By Yi-Wyn Yen
There’s a ton of videos to watch on the Web. YouTube uploads 13 hours of video each minute. But finding what to watch isn’t easy.
Video search is one of the biggest challenges on the Internet today. A number of online video companies are trying to figure out how to find and discover content for viewers, but a debate rages on the best method to deliver results.
VideoSurf is the latest startup that claims it has solved the video search problem. Major video search players like Blinkx, Truveo, and Everyzing rely on tagging video clips with descriptions and analyzing the audio portion of clips to make videos searchable through text. Google (GOOG) and Microsoft’s Live Search (MSFT) are also making headway with speech-to-text technology to index videos. VideoSurf cofounder Lior Delgo, a former Yahoo search executive, says his company has made a video search breakthrough by scanning and analyzing the images within videos as a way to organize the content.
Analyzing a video is an incredibly complex problem. VideoSurf does the heavy lifting for web viewers with algorithms that identify people and backgrounds in videos and then converts them into thumbnail frames that summarize sections of the video.
Visual computer scanning means search results will be more relevant, says Delgo. “Before this, there was no way to navigate videos visually. You had to rely on tagging content surrounding the video. But video is not text,” he says. “If you can visually see what a video is about, you can make better decisions about whether you want to watch it or not.”
For example, say you want to watch clips that feature Republican vice president nominee Sarah Palin on The Daily Show. VideoSurf will pull up video results and display frames in which Palin is shown. Viewers can then click on the frames and jump to a particular scene on the video. VideoSurf indexes video content from a number of top news and entertainment sites, including Comedy Central (VIA), Hulu, YouTube, and CNN (owned by Fortune’s parent company, Time Warner.) VideoSurf launched its new product last week and currently requires viewers to sign up to use the service.
VideoSurf is not the only company that uses visual scanning technology to index videos. Digitalsmiths has been cataloging digital videos for movie and TV studios for the past three years and is now providing its technology for web-based videos.
Last month Warner Bros. (which also is by Time Warner) launched TheWB.com and runs Digitalsmiths’s search technology so that viewers can find every scene, say, where Joey Tribbiani says “How you doin?” on Friends. TMZ.com (another Time Warner property), a popular celebrity blog, also powers Digitalsmiths to archive its videos.
The top search engines are pushing for more videos to be included in text-based searches. For example, when a search for Jamaican Olympic gold medalist Usain Bolt is entered, at least one video pops up in the first 10 results for Google, Yahoo (YHOO) and Microsoft’s Live Search.
Search engines have the computing power to process the massive amounts of video data, but don’t have the technology, which is why Gartner digital media analyst Allen Weiner predicts VideoSurf and Digitalsmiths will be acquired soon by the one of the major engines. “The search engines have been waiting for the next generation of technology to come alone. You can only do so much by metadata and tags. This is going fuel a lot of interest in acquisitions,” Weiner says.
Both VideoSurf and Digitalsmiths provide similar technology for video search, but their business approach is different. Digitalsmiths partners with media publishers to allow them to customize its video search products. VideoSurf, like YouTube, is essentially a destination site for consumers. Digitalsmiths CEO Ben Weinberger says, “Nobody should think they’re so powerful that you’ll be the place that everyone wants to come watch videos.”
Online video analysts say both VideoSurf and Digitalsmiths are making video search better, but there’s still a long way to go. “The reality is, when you compare performing a typical [text-based] web search on Google to searching for videos on any of the engines, it’s nowhere near as good,” says Will Richmond, a broadband video analyst for VideoNuze. “You’ve got speech-to-text, facial scanning, and tagging. There’s a real art to figure out which of these will yield the best result. At the end of the day, you can really only measure the approach by the quality of the results.”
