The goal here is to edit the commercials out of recorded TV content. This must be accomplished quickly; if it takes longer than around 2 minutes per hour of content, then simple fast-forwarding through the programs seems to take about as long.
Fully automatic solutions to this are just not good enough: they incorrectly flag actual show content as advertisements and sometimes show advertisements that are unwanted.
The data will be video from many different types of show. For example, music videos will probably be much more difficult to classify than things like movies or sitcoms.
The idea is to cluster similar areas of the recorded content together so that the user can easily categorize the scenes. The clustered layout will be similar to PhotoMesa's.
There will be many different clustering metrics available, some that apply more or less depending on the type of TV program. There must be some simple way for the user to select which algorithm to use quickly.
The toolkit to create this project has not been completely decided on yet. Something like Jazz is likely to be overkill. Most likely, something like GTK will be used.