Classification of video shots using activity power flow

We propose a new method that can successfully classify video shots into broadly defined video genres. The classification process is performed using a new low level metric, the activity power flow, which is able to describe coarsely the spatial content of a video shot, as well as the temporal variations of that content within a shot. We also present a more robust method of calculating the motion intensity of a video frame directly from an MPEG bitstream, by discarding unreliable motion vectors which do not represent the true motion within a video sequence. These metrics are used as inputs to a radial basis function network in order to classify video shots into the four video genres - sport, drama, scenery, and news. Experimental results show that this method is both efficient, as processing is undertaken in the compressed domain, and effective, providing an accurate method to transform low level visual features into high level semantics suitable in a video indexing and retrieval application.