx> > > > > > Automatic ad detection is ultimately doomed to fail, much like the > > > US "Star Wars" missle defense: the counter-measures will always be > > > easier to implement than the initial measures. > > > > I don't think so. What a human can recognize, a computer can > > recognize - at least some time in futur. > > That's theoretical. > > > > But today it would already be possible (but not cheap!) to use image > > recognition. Really anoying is to see the same ad over and over > > again. Once the system knows the ad it will filter it next time... > > > > Easier (in Germany) is to use some of my hints I posted some days > > ago. To recognize a blank/black screen should be quite easy. Films > > don't tend to show black in timeslices of n*5 seconds. > > To make it perfect, let the system react, when "it is sure" that the > > last 1-2 minutes ads were going on. > > > > The consumer is now just watching the film "near realtime" 30 > > minutes later. The ad-detecting system writes into a buffer of up to > > 30 min. When the system sees a block of ads and "thinks": hey, that > > must be ads and they started 1' 35'' ago! just remove these block from > > the buffer. The buffer will shrink, of course... but somewhen the film > > will be over. > > > > > However, the broadcasters won't bother with implementing the > > > counter-measures until the "measures" are widely used, so > > > early-adopter hackers could enjoy commercial-free viewing for a > > > while, until everyone starts doing it. > > > > Some of the "easy to recognize" ad-hints are enforced by law in > > Germany. But generally I think you are right here. > > Maybe in time of the "coldest ad war" [-: companies use a way to let > > me consume well adapted ads for me. - I would like it much more to > > see things I'm really interested in, but that's another story... > > (Do I have a dishwasher?? - No!) > > > > [web scrubbers] > > > They work mostly based on the size of the ads, which tend to be > > > standard. > > > > The most easiest are URL- and header-filter. I'm using junkbuster for > > this (linux). > > > > > Varying the size a few pixels in either direction without changing > > > the original image would be trivial, but I don't think anyone is > > > doing that yet. The reason: hardly anybody runs the scrubbers. I > > > don't, although I guess I would if there was one for Linux. > > > > Better than "junkbuster" is "muffin". It's written in java. It can > > rewrite header, html, gifs... It knows about java-script, picture size > > (whith ranges you like :), animation, flash, URLs and much more. > > > > > When scrubbers become common, then the owners of pages displaying > > > ads will care, and then scrubbers will be easily defeated. > > > > If owners of companies would care what I do, buy and need, they would > > write linux drivers for their hardware, would build 4HE 19"-ATX-cases > > with all ports on the front side, would make watches which can > > communicate with my linux-PC, would let me see the specs. - I would > > pay for that. > > > > I could consume things much faster this way... [-: > > > > > > Bye, > > Robert > > > > > > > > _______________________________________________ > > Video4linux-list mailing list > > Video4linux-list@xxxxxxxxxx > > https://listman.redhat.com/mailman/listinfo/video4linux-list > > > > _______________________________________________ > Video4linux-list mailing list > Video4linux-list@xxxxxxxxxx > https://listman.redhat.com/mailman/listinfo/video4linux-list
