Difference between revisions of "Talk:Commercial detection"
Latest revision as of 23:13, 5 August 2012
Two possible suggestions: - Sound processing (i.e. more than just images) - Commercials tend to repeat, so you could you this to get another metric for a match. Of course, it wll need a fairly well maintained database...
I am intersted in signal/image processing, so it'd be nice to have a bit of a page spring out from this about the current technical means/ideas for implementing commercial detection.
Do commercials have Teletext/Subtitles? If so could there be a database of Ad scripts?
Commercial Flagging Performance
I think it would be great to add in a section on the various performance penalties that the various commercial flagging options add. Obviously blank-frame seems to be the easiest, and scene change probably isn't too hard, but the logo detection probably yields some processor hit.
Another topic that I often don't seen mentioned is the accuracy of the commercial skip function given various recording qualities. I tend to record a plethora of kids shows in low quality and have noticed that the commercial breaks are missed more frequently on lower quality recordings.
It would also be nice to mention how the Strict Commercial Detection option affects each of the other features.
I've seen several networks (increasing in numbers) that have started adding or removing commercial skip identifiers around their show advertisements. Often the commercial skip detection leaves the advertisements for other ABC shows just prior to the real show returning from commercial. Although it may prove impossible to identify commercials when the network purposely tries to hide them, it would be interesting to see an option to adjust the sensitivity of the commercial detection options.
commercial flagging database
Perhaps it would be helpful to have some type of searchable database that has pointers as to what settings work best for different types of shows, on air and cable providers, locales, specific channels, etc. And some method to make it simple and painless enough that people will actually use it and contribute input.
I'm looking at one recording in final cuts pro, using the vectorscope, waveform monitor, and histogram tools. I wonder if there is some command line mpeg2 analyzer tool that does something similar and generates frame-by-frame data just writes that to some csv text file. Then by using that data, you can tweak your flagging paramaters for different shows, channels, etc.