Wednesday, December 22, 2004

Video searches on Web don't always click yet

IHT: "Last month's basketball riot in suburban Detroit was a disgrace, a shameful moment in the history of American sports - and the most entertaining TV show of the year.

Something bizarre, shocking or hilarious was happening all over the screen. Thank heaven for Internet video. A host of Web sites captured the spectacle, allowing us to replay it on our desktops whenever we like.

There is a vast amount of video stored online, footage of just about anything from congressional hearings to chimpanzees performing karate. But unlike the text content of the Web, or even still photographs, a video cannot be retrieved by typing a few words into an Internet search engine like Google.

Try Yahoo instead: video.search. yahoo.com, to be exact. That is where Yahoo is testing a new video search tool. It is actually an upgrade to a video search service that has been running for some time at alltheweb.com.

Yahoo video search is based on the same principle as the Yahoo image search. A "spider" program scours the Web, looking for files with names and file extensions like .mov., .mpg or .qt, which indicate they contain video data. These files and the Web pages that house them are then added to an index.

On a company blog, an engineer, Jeremy Zawodny, said that Yahoo was putting together a system called Media RSS, for adding more thorough identification tags to online media. If video publishers embrace Media RSS, every item they post will include a bunch of computer-readable information that will make it easy to index the file.

One of them is slap-your-forehead obvious: closed captioning. Most TV shows feature a captioning track for the benefit of the hearing-impaired. It is easy to capture and index this data.

Still, there is a lot of video that is not captioned. How can you index it? By listening.

That is the approach taken by a little company called Blinkx, in London and San Francisco. Its new video search, at www.blinkx.tv, could become the company's first hit. It is a visually enticing search tool that does a nice job of locating recent videos of major news events.

Unlike Yahoo's video search, Blinkx does not even try to find all the video on the Web. It focuses on video from 20 major news and entertainment producers, including CNN, the BBC, the Discovery Channel and MSNBC. Blinkx has a distribution agreement with Fox News. For the others, Blinkx uses its own software to locate and index video snippets, some by reading the closed captions.

The rest is done with voice-recognition software that listens to the narrator and translates his words into digital text. This text is indexed, providing a set of words that describe the video. Now, just punch a few likely words into the Blinkx engine - each search requires a minimum of three words. "

No comments: