Video Search To Use Internal Content

    October 31, 2008

Today the WSJ profiles a number of video search engines that actually seem to be getting smarter. Instead of relying on external meta data to determine the content of a clip, these engines are looking to data internal to the clip itself—including dialogue and people (or characters) appearing in the clip. And yep, one day YouTube might not be the #1 video search engine (although they may still be the #1 video hosting site).

Some of the more advanced video indexing technology is capable of indexing by images in the video, including characters and actors, using a form of facial recognition software:

Elsewhere, VideoSurf Inc. is analyzing the actual visual content of videos using technology known as “computer vision algorithms,” which produces more relevant search results, says Lior Delgo, the company’s chief executive. Computer vision is the science of programming computers to process and analyze images and video.

For example, VideoSurf’s technology can identify characters within search results. A search for the television show “Lost” brings up results for the show and also a thumbnail photo for each character. Clicking on the thumbnail of “Lost” actress Evangeline Lilly will bring up clips from the TV show and also other clips of Ms. Lilly, like her appearance on the “Late Show With David Letterman.” The site has indexed 10 million videos from 50 different online video sources.

Six weeks ago, Google premiered Gaudí GAudi, a video search engine that uses speech recognition to find words in clips and cue up the clips in SERPs to the uses of the keyword. So far, GAudi is only processing election videos, but the technology is being refined through this process and will hopefully be applied to other categories.

Mefeedia is also taking into account user ratings/votes in its ranking schemes. OVGuide relies on people in a different way—editors select the sites to be indexed and says they also have editorial control over ranking.

These two engines are meta search engines for video, pulling results from several different sources (licensed, possibly bootleg and UGC included)—and plenty of them don’t host original video content at all. Also in this category is the familiar name Blinkx, which indexes content from Hulu, CBS and Showtime. Blinkx also uses external clues, including links from Wikipedia or social media, to identify potential viral hits.

What advancements in video search are you most looking forward to?