Google, meet TiVo (09 Jun 2005)
What is striking is that despite all the buzz around video searches, none of these companies actually searches the visual content of the video. That is because actual video searching—finding all the clips that show a red car, perhaps, or George Bush—is an extremely complex problem. IBM, which is doing research in this area, is using a variety of techniques to determine the context of a clip: indoor, outdoor, sports and so on.
In February, IBM made a test version of its Marvel search engine available online. It uses a technique called “support vector analysis” to classify clips into particular categories, by scrutinising audio, video and transcribed speech. Having learned about a particular kind of content, it can then recognise it in future. It is, in short, a step towards an engine that can actually search video. IBM is working with CNN, the BBC and other broadcasters to assemble a library of clips on which to train the system. John Smith, a researcher in “intelligent information management” at IBM Research, thinks it could be ready for commercial use within two years. But searching video is one field where seeing really is believing.
Article URL: http://www.economist.com/science/tq/displayStory.cfm?story_id=4031167
Read 73 more articles from Economist sorted by
date,
popularity, or
title.
Next Article: Home alone
|