Video search
Sunday, July 31st, 2005The latest Economist Technology Quarterly (free access for the moment) has a brief status on video search. Searching the image itself is very difficult, but there is a lot of activity in searching associated text: the closed captions, the audio portion (via speech recognition), and human or machine produced video metadata.
IBM has a test version of a more sophisticated analysis/retrieval engine called Marvel; after manually tagging 1-5% of a set of video content, the system learns to recognize the rest — this is a huge reduction of the expensive, error-prone human tagging effort. IBM’s web has good detail on the project. From those pages I learned about MPEG-7, which is a video metadata (rather than encoding) standard.
The ETQ acknowledges that business models are lacking, but singles out Critical Mention as one that is working; they offer web-based search and alert (”reputation management”) services for TV news. I see in the news last week that they now have a licensing agreement with AP (Associated Press) Digital.
Since it looks like metadata is playing a big role in video search, it should be interesting to see which producers/consumers are doing the tagging, and where the analysis and search logic resides. The big search engines are an obvious home, fancy services like CM are another, but why shouldn’t I also have something to find stuff on my computer or LAN?